Table of Contents

EXECUTIVE SUMMARY.........................................................................................................6
1 INTRODUCTION.....................................................................................................................8
2 KEY AI TECHNOLOGY IN FINANCIAL SERVICES.....................................................9
2.1 MACHINE LEARNING.........................................................................................................10
2.1.1 Supervised machine learning..................................................................................10
2.1.2 Unsupervised machine learning..............................................................................10
2.1.3 Reinforcement learning...........................................................................................11
2.2 EXPERT SYSTEM................................................................................................................12
2.3 NATURAL LANGUAGE PROCESSING...................................................................................12
2.4 ROBOTICS PROCESS AUTOMATION...................................................................................13
3 BENEFITS OF AI USE IN THE FINANCE SECTOR.....................................................13
3.1 IMPROVING DECISION-MAKING PROCESS..........................................................................14
3.2 AUTOMATING KEY BUSINESS PROCESSES IN CUSTOMER SERVICE AND INSURANCE.......15
3.3 ALGORITHMIC TRADING IMPROVEMENT...........................................................................15
3.4 IMPROVING FINANCIAL FORECASTING..............................................................................16
3.5 IMPROVING COMPLIANCE & FRAUD DETECTION..............................................................16
3.6 REDUCING ILLEGAL INSIDER TRADING.........................................................................17
3.7 REDUCING OPERATIONAL COSTS......................................................................................17
3.8 IMPROVING FINANCIAL INCLUSION...................................................................................17
3.9 STRENGTHENING CYBERSECURITY RESILIENCE................................................................18
3.10 TAKEAWAYS FROM THE INSURANCE SECTOR...............................................................18
4 THREATS & POTENTIAL PITFALLS..............................................................................19
4.1 EXPLAINABILITY AND TRANSPARENCY OF AI-BASED MODELS........................................19
4.2 FAIRNESS OF AI-BASED MODELS......................................................................................19
4.3 LACK OF ACCOUNTABILITY FOR AI OUTPUT....................................................................19
4.4 DE-SKILLING OF EMPLOYEES IN THE FINANCIAL SECTOR..................................................20
4.5 JOB DISPLACEMENT...........................................................................................................20
4.6 DATA PRIVACY CHALLENGES...........................................................................................21
4.7 SYSTEMIC RISK..................................................................................................................21
4.8 HIGH COST OF ERROR.......................................................................................................21
5 CHALLENGES.......................................................................................................................22
5.1 AVAILABILITY AND QUALITY OF TRAINING DATA...........................................................22
5.2 USE OF SYNTHETIC DATA IN AI-MODELS.........................................................................22
5.3 SELECTING THE OPTIMAL ML MODEL.............................................................................23
5.4 LEGACY INFRASTRUCTURE................................................................................................23
5.5 LACK OF APPROPRIATE SKILLS.........................................................................................24
5.6 REQUIREMENT OF BETTER AGILITY AND FASTER ADAPTABILITY...................................24
5.7 AI MODEL DEVELOPMENT CHALLENGES.........................................................................24
6 REGULATION OF AI AND REGULATING THROUGH AI........................................25
6.1 REGULATION OF AI..........................................................................................................25
6.1.1 Risk-based approach to regulating AI....................................................................26
6.1.2 Existing regulation on the use of AI in the finance sector......................................26
6.1.3 The need for Human-In-The-Loop..........................................................................27

1
 6.2 REGULATING THROUGH AI...............................................................................................27
6.2.1 The regulatory context.............................................................................................27
6.2.2 The opportunity........................................................................................................28
6.2.3 The adoption of AI by regulators............................................................................28
6.2.4 The horizon..............................................................................................................29
6.2.5 The challenges.........................................................................................................29
6.3 REGULATORY TESTING OF AI...........................................................................................30
7 RECOMMENDATIONS........................................................................................................33
7.1 ACADEMIA.........................................................................................................................33
7.2 INDUSTRY..........................................................................................................................33
7.3 REGULATORS.....................................................................................................................34
REFERENCES...........................................................................................................................35

2
 Abbreviations:
AI Artificial Intelligence
API Application Programming Interface
DARPA The Defense Advanced Research Projects Agency ES
Expert System
EU European Union
FCA Financial Conduct Authority
FINRA Financial Industry Regulatory Authority GAN
Generative Adversarial Networks
GDPR General Data Protection Regulation
IDE Integrated Development Environment
LDA Latent Dirichlet Allocation
LLM Large Language Models LSE
London Stock Exchange MDP
Markov Decision Process
MiFID Markets in Financial Instruments Directive ML
Machine Learning
NER Named Entity Recognition NLP
Natural Language Processing
NMF Non-negative Matrix Factorisation
NMT Neural Machine Translation
NN Neural Network
PCS Principal Component Analysis
RegTech Regulatory Technology
RL Reinforcement Learning
RPA Robotics Process Automation
SME Small and Medium-sized Enterprises
SMT Statistical Machine Translation
SVM Support Vector Machines
VAE Variational Autoencoders
XAI Explainable AI

3
Executive Summary
This report examines Artificial Intelligence (AI) in the financial sector, outlining its potential
to revolutionise the industry and identify its challenges. It underscores the criticality of a
well-rounded understanding of AI, its capabilities, and its implications to effectively leverage
its potential while mitigating associated risks.
In its various forms, from simple rule-based systems to advanced deep learning models, AI
represents a paradigm shift in technology's role in finance. Machine Learning (ML), a subset
of AI, introduces a new way of processing and interpreting data, learning from it, and
improving over time. This self-learning capability of ML models differentiates them from
traditional rule-based systems and forms the core of AI's transformative potential.
The potential of AI potential extends from augmenting existing operations to paving the way
for novel applications in the finance sector. The application of AI in the financial sector is
transforming the industry. Its use spans areas from customer service enhancements, fraud
detection, and risk management to credit assessments and high-frequency trading. The
efficiency, speed, and automation provided by AI are increasingly being leveraged to yield
significant competitive advantage and to open new avenues for financial services.
However, along with these benefits, AI also presents several challenges. These include issues
related to transparency, interpretability, fairness, accountability, and trustworthiness. The use
of AI in the financial sector further raises critical questions about data privacy and security.
Concerns about the 'black box' nature of some AI models, which operate without clear
interpretability, are particularly pressing.
A pertinent issue identified in this report is the systemic risk that AI can introduce to the
financial sector. Being prone to errors, AI can exacerbate existing systemic risks, potentially
leading to financial crises. Furthermore, AI-based high-frequency trading systems can react
to market trends rapidly, potentially leading to market crashes.
Regulation is crucial to harnessing the benefits of AI while mitigating its potential risks.
Despite the global recognition of this need, there remains a lack of clear guidelines or
legislation for AI use in finance. This report discusses key principles that could guide the
formation of effective AI regulation in the financial sector, including the need for a risk-
based approach, the inclusion of ethical considerations, and the importance of maintaining a
balance between innovation and consumer protection.
The report provides recommendations for academia, the finance industry, and regulators.
For academia, the report underscores the need to develop models and frameworks for
Responsible AI and the integration of AI with blockchain and Decentralised Finance (DeFi).
It calls for further research into how AI outcomes should be communicated to foster trust and
urges academia to lead the development of Explainable AI(XAI) and interpretable AI.
The finance industry players are advised to be cognizant of data privacy issues when
deploying AI and to implement a robust 'human-in-the-loop' system for decision-making.
Emphasis is placed on maintaining an effective governance framework and ensuring

4
technical skill development among employees. Understand the systemic risks that AI can
introduce is also emphasised.
The regulatory authorities are urged to shift from a reactive to a proactive stance on AI and
its implications. They should focus on addressing the risks and ethical concerns associated
with AI use and promote fair competition between AI-driven FinTech and traditional
financial institutions. The report advocates for regulatory experimentation to better
understand AI's opportunities and challenges. Lastly, fostering collaboration between
regulators, AI developers, and ensuring international coordination of regulations are deemed
pivotal.
These recommendations pave the way for the effective integration of AI in the financial
sector, ensuring its benefits are optimally harnessed while mitigating the associated risks.

5
1 Introduction
This report aims to study the impact of artificial intelligence (AI) on the finance sector, focusing
on its practical applications, challenges, and potential benefits for driving innovation and
competition. As a high-level concept, AI is a broad field of computer science that focuses on
creating models capable of performing tasks that typically require human-like intelligence,
such as understanding natural language, recognising images, making decisions, and learning
from data. These tasks encompass complex problem-solving abilities and human-like decision-
making, which have been a subject of interest for researchers for over seven decades (Agrawal
et al., 2019; Furman and Seamans, 2019; Brynjolfsson et al., 2021).
In recent years, there has been a significant increase in practical applications across
various industries, such as finance, healthcare, and manufacturing, thanks to advancements in
computing power, data storage, and low-latency, high-bandwidth communication protocols
(Biallas and O'Neill, 2020). One reason for AI widespread adoption in sectors like financial
services is its versatility (Milana and Ashta, 2021). A Bank of England and FCA survey in
2022 found that 72% of surveyed firms reported using or developing machine learning
applications (Blake et al., 2022).
The increased use of AI in finance can be partially attributed to intense competition within the
sector (Kruse et al., 2019). The term 'Fintech' has been coined to describe companies that use
digital technologies in their services. Compared to traditional financial institutions, fintech
companies leverage technology to offer innovative and user-friendly financial services,
including mobile payments, online banking, peer-to-peer lending, and automated investment
platforms. Since these are often more convenient, efficient, and affordable financial services
for consumers, they may disrupt traditional financial services through heightened competition,
innovation, and a focus on customer satisfaction. Although traditional financial institutions
were initially reluctant to adapt to these changes, many are now investing in digital technology
and collaborating with emerging fintech firms to stay competitive. For example, Deutsche
Bank sought to invest in supply chain financing and establish a partnership to incorporate
supply chain solutions and technologies into their offerings. They collaborated with Traxpay,
a German fintech company that provides discounting and reverse factoring solutions to its
corporate clients (Hamann, 2021). This partnership has allowed Deutsche Bank to become a
prominent player in the global supply chain financing industry. Using AI, financial institutions
can gain a competitive advantage by introducing innovative services and improving operational
efficiency (Ryll et al., 2020).
The finance industry generates a vast and constantly growing amount of data, including daily
transactions, market trends, and customer information. Such rich data can be harnessed to train
AI algorithms and create predictive models, making it an ideal domain for AI applications
(Boot et al., 2021). These applications can identify patterns and predict future trends in the
market, enabling financial institutions to make more informed decisions about investments and
other financial operations. Moreover, AI can also analyse customer behaviour and preferences,
offering tailored recommendations and personalised services (Zheng et al., 2019). By
leveraging AI applications, financial institutions can optimise operations, reduce costs, and
provide better customer service. With the increasing volume of data generated by the financial
industry, the integration of AI technology is expected to become even more prominent, leading
to more sophisticated applications and further transforming the financial landscape.
Despite the potential benefits of AI in the finance sector, industry experts and academic
research suggest that financial institutions have not been able to fully leveraged potential
(Cao, 2022; Fabri et al., 2022). This is partly because of the numerous challenges and pitfalls

6
of developing and using AI models. One of the primary concerns for customers is the issue of
data bias and representativeness, as improper use of AI can lead to discriminatory decisions
(Ashta and Herrmann, 2021). Furthermore, over-reliance on similar AI models or third-party
AI providers can worsen the situation, potentially leading to the exclusion of certain groups of
customers from the entire market rather than just a single financial institution (Daníelsson et
al., 2022). A comprehensive approach is needed to address these challenges, including
transparent audits of AI models and data sources, regular audits to assess AI algorithms'
accuracy and fairness, and engagement with customers to ensure that their concerns are being
addressed.
As AI becomes increasingly popular in the financial sector, firms face challenges related to the
explainability and interpretability of AI models (Fabri et al., 2022). Such challenges can lead
to reputational damage and a reluctance to adopt AI applications. Firms' stakeholders, including
customers, investors, and regulators, demand transparency and accountability in decision-
making processes. Lack of interpretability can make it challenging for firms to identify and
address errors or biases in their decision-making mechanisms using AI, resulting in legal and
financial consequences. Hence, explainability and interpretability are critical factors for AI's
responsible and ethical use in the financial sector (Fabri et al., 2022). Such challenges emerging
due to the use of AI in finance highlight the importance of effective model governance and
regulations to ensure ethical and responsible use of the technology (Ryll et al., 2020). Such
measures not only increase consumer trust in AI but also help financial firms avoid negative
consequences like legal liability, reputational damage, and the loss of customers. However,
stringent regulations may impose a significant burden on firms seeking to implement AI
systems, including additional costs related to data privacy, security measures, or hiring more
staff to monitor and maintain the technology, these costs may deter some firms, especially
smaller ones with limited resources, from adopting AI.
Regulators are also leveraging AI to enhance the efficiency and effectiveness of regulatory
processes. Organisations such as the London Stock Exchange (LSE) and the Financial Industry
Regulatory Authority (FINRA) are embracing AI as a means of improving their regulatory
capabilities (Prove., 2021). For example, the LSE has partnered with IBM Watson and
SparkCognition to develop AI-enhanced surveillance capabilities, while FINRA uses AI to
identify and prevent stock market malpractices (Prove., 2021). Regulators must balance the
benefits and risks of AI and ensure appropriate safeguards are in place to mitigate negative
outcomes. The recent surge in publications has shed light on various opportunities, challenges,
and implications of AI in the financial services industry (Bahrammirzaee, 2010; Cao, 2020;
Hilpisch, 2020; Königstorfer and Thalmann, 2020).
This report seeks to complement and update previous surveys and literature reviews by
achieving the following objectives: 1) summarising the key AI technology in finance services
based on research from finance and information systems studies; 2) examining the benefits of
AI use and adoption in the finance sector; 3) highlighting potential negative consequences and
threats associated with AI use in the finance sector; 4) addressing and evaluating the challenges,
5) role of regulators in addressing these unintended outcomes while exploring the use of AI to
enhance regulatory work; and 6) providing recommendations to academia, industry, and
regulators.

2 Key AI Technology in Financial Services

AI models provide a set of tools and techniques that can enhance decision-making processes,
reduce costs, and identify patterns that would otherwise require significant time and effort to
7
uncover. This section aims to present an overview of the fundamental AI technologies used in
the finance sector and delineate some of the principle applications, benefits and challenges of
this technology for both researchers and practitioners in the field.

2.1 Machine Learning

Machine learning (ML) algorithms are developed to recognise patterns in data and make
predictions or decisions without being explicitly programmed. In high-level classification, ML
is classified into supervised, unsupervised, and reinforcement learning. Supervised learning
models are trained on labelled data to predict an output variable based on input variables
(Kotsiantis, 2007), but unsupervised learning involves discovering hidden patterns or structures
in the data without any predefined output variables (Dougherty et al., 1995). Reinforcement
learning is a relatively recent development in the field. It considers that an agent learns to make
decisions based on the feedback from its environment in the form of rewards or penalties. This
section will provide an overview of the three types of ML models.

2.1.1 Supervised machine learning

Supervised learning involves developing and training a model by providing a set of input-
output pairs to learn a mapping function from the input to the output. After this phase, the
model is tested and confirmed through a testing phase and can then be used to predict the most
probable value of the target variable for unlabelled data. Several supervised learning algorithms
are used, including well-known linear and logistic regressions. The other main techniques
include the following. These techniques are used in a wide range of applications, such as
financial prediction, portfolio management, credit evaluation, and fraud detection (Jemli et al.,
2010; Krauss et al., 2017; Kvamme et al., 2018; Thennakoon et al., 2019; Zhong and Enke,
2019).
Decision trees: A tree-based algorithm is used for both classification and regression
tasks, which splits the data into smaller subsets based on the values of the input
variables and decides based on these subsets.
Random forests: An ensemble learning algorithm that combines multiple decision
trees to improve performance and avoid overfitting.
Support vector machines (SVM): A classification algorithm that finds the best
hyperplane to separate the data into classes.
Neural networks (NN): A multiple layers of interconnected nodes, each of which
performs a simple computation on its inputs, inspired by the structure and function
of the human brain, are used for both regression and classification tasks.

2.1.2 Unsupervised machine learning

The primary objective of unsupervised learning is to identify meaningful patterns or
relationships within a set of input data. This can be achieved through various techniques, such
as clustering, which groups similar data points together based on their similarity or distance
from each other (Hofmann, 2001). Another major application of unsupervised learning is
dimensionality reduction, where the algorithm finds a lower-dimensional representation of the
data that still captures its essential features (Muscoloni et al., 2017; Saeys et al., 2016; Torshin
and Rudakov, 2015). Additionally, unsupervised learning can be leveraged for anomaly
detection, where the algorithm identifies data points that deviate significantly in the data.
Unsupervised learning is used in a wide range of applications, such as fraud detection, customer
segmentation, portfolio optimisation, credit risk analysis, and market analysis (Bao et al., 2019;

8
Gomes et al., 2021; Kedia et al., 2018; Mittal and Tyagi, 2019; Umuhoza et al., 2020).
Compared with supervised learning, unsupervised learning can be less interpretable as patterns
and relationships discovered may not be straightforward and intuitive, and it is challenging to
evaluate and validate the results of the analysis as there are no predefined labels or classes to
compare against (Lee and Shin, 2020). Moreover, unsupervised learning can be more
computationally intensive than supervised learning, given the more data and more complex
algorithms required, and it can be more prone to overfitting as it does not have the same level
of guidance or constraint as supervised learning. Some of the commonly used models of
unsupervised learning include:

Principal component analysis (PCA): A technique used for dimensionality

reduction that transforms high-dimensional data into a lower-dimensional space,
while retaining as much of the original information as possible.
Association rule mining: A technique used to discover relationships between
variables in a large dataset, such as the method called Apriori.
Autoencoders: A type of NN used for unsupervised learning, trained to reconstruct
the input data. They are used for tasks such as image denoising and anomaly
detection.
Generative models: A class of models that can generate new data resembling the
input data, such as generative adversarial networks (GANs) and variational
autoencoders (VAEs).

2.1.3 Reinforcement learning

The concept of reinforcement learning (RL) is that an agent interacts with an environment to
learn how to maximise a reward signal. In this setting, the agent receives feedback from the
environment in the form of rewards or punishments, and aims to learn a policy that maximises
the cumulative reward over time, improving its performance through an iterative process of
trial and error. The RL problem can be formalised as a Markov Decision Process (MDP),
consisting of a set of states, actions, transition probabilities, and rewards (Arulkumaran et al.,
2017). The agent learns a policy that maps states to actions and uses this policy to decide which
actions to take in each state.
Reinforcement learning is less prevalent as compared with other ML approaches. However, its
uniqueness led to its irreplaceability. It is particularly well-suited to problems with unknown
optimal actions (i.e., correct answers). In financial services, it could be used for trading
execution and dynamic pricing (Culkin and Das, 2017; Nevmyvaka et al., 2006; Yu et al., 2019;
Zhang et al., 2020). RL algorithms can be grouped into model-free RL and model-based RL.
Model-based RL entails indirect learning of optimal behaviour by constructing a model of the
environment through observation of outcomes, including the next state and immediate reward
resulting from actions taken (Abbeel et al., 2006; Levine et al., 2016; Levine and Abbeel,
2014). Model-based RL tends to learn with significantly fewer samples than model-free RL,
while model-free RL is more generally applicable and easier to implement (Clavera et al.,
2018). For model-free RL, two main approaches are policy optimisation and Q-learning:
Policy optimisation: Enables an agent to learn the policy function that maps a state
to an action without using a value function (Bennett et al., 2021; Yu et al., 2019).
Q-Learning: Learns the optimal action-value function for a given MDP (Hasselt et
al., 2016; Kumar et al., 2020). It works by iteratively updating the Q-values for
each state-action pair, using the Bellman equation to estimate the expected reward
for each possible action.

9
2.2 Expert System
An expert system (ES) is a type of AI system that imitates the expert decision-making abilities
of a specific domain or field. ES utilises information in a knowledge base, a set of rules or
decision trees, and an inference engine to solve problems that are difficult enough and require
human expertise for resolution (Harmon P, King D, 1985). ES consists of three main
components (Metaxiotis and Psarras, 2003; Sotnik et al., 2022):
Knowledge base: It contains domain-specific knowledge and rules that the expert
system utilises to solve specific problems. The knowledge base is typically created by
domain experts and is organised to enable efficient access. The most used technique is
the if-then rule.
Inference engine: This expert system component uses knowledge in the knowledge base
to draw conclusions and make recommendations. It utilises a set of rules or decision
trees to guide its reasoning.
User interface: This enables users to interact with the system, ask questions, and receive
recommendations or advice. It mainly consists of screen displays, a consultation dialog
and an explanation component.

Several factors differentiate expert systems from other mathematical models (Jackson, 1986).
For example, (a) they can handle and process qualitative information; (b) inflexible
mathematical or analogue methodologies do not restrict them and is capable of managing
factual or heuristic knowledge; (c) their knowledge base of ES can be continually expanded
with accumulating experience as necessary; (d) they can deal with uncertain, unreliable, or
even missing data; and (e) they are capable of reflecting decision patterns of users.
ESs are used in various applications, such as financial prediction, credit risk analysis, and
portfolio management (Bisht and Kumar, 2022; Mahmoud et al., 2008; Nikolopoulos et al.,
2008; Shiue et al., 2008; Yunusoglu and Selim, 2013). They can be particularly useful in
situations where the knowledge required is complex and difficult to acquire or where there is a
shortage of human experts in a particular field. However, developing an expert system can be
time-consuming and expensive, and the accuracy depends on the quality of the
knowledge base and the rules used by the inference engine.

2.3 Natural Language Processing

Natural Language Processing (NLP) is a subfield of AI that enables computers to understand,
interpret, and generate human language, thus facilitating natural communication between
humans and computers (Mei, 2022; Ruffolo, 2022; Xu, 2022). NLP has many practical
applications in financial services, such as chatbots for improving customer experience. It can
help both regulators and financial firms by extracting relevant information and managing
voluminous documentation efficiently (Mah et al., 2022). Some commonly used algorithms in
NLP include:
Text Pre-processing: This involves cleaning and transforming raw text data into a
format suitable for analysis and quickly reducing the data space required. Common pre-
processing techniques include tokenisation (breaking down a text into smaller chunks)
(Forst and Kaplan, 2006; Habert et al., 1998; Webster and Kit, 1992), lemmatisation
and stemming (allowing tracing
al., 2016).
Sentiment Analysis: Useful in determining the emotional tone or sentiment in text by
analysing text data. Both supervised and unsupervised algorithms can be used for
sentiment analysis (Adamopoulos et al., 2018; Fang et al., 2014).
10
 Named Entity Recognition (NER): Identifies and classifies named entities (such as
people, organisations, and locations) in unstructured text into a set of predetermined
groups (McCallum and Li, 2003; Sang and De Meulder, 2003; Sazali et al., 2016). NER
can be performed using rule-based approaches or machine learning models.
Topic Modelling: This refers to identifying topics or themes in a corpus of text data
(Blei, 2003; Sridhar and Getoor, 2019). Popular techniques for topic modelling include
Latent Dirichlet Allocation (LDA) and Non-negative Matrix Factorisation (NMF).
Machine Translation: It involves translating text from one language to another.
Although this is not new to many people, the necessary level of quality is still not
satisfied. The main challenge with machine translation technologies is to keep the
meaning of sentences intact along with grammar and tenses (Khurana et al., 2023).
Machine translation can be done using rule-based approaches or statistical machine
translation (SMT) models, or more recently, with neural machine translation (NMT)
models.
Speech recognition: This can help covert spoken words into written text, which has
been applied in various settings such as interviews and conversations. Symbolic,
statistical or hybrid algorithms can support speech recognition.

2.4 Robotics Process Automation

Robotics Process Automation (RPA) is increasingly used in financial services to automate
repetitive, rule-based human tasks. It aims to reduce operational costs, increase efficiency,
reduce human errors, and enhance the customer experience (Driscoll, 2018; Gotthardt et al.,
2020). Some of the common applications of RPA in the financial service include:
Account opening and closing: RPA can automate the process of opening and closing
accounts, including data entry, verification, and document processing (Romao et al.,
2019).
Claims processing: RPA can automate the process of claims processing, including data
entry, verification, and validation, thereby reducing the time and cost involved in claims
processing (Oza et al., 2020).
Fraud detection and prevention: RPA can identify and prevent fraud by monitoring
transactions, identifying anomalies, and alerting the relevant stakeholders in real-time
(Thekkethil et al., 2021).
Customer service: RPA can automate customer service processes, including query
handling, complaint management, and resolution, thereby improving customer
satisfaction and experience (Kobayashi et al., 2019; Lamberton et al., 2017; Willcocks
et al., 2015).
Compliance and reporting: RPA can automate compliance and reporting processes,
including data collection, validation, and reporting, thereby reducing the risk of errors
and improving regulatory compliance (Anagnoste, 2018; Radke et al., 2020; Sibalija
et al., 2019).

3 Benefits of AI use in the Finance Sector

Researchers and practitioners have outlined numerous benefits of using AI in finance.
However, they have cautioned that the realisation of these benefits depends on the scale of the
organisation, with large financial organisations benefiting more than smaller ones (Ashta and

11
Herrmann, 2021). This reflects financial institutions' different capabilities and resources and
the scope of their services.

3.1 Improving Decision-making Process

One of the biggest benefits of using AI in the finance sector is improving decision-making
related to credit assessment, lending, and investment. By harnessing the vast amount of data
generated by financial institutions, AI models are being used to automate and augment
decision-making. This allows for a more accurate, faster, and informed decision-making
process. For example, AI-based models are increasingly used for automated decision-making
in lending (Königstorfer and Thalmann, 2022), as they can significantly improve the credit risk
assessment of a loan applicant due to their reliance on diverse, often non-traditional, data sets.
When AI and big data are used together, they can detect weak signals, such as interactions or
non-linearities across explanatory variables, which seem to increase prediction accuracy
compared to traditional creditworthiness metrics. This results in estimates of increased
economic growth at the macroeconomic level. AI disrupts the banking industry because it
allows for the utilisation of more varied types of data that can produce more accurate credit
risk projections. AI can manipulate "big data" gathered due to consumer behaviour, the
digitisation of customer contacts, and information made available through sources like social
networks.
The effect that AI-based models (using large amounts of data produced by social media
activity) might have on the quality of credit scores has been the subject of theoretical analysis
by researchers (Wei et al., 2016). Researchers have concluded that they might backfire by
causing strategic changes in the behaviour patterns of potential borrowers on social media sites.
For instance, they can limit their social media connections or prioritise relationships with
people in socio-professional groups, such as civil servants, who are less susceptible to losing
their jobs. This indicates that AI-based scoring models intended to enhance prediction results
may eventually lead to behavioural adjustments that perform well according to the selected
indicators.
Empirically, there is still substantial disagreement over how AI may assist with financial
decision-making. Lenddo and Big Data Scoring, two fintech companies whose major business
is treating massive amounts of data using AI algorithms, predictably support this practise.
Because they believe that the low likelihood of payback and potentially high loan risk will not
even cover the evaluation costs, traditional banks may frequently choose not to evaluate the
creditworthiness of small borrowers (Bazarbash, 2019). In this way, the use of AI and its
capacity to handle a larger variety of data allows organisations like FinTechs to venture into
territory that has, up until now, been uncharted.
Additionally, AI-based robo-advisors have been employed to improve or fully automate
investment decision-making. For example, AI-based models can recommend investment
portfolio construction and re-construction (Ahmed et al., 2021) and efficiently drive ESG
investment targets (Ashta and Herrmann, 2021). Further, NNs have been deployed to estimate
and recommend optimal investment strategies (Chen and Ge, 2021). Given their risk appetite,
robo-advisors have also been utilised to help lenders decide on the most optimal P2P loan
investments (Ge et al., 2021). Most lenders on P2P lending platforms such as Lending Club are
relatively young and unexperienced. To avoid making bad investment decisions these
platforms are increasingly relying on robo-advisors.

12
3.2 Automating Key Business Processes in Customer Service and Insurance
Financial institutions have benefited from automating key business processes using RPA
algorithms (Wittmann and Lutfiju 2021). Such RPA are mainly used as robo-advisors to
support customer services in retail banking and, more recently, in wealth management (Kruse
et al., 2019). Such robo-advisors can offer automated financial planning services like tax
planning guidance, opening a bank account, recommending insurance policies, giving
investment advice, and many other essential financial services.
Banks can make easy wins in key areas such as untapped client segments, lower acquisition
costs, stronger usage of existing products and services, and improved access and scale by
adopting an AI-first approach to customer interaction (Mckinsey, 2020). Customers are asking
for financial services to be delivered with a wider range of goods and services anytime,
anywhere (Zeinalizadeh et al., 2015). Financial organisations can no longer ignore the
extraordinary advantages of integrating and utilising Robotic Process Automation (RPA)
solutions in their environment. Data collection enhances the user experience and offers
numerous benefits to customers by creating the impression that AI interactions are on par with
those of humans. By providing personal data, consumers can obtain personalised services,
information, and entertainment, frequently for little or no cost.
Access to personalised services also suggests that users will benefit from the choices made by
digital assistants, which successfully match preferences with accessible possibilities without
subjecting users to the cognitive and affective exhaustion that can come with decision-making.
To maintain their competitive advantage and boost profitability, banks are placing increasingly
more strategic importance on RPA. The main advantage of using RPA services in retail banking
is that it allows banks to operate around the clock, deliver cutting-edge services, and improve
client experiences while increasing efficiency and accuracy (Villar and Khan, 2021). The
sharing economy has developed to give consumers more power. Real-time analytics and
messaging require the end-to-end integration of internal resources to fully utilise those
potentials. Banks must modernise their IT architecture and analytical skills to acquire, process,
and accurately analyse client data.
Similar automation is also often used in insurance, for example, in the pre-validation of pre-
approved claims. Dhieb et al. (2019) propose an automated deep learning-based architecture
for vehicle damage detection and localisation. Cranfield and White (2016) explain how
insurance claims outsourcing and loss adjusting firm managed to implement RPA (Robotic,
Cognitive robotic, AI), leading to a team of just four people processing around 3,000 claims
documents a day. Thus, robo-advisors have been helping insurers collect information about
claims and process the gathered information quickly.

3.3 Algorithmic Trading Improvement

Another key area where AI has been applied is trading, such as equity trading and, more
recently, trading in the foreign exchange market. Algorithmic trading has proven to be more
efficient than human traders as, relying on diverse, real-time data sets, it can factor in market
anomalies and account better for price differences (Cohen, 2022). AI-models can also send
better trading signals to human traders, as they can identify unexpected market trends within a
limited time. Unlike human traders, such models do not rely on sentiments, which have been
proven to cloud the judgement of human traders. Thus, AI models used for equity trading can
be more efficient than human traders.
Recognising the potential of AI to perform high-frequency trading, many researchers and
practitioners have tried to incorporate advanced AI models, relying on, for example, neural

13
networks and fuzzy logic to advance algorithmic trading. There are many examples of these
AI-based models, such as an AI-model based on a reinforcement learning algorithm that can
improve stock trading (Luo et al., 2019), an NLP-
to predict stock trading returns (Martinez et al., 2019), and fuzzy logic models for predicting
trends in financial asset prices (Cohen, 2022).
Recently, researchers have also become interested in studying the impact of social media data
on stock performance. The proliferation of advanced AI and ML techniques has facilitated this
research. For example, Valencia et al. (2019) developed a ML to analyse how Twitter data can
predict the price movements of several cryptocurrencies, such as Bitcoin, Ethereum, Ripple,
and Litecoin. Similarly, Wolk (2019) has shown that advanced social media sentiment analysis
can be used to predict short-term price fluctuations in cryptocurrencies.

3.4 Improving Financial Forecasting

Forecasting models are the most widely developed AI-based models in the finance sector.
These models utilise diverse sets of traditional and non-traditional data compared to traditional
techniques. For example, Óskarsdóttir et al. (2018) use smartphone-based data in combination
with socio-demographic data to forecast consumer loan default. By utilising a diverse data set,
AI-based forecasting models produce better, richer insights within a limited amount of time,
thus vastly reducing the time and cost of producing a forecast.
Various AI-based models predict companies' bankruptcy, loan defaults of small and medium-
sized enterprises (SMEs), and stock price fluctuations (Ahmed et al., 2022). For instance,
Sigrist and Hirnschall (2019) developed a model to predict the default of SMEs in Switzerland.
Li and Mei (2020) utilised deep learning neural network with two hidden layers to predict asset
returns, while Ruan et al. (2020) employed ML models to forecast stock market returns based
on investor sentiment. Petrelli and Cesarini (2021) combined different artificial intelligence
techniques to predict high-frequency asset pricing. Furthermore, credit risk forecasting models
have also been developed by Sigrist and Hirnschall (2019).
AI models have also been used for insurance claim prediction. A claim is a request that the
insurer s business pay for an occurrence that is covered by a policy, such as a car accident, a
house fire, or a trip to the ER. An insurance claim is a request for reimbursement for unlucky
occurrences like a car accident, medical emergency, or house fire. An insurance client may
request an explanation of why their claim was refused by using AI to predict insurance claims
(Rawat et. al, 2021). A first-party insurance claim must be made during an accident or other
occurrence. An automobile insurance claim, for instance, can demand payment for property
damage, personal injury, or accident-related medical costs. Both feed-forward and recurrent
neural networks make up these neural networks. Annual claims are predicted using these
methods. After training the test data, one may examine the association between the test data
and standardised data. They also built model accuracy and stopping criteria into constructing
these models.

3.5 Improving Compliance & Fraud Detection

AI-based models can help financial institutions achieve compliance faster, in real-time, and
with limited resources (Ashta and Herrmann, 2021; Deshpande, 2020; Fabri et al., 2022). Such
models enable real-time monitoring of data, which is paramount for the timely detection of
suspicious activities and considered one of the main benefits of using AI for fraud detection.
AI can also make financial regulatory reporting more efficient by uncovering previously
undetectable patterns (Kerkez, 2020). Trained on large sets of historic data, models utilising

14
ML techniques can find hidden fraud patterns by considering non-traditional financial data
(Milana and Ashta, 2021). Reinforced ML will most likely be used to model unusual financial
behaviour (Canhoto, 2020; Milana and Ashta, 2021). For example, AI-based models can
recognise fraud patterns by studying annual financial statements to identify the risk of financial
irregularities within an organisation (Wyrobeck, 2020). ML techniques can also be used
successfully for identifying money laundering activities (Ahmed et al., 2022).

3.6 Reducing Illegal Insider Trading

Insider trading in the stock market involves trading based on non-public information, which
can be legal or illegal. Insider trading is legal, providing that it adheres to specific regulatory
guidelines. On the other hand, illegal insider trading occurs when trading is conducted based
on non-public information, such as private, leaked, or tipped information, before it is made
public (Varma and Mukherjee, 2022; Islam, 2018). This could include information on new
product launches, quarterly financial status, or acquisition and merger plans.
AI can be used to detect potential insider trading around price-sensitive events. This is achieved
through clustering to identify discontinuities in an investor s trading activity and identify small
groups of investors that act coherently around such events, indicating potential insider rings
(Mazzarisi et al., 2022).
Such AI-based systems can help financial industries and regulators stay ahead of insider trading
and other financial crimes, ensuring that the financial sector remains transparent, fair, and free
from illegal activities. Ultimately, using AI to detect and prevent illegal trading can help
maintain the integrity of the financial markets and protect investors from potential losses.

3.7 Reducing Operational Costs

Scholars have pointed out that the adoption of AI by financial institutions can lead to a
reduction in operational costs. For example, AI can reduce loan default rates because, due to
robust credit score assessment, financial institutions and FinTech lenders can improve
customer targeting (Königstorfer & Thalmann, 2020). Due to automation, AI can contribute to
reducing costs in terms of, for example, compliance (Kerkez, 2020) and detecting financial
fraud (Ashta and Herrmann, 2021). Similarly, using AI chatbots can help reduce labour costs
(Patil and Kulkarni, 2019). Further, they can also improve speed in providing customer service
(shorter time) and availability (a robot can run 24/7 and has no sick leave) (Wittmann and
Lutfiju 2021).
AI can also be used to reduce the cost of transactions. By minimising the involvement of
humans, AI can be utilised to increase the speed and efficiency of the payment process. By
automating workflows, offering decision support, and applying image recognition to
documents, AI can enable the straight-through processing of payments (Barclays, 2019).
Typically, businesses may use RPA to automate low-value jobs to scale up the advantages
through the number of transactions processed. However, the advantages naturally end at a
certain point. AI systems can learn, foresee, and anticipate based on available knowledge and
past data, whereas RPA systems can validate, analyse, compile, calculate, and orchestrate
repetitive and rule-based activities. In order to cut costs and boost revenue potential, firms can
deliberately design intelligent automation systems by taking a holistic view of end-to-end
processes (Deloitte, 2020).

3.8 Improving Financial Inclusion

Millions of previously uninsured and underserved poor people are moving away from cash-

15
based transactions and into formal financial services, where they can access a range of services
like payments, transfers, credit, insurance, stocks, and savings (WorldBank, 2020). The
problem of information asymmetry between financial institutions and individuals can be solved
by providing digital financial inclusion through AI access to various social networks and online
shopping platforms, which generate a wealth of personal data (Yang and Youtang, 2020).
People might be able to access credit, save money, make deposits, withdraw, transfer, and pay
for goods and services using a mobile device with AI intelligence. This enables those with
modest incomes to obtain services that are not available to them through the traditional banking
system (Van and Antoine, 2019).

3.9 Strengthening Cybersecurity Resilience

Financial organisations have been traditionally exposed to many cybersecurity attacks, which
have recently increased. AI can be used successfully to further strengthen financial
organisations' cybersecurity resilience. For example, AI techniques can provide better
protection from social engineering attacks such as phishing. The primary use of NLP in cyber
security for the insurance sector will be to promote interactions between humans and machines
(Ursachi, 2019). Insurance companies may use NLP to search enormous databases for email
conversations to detect the possibility of a phishing attempt. NLP can spot harmful behaviour
patterns by monitoring all emails that enter the organisation s network (Mansour, 2020). For
instance, if an email containing hazardous links, files, or malware were to infiltrate the
networks, NLP could scan and analyse the email, delete it from the system, and ensure it has
no negative effects on any important procedures or data.

3.10 Takeaways from the Insurance Sector

The necessity for profitability, financial regulation, and company competition all contributed
to adopting AI in financial services (Akyuz and Mavnac o, 2021). It has been noted that before
using AI, a few important obstacles need to be considered. Information asymmetry is a serious
issue because algorithms that receive insufficient data can produce inaccurate predictions with
direct and indirect implications in financial engineering and decision-making (Jan, 2021). Any
new technology can potentially expand the threat surface, raising the organisation s risks. Data
privacy has grown to be a major problem regarding the usage of AI in the industry due to the
extensive use of data for forecasts. Additionally, there is concern that substantial investments
in new technology may drive up insurance costs, making it unaffordable for economically
disadvantaged groups. AI makes data administration easier, which is its key benefit in the
insurance industry. This might also be advantageous for the finance sector, given how data is
processed for decisions in both industries.
Machine learning can be used to organise unstructured and semi-structured datasets. Scholars
and data analysts can use datasets from different insurance companies. In the insurance sector,
machine learning may be used to more accurately anticipate risk, claims, and consumer
behaviour (Vandrangi, 2022). Artificial intelligence has also been used to power conversation
interfaces that intelligently present clients with different types of information. These interfaces
use current data, machine learning, and natural language processing. Chatbots receive natural
language data from previous customer encounters, which an intelligent system evaluates and
instantly uses to learn how to reply to users in text. In the insurance sector, artificial intelligence
may be applied in various ways, including responsive underwriting, premium leakage, cost
control, arbitration, litigation, and fraud detection (Pirilä et al, 2022). Strong artificial
intelligence techniques are being incorporated into insurance data to handle this issue in great
depth.

16
4 Threats & Potential Pitfalls
While AI can offer significant benefits for the financial industry, researchers and practitioners
point out that the use of AI comes with many threats/potential pitfalls. Therefore, organisations,
users and regulators must remain cognizant and vigilant of the potential drawbacks associated
with using AI to ensure that this technology is utilised fairly and efficiently.

4.1 Explainability and Transparency of AI-based Models

The decision-making process of AI models is often compared to a black box that lacks
transparency (Buckley et al., 2021; Milana and Ashta, 2021; Ryll et al., 2020). As a result,
users are unable to comprehend how the system operates, makes decisions, and the underlying
reasons behind those decisions. This creates a challenge in identifying errors and biases in the
system, which may result in inaccurate and unjust decisions. Furthermore, the lack of
transparency can constrain the ability to enhance the AI system over time. Without an
understanding of its operations, it becomes difficult to identify areas for improvement or
optimise performance, limiting the potential of AI to its fullest extent. Thus, one of the major
pitfalls is incorporating transparency into AI solutions to unlock its full potential, including
increased efficiency, accuracy, and cost savings.
Researchers have called for increased explainability of AI models to address this lack of
transparency. However, while academic research is striving to develop explainable and
interpretable AI models, incorporating explainability in AI models can result in lower
efficiency (Adadi and Berrada, 2018) and higher costs when applying AI models (Fabri et al.,
2022). Furthermore, it remains unclear among researchers and industry experts what constitutes
a satisfactory explanation concerning explainable AI. The varying explanations required by
different stakeholders further complicate the development and comparison of different
explanations (Fabri et al., 2022).
These issues can have negative consequences for the financial industry in multiple ways. First,
if the decision-making process of an AI system is obscure, it becomes challenging to detect
and rectify errors or biases in the system, potentially resulting in flawed or biased decisions
with detrimental outcomes. Second, if people cannot comprehend how an AI system arrived at
its conclusions, they may be less inclined to trust the system, leading to limited adoption and
effectiveness. Third, using AI may heighten systematic risk as more financial firms implement
the same tools and algorithms (Daníelsson et al., 2022). In such cases, opacity hinders the
proper modelling and monitoring of such risks, raising the likelihood of market crashes.

4.2 Fairness of AI-based Models

AI systems can replicate and amplify biases present in the data used to train them. Failing to
conduct a comprehensive investigation of the data utilised to train AI models could result in
outliers and spurious patterns in the data leading to AI models producing inaccurate and biased
decisions that perpetuate existing biases and discrimination in society. Moreover, historical
data largely used for AI and ML training have inherent limitations in fully representing the
future, particularly when crucial extreme events are absent from the available financial data.
This increases the likelihood of AI model failures during a crisis.

4.3 Lack of accountability for AI Output

One of the main pitfalls in using AI systems within financial organisations is the lack of
accountability for AI output. This becomes particularly problematic when AI is employed to

17
make critical decisions with important implications, such as assigning falsely bad credit scores,
which can deny access to a loan. In cases where such AI-based critical decisions are made
based on inaccurate training or biased and unrepresentative data, it is challenging to determine
who is accountable for these decisions (Ashta and Herrmann, 2021; Fabri et al., 2022).
Machine learning techniques and associated artificial intelligence technologies use historical
training data to determine how to respond in various circumstances. They frequently update
their databases and educational materials in response to fresh knowledge. Two significant
issues arise when attempting to raise awareness of these technologies, which must be
considered. First, decisions are made automatically without human involvement, and mistakes
cannot be tracked. Second, the justification for a decision s formulation might not always be
clear to auditors (DRCF, 2022).
AI is used in a variety of processes, including damage assessment, IT, human resources, and
legislative reform. AI systems can quickly pick up on petitions, policies, and changes made
due to those policies. They are also quick to decide. This strategy raises concerns about
security, social, economic, and political dangers and decision-making accountability. This
further erodes trust in AI-based systems and reinforces the need for AI transparency and
explainability.

4.4 De-skilling of employees in the financial sector

The development of advanced AI techniques, coupled with the increased availability of data,
has resulted in a growing number of companies and individuals becoming attracted to AI and
utilising it in their operations. However, excessive reliance on AI can present various risks. For
instance, it can diminish human skills (Milana and Ashta, 2021) and discourage people from
developing the necessary skills to make decisions independently. Researchers, for example,
have pointed out that human skills related to financial forecasting, planning and decision
support will soon be in less demand as financial organisations adopt more AI systems (Kruse
et al., 2021). At the same time, in other areas within the finance sector, employees will be
undergoing upskilling to train how to work more efficiently and safely with AI.

4.5 Job Displacement

Implementing AI on a large scale in the financial sector, particularly in commercial banks, will
likely result in job displacement for many workers. As automation of routine tasks replaces
human tasks, financial institutes will require fewer employees, with fewer recruitment drives
and the potential for early retirements or even layoffs. This could lead to discontent among
bank employees, resulting in productivity losses that could offset some of the gains from
technological advancement (Juneja, 2021).
AI also has the potential to automate many non-routine tasks that humans currently perform.
This could result in significant changes to labour demand, job polarisation, and inequality. For
example, there may be a shift towards stronger relative employment growth in high-paid or
low-paid occupations, depending on which AI automates non-routine tasks. This shift could
lead to significant changes in the workforce and potentially exacerbate existing inequalities,
leading to economic instability (WhiteHouse, 2022).
While there is a growing demand for AI skills in the finance sector in the UK, particularly in
financial trading, projections over the next 5, 10, and 20 years indicate significant estimated
net employment reductions. This raises concerns about job displacement and the need for
financial institutions and policymakers to develop strategies to mitigate any negative
consequences (BEIS, 2021).
18
4.6 Data Privacy Challenges
Based on the survey conducted by (Kruse et al., 2019), the financial service industry is
apprehensive about losing control of their data, which is a valuable asset for their business. If
financial institutions lose control of their data, it can lead to significant financial losses, legal
liabilities, and damage to their reputation. However, collecting and storing a large amount of
data can pose challenges to data protection. Hence, it is imperative to ensure that data is
collected and processed in compliance with relevant data protection regulations (Lee and Shin,
General Data Protection Regulation (GDPR) and other
industry-specific regulations, and to implement appropriate security measures to safeguard
sensitive data. If data protection issues are not addressed, AI technologies can impede adoption
in the financial industry by eroding customer trust and confidence in financial institutions.
Further, it is tenable to argue that the enormous potential of technological platforms, which use
risk prediction models based on machine learning algorithms, to obtain and analyse data from
a variety of sources, such as internet searches, social media accounts, shopping, and purchase
information obtained from credit card companies, is a potential threat to user/client privacy in
the context of financial services. For example, consumers seeking auto insurance may not be
aware of the information gathered about them or the methods used to utilise it as a basis for
risk assessment (Riikkinrn et al., 2018). This data may be collected without the content
providers awareness and occasionally without their consent.
There is a chance that the data is inaccurate even though the claimant is not allowed to change
it. Such a privacy violation might have detrimental effects on customers (Davenport et al.,
2020). Some people may even deactivate their social media accounts if concerned that their
online behaviour may increase their insurance prices. The lack of a time limit on using a
person s information gleaned from a social media account or another source when assessing
risk is arguably the most worrisome problem, becoming more prevalent as credit assessment
models rely increasingly on social media data for their scores.

4.7 Systemic Risk

Although there is currently limited evidence, researchers and practitioners have warned that
using AI can increase systemic risk in the finance sector (Danielsson et al., 2021). For example,
algorithmic trading, which relies heavily on advanced AI techniques, allows an algorithm to
learn and adapt its trading strategies independently. In unstable markets, this may lead to
increased volatility, which, as financial markets are increasingly inter-connected, can create
spillover effects and increase systemic risk (Svetlova, 2022). Researchers have also warned
that the use of similar AI-based models, trained on largely similar type of financial data sets,
can significantly increase the herding behaviour among human traders due to the similarity of
the AI output, which can further destabilise the finance system (Svetlova, 2022).

4.8 High Cost of Error

Implementing AI can be very costly for organisations within the financial industry, with the
added risk of significant financial losses in the case of errors. This is especially true for
commercial banking, where loans can amount to millions. While humans have traditionally
evaluated such loans, the rise of AI means that systems will increasingly play a major role,
with humans in an ancillary position. If these systems make an error, such as disbursing a loan
to a non-creditworthy counterparty, the bank will bear the consequences (Juneja, 2021).

19
5 Challenges
While the above section on pitfalls outlines some of the important issues that may stem from
the continuous and wide-spread use of AI in the finance sector, various challenges associated
with using AI would remain. If not addressed appropriately, these challenges may slow the
adoption of AI-based systems in the finance sector.

5.1 Availability and Quality of Training Data

AI models need to be trained on a large amount of data. The more data the AI model has access
to, the more accurate and reliable its predictions and decisions will be. By training on a vast
amount of data, the AI model can learn from diverse examples, enabling it to identify more
subtle patterns and relationships that would not have been possible with smaller datasets.
Additionally, using a large dataset helps to reduce the risk of overfitting, where the AI model
becomes too specialised in the training data and fails to generalise to new data. Therefore, a
large amount of data is crucial for achieving high accuracy and avoiding overfitting during AI
training.
Although the financial industry has access to more data compared to other industries, most of
this data cannot be used to train AI models effectively. This is mainly because many established
financial service providers have not fully digitalised their business processes. This results in
insufficient amounts of digitally available data, which presents a challenge for adopting AI in
the financial sector (BoE and FCA, 2022); Cao, 2022; Kruse et al., 2019; Milana and Ashta,
2021).
Data quality is also paramount when training AI models. If the data used is incomplete,
inaccurate, biased, or inconsistent, it can negatively affect the model's performance and lead to
inaccurate or unfair predictions. Ensuring data quality is crucial for deep learning and when
data is more unstructured and sourced from multiple sources (Greenspan et al., 2016; Lee,
2017).
Another issue that erodes data availability and quality, particularly user data, for training AI
models in the finance sector is the various data privacy requirements, which are stricter for
financial institutions than for other industry sectors (Kruse et al., 2019). However, advanced
AI techniques, such as federated learning, can allow AI models to be trained on user data
without compromising user privacy (Ashta and Herrmann, 2021). Thus, financial organisations
need to ensure that the development and deployment of AI models also safeguard user privacy
requirements satisfactorily.
Therefore, it is essential to ensure that AI models' data are high quality, accurate, and
representative of the real-world situations it is intended to model. This can be achieved by
implementing data quality controls (Lee and Shin, 2020), such as data cleaning, validation, and
profiling, to ensure that the data is accurate, consistent, and bias-free. By ensuring high-quality
data, AI models can make accurate predictions and help organisations to achieve their goals
more effectively.

5.2 Use of Synthetic Data in AI-models

Given the challenges in accessing relevant, high-quality data, financial organisations and
regulators are increasingly turning to generate synthetic data. One of the key challenges
associated with using synthetic data in AI is ensuring the quality of this data. The synthetic
data must represent diverse and unbiased real-world scenarios to ensure accurate and reliable
results. Developers must carefully account for data bias during generation to avoid biased

20
models. Synthetic data can also be limited in its representativeness of real financial data, and
outliers must be considered during the generation process to avoid compromising privacy
(FCA, 2022).
Another significant challenge is ensuring security and privacy when undertaking synthetic data
generation. Synthetic data generation techniques require real data as input, which poses a risk
to consumers privacy rights. Developers must comply with data protection laws to protect
consumers privacy and avoid infringing on their rights. Adequate measures must be in place
to secure the synthetic data and prevent any unauthorised access or misuse of the data. Financial
institutions must address these challenges to effectively leverage synthetic data to develop
accurate and reliable AI models while ensuring the privacy and security of their customers
data (FCA, 2022).

5.3 Selecting the Optimal ML Model

Like the traditional statistical methods, no single AI algorithm is effective for all problems.
Using an unsuitable algorithm can result in poor performance, inaccurate predictions, or even
the inability to solve the problem. Moreover, as highlighted in section 2, there are instances
where traditional methods outperform AI models in addressing specific issues. Therefore,
selecting the appropriate algorithm involves comprehending the problem s nature and data
characteristics and understanding the strengths and limitations of various algorithms.
Organisations might employ different algorithms for different purposes based on the desired
accuracy, interpretability, and dataset nature (Lee and Shin, 2020).
The use of RPA is also challenging as most financial institutions struggle to understand where
to use RPA in their business. As listed in (Lamberton et al., 2017), targeting RPA at a highly
sophisticated process is one of the top 10 common issues in failed RPA projects, leading to
significant automation costs that could be spent on automating multiple other processes.
Partially this is because most organisations do not completely understand the capabilities of
bots or how they operate (Cooper et al., 2019). Another concern that deters these organisations
from using RPA is the protection of business processes and the flow of information between
different jurisdictions (Cooper et al., 2019).

5.4 Legacy Infrastructure

The expansion of Information Technology (IT) architectures has been ongoing since the 1980s;
however, it has not been consistently updated. As a result, many IT architectures today,
including outdated hardware and software systems, have become a burden, making it difficult
and complicated to incorporate modern AI techniques into these legacy systems
(Kalyanakrishnan et al., 2018). Consequently, this discourages the use of AI.
One reason for this challenge is that legacy systems may not possess the necessary processing
power or storage capacity to effectively train and operate AI models (Ryll et al., 2020), which
rely heavily on these resources. This can lead to longer processing times and reduced accuracy.
Further, traditionally, financial organisations have gathered data in silos, that is, in isolated IT
systems. Hence, organisations must first transfer the isolated data to shared data lakes to
facilitate significant training and modelling [Link] IT initiatives are complex,
expensive, and time-consuming, which can further slowdown the adoption of AI in financial
organisations.
Additionally, legacy systems may not be able to integrate with modern AI tools and platforms,
limiting an organisation s ability to utilise the latest AI technologies and resulting in missed
opportunities for innovation and reduced social welfare.

21
5.5 Lack of Appropriate Skills
Using AI also challenges financial organisations as most employees do not possess the
technological expertise required to effectively operate AI systems (Kruse et al., 2019). Many
AI systems require specialised programming, data analytics, and machine learning knowledge.
Without these skills, employees may encounter difficulties comprehending how to properly use
and interpret the outcomes produced by AI systems.
Furthermore, the rapid pace of advancements in AI technilogies can make it hard for employees
to stay up to date with the latest trends and optimal techniques. Hence, organisations might
need to invest in ongoing education and training programs to guarantee that their employees
have the necessary skills and knowledge to use AI systems competently.
Moreover, adopting AI can change job responsibilities and roles (Culbertson, 2018; LinkedIn.,
2017). Certain tasks might become automated, making them redundant, while others may
require fresh skills or an alternative approach to problem-solving. Thus, it is crucial for
organisations to anticipate and plan for these changes and to provide their employees with the
essential support and training to adapt to their new roles and responsibilities.

5.6 Requirement of Better Agility and Faster Adaptability

Agility and adaptability are crucial for managing the risks associated with using AI in the
financial industry (Thowfeek et al., 2020). As previously discussed, these risks include data
bias, security and privacy concerns, and the opacity of AI models. Companies must be agile
and adaptable in addressing these risks, which can have significant consequences for the
business and its customers. Also, the use of AI can result in increased competition in the
financial industry, as companies with advanced AI capabilities can quickly gain an advantage
over those lacking such technology. To remain competitive, companies must be agile and
adaptable in adopting and incorporating AI into their operations. Furthermore, using AI can
lead to changes in how businesses operate and make decisions, which can require adjustments
to existing processes and structures. Therefore, agility and adaptability are crucial for
effectively leveraging the benefits of AI in the financial industry.

5.7 AI Model Development Challenges

While AI techniques have advanced significantly in recent years, financial organisations still
struggle to develop accurate, well-performing AI models. Apart from the issues outlined above,
some challenges relate to the nature of the AI techniques.
For example, NLP is used in sentiment analysis for predicting stock prices or generating trading
signals from processing rich sets of financial text data (Osterrieder, 2023) presenting unique
challenges related to language interpretability. While humans can easily detect the same words
having different meanings by evaluating the context within which they occur, NLP models
struggle with this task (Khurana et al., 2023). Similarly, humans often use different words to
express the same idea, making it challenging to process the language and design algorithms.
Homonyms are particularly problematic for question-answering and speech-to-text
applications because they are not in written form (Khurana et al., 2023).

22
6 Regulation of AI and Regulating through AI
Given the rapid proliferation of AI-based systems in the finance sector and the threats/pitfalls
they may create for individuals, organisations and society, regulators across various
jurisdictions have been investigating how and to what extent they should regulate the use of AI
in the finance sector. To better understand this emerging technology and its benefits, as well as
its associated threats, regulators have also become increasingly interested in harnessing its
innovation potential for regulatory work. This has led to the emergence of algorithmic

decision-
issues: regulation of AI and regulating through AI (Ulbricht and Yeung, 2021; Yeung, 2017).
To address this, regulators use regulatory testing of AI, which we explore in detail below.

6.1 Regulation Of AI
A fundamental aspect of good financial regulation is enhancing public trust by ensuring
markets function well. Consumers should feel safe investing in financial offerings that suit
them without the risk of being defrauded or even misinformed when making the investment.
For regulators to continue maintaining consumer trust in the market in the context of a changing
technology innovation landscape, there is a responsibility to be cognizant of how markets are
evolving and keeping sight of risks that could emerge for consumers when adopting emerging
technologies such as AI along the way to establish the right safeguards.
Digital technologies and data have disrupted entire industries and in many cases, have brought
about products and business models that do not fit well with existing regulatory frameworks
particularly in finance, transport, energy, and health industries. Regulation in these industries
is paramount to safeguard safety and quality standards to ensure the ongoing provision of
critical infrastructures (OECD, 2019). It is now ever more difficult to know what, when, and
how to structure regulatory interventions in a rapidly evolving technological landscape with
immense disruptive potential (Fenwick et al., 2017).
Four main regulatory considerations arise from the growing adoption of new technologies:
Consumer protection: What are the implications for consumers, especially regarding
how their data might be used in the provision of offerings leveraging new technologies,
as well as the risks around investing in the offering.
Competition concerns: What are the implications for the diverse players in the market,
especially smaller firms looking to compete with well-established tech firms that start
providing financial services.
Market integrity: What are the implications for financial stability, especially if many
consumers start investing in risky, unregulated offerings without being subject to
protections.
Operational resilience: What are the implications for the financial market
infrastructure, especially in operational disruption or large-scale cyberattacks within a
rapidly growing dependence on technology. The Covid-19 pandemic showed the
importance of ensuring operational resilience to protect consumers and market
integrity.

With these considerations in mind, deciding on the scope of any AI regulation is not a simple
task, as evident by the cautionary approaches undertaken by regulators around the globe. One
of the challenges is related to the established principle of technology neutrality, which

23
2007, p. 264). Already, however, certain regulations have focused specifically on AI. For
example, while claiming to be technology-neutral, the European Union (EU) AI Act focuses
on one specific class of technology AI. Further, the act has been criticised for evading the
technology neutrality principle by providing a too broad definition of AI as part of its scope,
which currently encompasses AI techniques that are not considered to pose significant risks to
consumers (Grady, 2023).
Other exiting regulatory initiates have also proven that it is not easy to regulate AI use in the
finance sector, given that AI relates to a wide spectrum of analytical techniques applied across
various finance areas. The rapid development of AI techniques also makes determining the
right scope and timing of legislation problematic as regulators want to avoid overregulating,
which may stifle innovative AI use in the finance sector and thus deprive us of AI-related
benefits (see above).

6.1.1 Risk-based approach to regulating AI

Abiding to the technology neutrality principle includes focusing on the outcomes of the
technology use rather than the technology itself. However, not all outcomes of the use of AI in
finance should be regulated, as this can easily result in overregulation. The risk-based approach
to understanding how and when to regulate AI, which the EU has promoted, allows us to adopt
a more granular understanding of AI use in the finance sector and to associate those outcomes
with different levels of risk. At the heart of this approach is the desire of the regulators to
achieve an optimal balance between promoting and ensuring fair use of
AI. The risk-based approach to regulating AI allows us to move away from focusing on finding
the precise legal definition of AI, which given its broad scope of techniques, applications and
rapid development, is elusive and difficult to define, but rather focus on any negative,
unintended outcomes that such AI use may cause. Such an approach, based on the principle of
proportionality (higher risk comes with higher requirements), can help regulators avoid
overregulating AI.
On the EU level, the EU AI Act has put forward this risk-based approach to regulating AI. In
its current version, the proposal of the EU AI Act views predominantly credit scoring services
as high risks, which will be subjected to strict requirements in terms of development,
deployment, monitoring, and reporting, all of which will require human oversight and a high
degree of transparency and explainability. Ultimately, the EU AI Act, which cuts across
different industry sectors, is not expected to substantially impact other finance sector areas.
However, the extent of its influence remains uncertain and subject to interpretation.

6.1.2 Existing regulation on the use of AI in the finance sector

It is important to note that the finance sector is one of the most heavily regulated industries. As
such, robust legal rules are already in place that can address some of the challenges that AI can

provides robust rules concerning algorithmic trading. Further, various anti-discrimination laws
forbid using statistics that severely denigrate protected characteristics by posing a serious threat
of bias. This type of legal protection is illegal based on the Equality Act of 2010, which forbids
insurers from utilising algorithms that may result in discrimination based on appearance and
physical attributes. This is an undisputed and obvious point. Indirect discrimination may occur
even though the algorithms used in the risk individualisation process are not designed to
analyse physical attributes (Mann and Matzner, 2019). However, the actual outcomes of the
individualisation achieved by the algorithms would be particularly harmful to people who have
a protected attribute. This type of discrimination, sometimes known as unintentional proxy

24
discrimination, is widely believed to be inevitable when algorithms are used to look for
relationships between input data and goal variables, regardless of the nature of these
relationships (Prince and Schwarcz, 2019) For example, the programme would not
purposefully discriminate against people based on their gender. Some proxies, such as the
colour or brand of the car, on the other hand, may unintentionally reproduce biases or
unintended outcomes that a person would not deliberately incorporate into the system.
Regulators, however, have continued to evaluate whether the growing use of AI in the finance
sector can negatively impact consumer protection, competition, financial stability, and market
integrity (Bank of England and FCA, 2022).

6.1.3 The need for Human-In-The-Loop

Despite the growing efforts of regulators, several legal scholars have stated that regulatory
efforts under the form of legal frameworks, principles, laws, and other measures, may not be
sufficient to ensure the fair use of AI in the finance sector (Buckley et al., 2021; Zetzsche et
al., 2020). This is largely because AI systems are often conceived as black boxes, which makes
their regulatory supervision challenging (Buckley et al., 2021; Zetzsche et al., 2020). Instead,
legal scholars advocate that financial organisations that utilise AI should adopt strict AI internal
governance policies by promoting personal responsibility of senior managers who are
responsible for an organis -based systems. By adopting a personal responsibility,
managers can demand more transparency and explainability (at least for high-risk areas) of the
AI systems their organisations develop, deploy and use. This may also require independent AI
review committees, as Zetzsche et al., 2020 suggested.
This drive towards more AI explainability could, however, lead to a potential clash between
AI explainability as desired by senior managers vis-à-vis as preferred by regulators. Scholars
have already reported such disparity between regulators (preferring a wider scope) and
financial institutions (preferring limited scope) with regards to AI explainability (Kuiper et al.,
2020). In particular, while AI explainability may be desired, it is often expensive and can come
at the expense of the and accuracy. Thus, defining the optimal,
desirable point of the challenging.

6.2 Regulating Through AI

The continuing adoption of AI in financial services drives debate among regulators and
industry on the most appropriate approach to regulatory oversight. A key question concerns
whether existing frameworks will suffice or new measures will be required (Bank of England,
2022). In order to contribute to these discussions, regulators must develop their expertise in AI
(through algorithmic regulation). Moreover, AI has significant potential to improve the work
of regulators too, so acquiring hands-on experience is an essential activity in modern regulatory
practice. This section, therefore, explores aspects of AI use from the regulators' perspective.

6.2.1 The regulatory context

Scholars describe the role of the regulator as intentional attempts to manage risk or alter
behaviour to achieve some pre-specified goals (Black, 2014). The control mechanisms are three
core components of setting standards, gathering and monitoring information, and making
interventions or implementing sanctions to align with the desired goals (Ulbricht & Yeung,
2019). To be able to oversee the safety and soundness of their regulatory environments,
regulators need to be able to make decisions that are both timely and informed. With a move
away from the prescriptive practices of regulation and a growing emphasis within regulatory

25
regimes on targeting outcomes, there is a greater need for more and better information. These
same information resources low-
probability but high-impact events that cause the most harm to markets and consumers (Black,
2014).

6.2.2 The opportunity

An important focus for regulators in using AI is on better informed decisions, while stopping
short of automating decision-making. Improving analytical techniques, exploiting the speed,
scale, and volume of modern information processing are key enablers. The evolution in
maturity from descriptive to predictive and prescriptive analytics are helping move the time
window of oversight from what has happened to what might happen to what action could be
taken (Lepenioti, 2020). This supports a shift from reactive to proactive supervision, extending
capabilities from the core of gathering and monitoring information to providing ever greater
direction for action. As a result, the feasibility of taking preemptive action before any potential
harm arises becomes more realistic.. These analytical methods depend on good quality data
from improved data collection and processing activities. Increasingly, to pick up on leading
indicators involves capturing more varied data including less structured forms like text and
image from websites, open-source locations, and API interfaces, then structuring and
organising the data for analysis. Automation capabilities also play an important role in
strengthening the efficiency and effectiveness of regulatory workflows for the increasingly
essential end-to-end data pipelines and for the scalable execution of business operations. The
combination of established workflow and business rule technologies with the newer generation
of analytical capabilities has the potential to extend the reach of regulatory process automation.

6.2.3 The adoption of AI by regulators

The adoption of technology by regulators to supervise markets has been increasing
phenomenally. Such use of digital technology, including hardware and software, has been
termed Supervisory Technology, or SupTech. While Suptech is used in many regulatory fields,
finance is seen as a leader, and the development and deployment of technology in the financial
regulation space are becoming more widespread and more sophisticated. This change has been
fuelled by developments in AI that have significantly impacted financial regulation bodies'
collection, processing and analytics functions. A typical example of AI-driven predictive
analytics is using supervised machine learning to predict the risk of misconduct among
financial advisors, with the findings being passed to supervisors for follow-up (FSB, 2020).
Another example describes an unsupervised model that has been used to help regulators assess
whether firms have categorised the risk levels of their customers appropriately (CCAF, 2022).
A further example concerns using NLP and machine learning to compare filings against
historical patterns to flag those that may be more likely to be problematic. In all these instances,
the supervisor is provided with insights through analytics and then follows up with further
investigation. The essence of this role of AI is to help regulators work out where to look
(Toronto Centre, 2022). There is interest, too, in prescriptive analytics, but at present little sign
of active use. This position may be more behavioural than technical in origin as it aligns with
the consensus view that there has to be a human-in-the-loop. Elsewhere, there are numerous
examples of regulators exploiting AI techniques to capture and process more granular, diverse,
and timely data than can be used to provide insights that had not previously been possible
(CCAF, 2022). These insights inform various regulatory activities, including risk scoring,
triaging, monitoring misconduct, and detecting fraud. Collectively, these techniques are
showing signs of improving the efficiency and effectiveness of regulatory activities, and as
organisations with limited resources, regulators have much to gain.
26
6.2.4 The horizon
The trends for the future use of AI for regulation are already apparent. The move towards
predictive supervision, in particular, appears to be established among leading regulators and
will bring benefits to consumers and markets through quicker prevention of harm and
regulators through more efficient targeting of supervisory resources. In forecasting future ways
of working, looking at other domains for transferrable lessons is informative. Scholars of legal
practice suggest that judges' role may evolve from predicting outcomes and making judgments
to specialising more on the judgement itself while drawing increasingly on AI-developed
predictions. We could see similar patterns in regulation whereby greater use of predictions by
supervisors leads to more specialised decision-making roles (Legg, 2019). In healthcare,
machine learning techniques applied to medical imaging are helping improve the effectiveness

(Jussopow, 2022). Using could be its

parallel in regulation.
The rapid improvement in generative AI drawing on foundation Large Language Models
(LLM) also offers significant potential for use by regulators. Considering the case of
authorising firms, documentation gathered from a firm such as business plans, correspondence
and reports could be passed into a pre-trained generative model that has learned to identify
those characteristics: high risk, medium risk, or low risk. Similar techniques would be relevant
to enforcement activities whereby generative AI could be used to summarise evidence for
supervisors' review or further to highlight potential areas of concern. Recent research on the
effectiveness of LLMs has shown an improving ability to exceed the passing score in
examinations used to assess competence across a range of professional disciplines following
training on practice materials (Nori, 2023). Such AI capabilities could offer ways for regulators
to help firms better understand their responsibilities as regulated entities, whether by providing
simpler summaries of sometimes complicated legal text or natural language interfaces that
allow firms to ask questions interactively of a corpus of compliance obligations.

6.2.5 The challenges

The adoption of AI by regulators poses numerous challenges, which mirror those relevant to
the industry. Predictive supervision techniques will inevitably increase as regulators learn
lessons from leading practitioners. This will bring a risk that the efficiency of these predictive
methods leads to a gradual rebalancing in the workings of regulatory decision-making. This
may be mitigated by requiring a human-in-the-loop and ensuring specific accountabilities.
However, while human oversight will help reduce some of these challenges, there will also be
a need to determine the trustworthiness of AI-generated contributions, especially as models
become increasingly complex. Effective techniques for explainability will therefore be
increasingly important. In addition, there are risks around bias and discrimination, whereby
historical data that is not sufficiently representative has a negative influence on outcomes, then
imperceptibly perpetuates problems as it becomes the future training data. Further work will
be required to understand these concerns and to develop practical tools and solutions.
AI has great potential to improve the efficiency and effectiveness of regulatory activities. Some
of the most significant benefits can be achieved by improving data collection, processing, and
analysis. As the intelligence aspect of AI becomes more pronounced, further benefits will be
found in better supporting the higher-value work associated with regulatory decision-making.
However, regulators must always be mindful of their oversight role and ensure that they exhibit
appropriate behaviours in their use of AI.

27
6.3 Regulatory Testing of AI
In order to better understand the benefits, threats and challenges of regulating AI and regulating
through AI, a common approach explored is regulatory testing of AI. To achieve this, regulators
worldwide have begun implementing
new technologies such as AI. This focus towards regulatory testing and learning as opposed to

potential of new technologies, the need for a more responsive regulatory design, and the
growing interest in innovative products and services. The examples provided in this section
will largely be grounded in the context of financial regulation. However, many of the themes
and principles can also be extrapolated to regulation in other sectors.
Testing environments such as test beds, living labs, and sandboxes have provided an avenue
for evidence-based regulatory testing to support innovation and regulatory governance. Each
testing environment has its own distinctive features that can support regulatory decision-
making and learning by bringing in various stakeholders (Kert et al, 2022).
Digital Sandboxes are environments that provide a controlled space for experimentation,
development, analysis and evaluation. There have been a plethora of Digital Sandboxes that
regulators around the world have developed. These sandboxes have had various use cases and
seen strong industry engagement. The Monetary Authority of Singapore (MAS) recently
launched Project Guardian, a collaboration with the financial industry to explore the economic
potential of asset tokenisation (representing assets through a smart contract on a blockchain).
The FCA held a 3-week DataSprint, which convened 120 participants from across regulated
firms, start-ups, academia, professional services, data scientists and subject matter experts to
collaborate on developing high-quality synthetic financial datasets to be used by the members
of the digital sandbox pilot (FCA, 2020). Members of the digital sandbox gain access to a suite
of features such as readily accessible synthetic data assets; an Integrated Development
Environment (IDE) with access to solution development tools; an observation deck to view the
in-flight testing of solutions; a showcase page to examine solutions relating to different themes;
an ecosystem of relevant stakeholders in order to facilitate solution development from both
technological and conceptual angles; and an application programming interface (API) where
vendors can list their solutions and APIs to encourage greater interoperability and foster a
thriving ecosystem. To provide an example of the solutions that have emerged from the digital
sandbox, the firm Financial Network Analytics developed a solution that uses NNs to establish
the usual patterns of behaviour between organisations and individuals to highlight anomalies
that can be used to detect fraudulent payments (FCA, 2021).

sandbox is for firms at the early proof of concept stage, whereas the regulatory sandbox helps
firms prepare to take their services to the market. The digital sandbox can be seen as a
mechanism to support the early testing of emerging technologies using the development
features and the datasets available and forms part of the experimentation wing of the regulator.

participants access to datasets for solution development and validation. TechSprints are
regulatory-led hackathons that facilitate collaboration among experts across and outside
financial services to identify and develop solutions to key problem statements. These solutions
often form proofs of concept that regulators and industry can explore and develop further.
-and-learn approach to emerging technologies to
understand its potential in addressing challenges and unveiling opportunities. It is geared more
towards understanding the various possibilities in which emerging technologies can be
harnessed to meet desirable outcomes instead of one that is immediately ready to implement

28
and fit for purpose. Instead, it establishes the groundwork for new use cases to build upon by
understanding the art of the possible. Some notable examples of TechSprints have been on
, which shed light on the potential of Privacy
Enhancing Technologies to facilitate sharing information about money laundering and
financial crime concerns while remaining compliant with data security laws.
The two examples above, alongside other regulatory testing initiatives, have some common
themes and practices that tend to underpin them.

preference for a particular technology or solution driving the process. As opposed to building
a technological solution and finding ways to apply it, regulatory approaches tend to identify
the problems first and start considering solutions that could potentially address those problems.
Second, it is often an
focus on ensuring that consumer privacy and protection are kept at the core while exploring
the possible ways solutions can address a particular problem statement. More importantly, it
also acknowledges that emerging technologies may not offer the best solution out of a range of
options in some cases. Third, it helps explore which technologies could lend themselves to
ideration for the future.
This is especially relevant when understanding the implications of scaling up prototypes,
developing an operational tool, and maintaining it over time. Fourth, there is a strong
component of learning from other players' experiences beyond financial services, including
other regulators (BEIS, 2020). Finally, there is a forward-looking aspect which is still very
much grounded in existing tooling capabilities (BEIS, 2022), allowing regulators and other
stakeholders to explore the trajectory for technologies concerning specific use cases and,
consequently which areas could benefit from further policy guidance.
At its core, regulatory testing aims to understand how an uncertain future can impact specific
outcomes within an industry. While new technologies bring more uncertainty about their
implications, it is also worth noting that it is a challenge that has always arisen in response to
any change, and it has been met with approaches that involve scenario analysis and hypothesis
testing. As such, the principles underlying innovative approaches to testing new technologies
remain fairly similar, even if the approaches taken might become more advanced as they iterate.
Increasingly, sandboxes for specific types of technologies are becoming more widely adopted.
The EU AI Act is a recently proposed regulatory framework for AI that aims to promote the
development and adoption of trustworthy and ethical AI systems while ensuring that these
systems are developed and used responsibly and transparently. The AI Act includes several
key provisions, including requirements for risk assessment, transparency, human oversight,
and data protection.
A key element of the AI Act is the proposal for EU member states to set up national AI
regulatory sandboxes to provide a platform for companies to test their AI systems in a
controlled environment without facing the full burden of regulatory compliance. These
sandboxes aim to encourage innovation while ensuring that AI systems are developed
responsibly and safely (European Parliament, 2022).
Similarly, the European Commission has recently launched the European Blockchain
Regulatory Sandbox for innovative use cases involving Distributed Ledger Technologies
(DLT) in order to establish a pan-European framework to facilitate the dialogue between
regulators and innovators for private and public sector use cases (European Commission,
2023).
These initiatives fall under the wider bucket of anticipatory regulation and involve engaging

29
with stakeholders, monitoring trends and developments in the market, developing new
regulatory frameworks and sandboxes to support emerging technologies, promoting
collaboration between industry participants and regulators, and actively shaping the regulatory
environment to promote innovation, competition, and consumer protection (OECD, 2020;
Nesta, 2020).
In moving towards anticipatory regulation, regulators are increasingly becoming "market
makers" rather than "market fixers" (Mazzucato, 2016). This concept was introduced by
economist Mariana Mazzucato, who argues that regulators should take a more proactive role
in shaping markets rather than simply responding to market failures or crises.
According to this concept, regulators should promote innovation and investment in key areas,
such as green technologies, healthcare, and education, by providing the necessary
infrastructure, funding, and regulatory frameworks to support these industries. This approach
involves a greater emphasis on collaboration between regulators, industry stakeholders, and
other actors in the market rather than relying solely on top-down regulation.
In understanding the strategic rationale for the regulator in expanding into the realm of tech
exploration, the considerations need to be rooted in the regulatory objectives. Ultimately, the
regulatory objectives drive and justify the undertaking of these initiatives. Most regulators have
a mandate to protect consumers and enhance market integrity, with the UK FCA having a third
objective to promote competition in financial services in the interest of consumers. In a rapidly
changing landscape, with technologies like blockchain, AI, and quantum computing becoming
increasingly disruptive while also posing many opportunities, there is a role for the regulator
in keeping pace with these developments not just as an observer but as an active player in
channelling the use of these technologies down the right and responsible avenues.
Anticipatory regulation is not just about becoming aware of risks and developments earlier but
also carries - t. Beyond more formalised procedures of standards
and legislation, the act of regulatory signalling in itself has a market-making component.
Through initiatives such as a digital sandbox programme, or a TechSprint initiative, the
regulator can signal that they would like to see more innovation in a specific area while actively
providing guidance and policy steers. In signalling their appetite to encourage innovation and
help provide the right environment for firms to optimise their development, regulators can
actively shape the currents of innovation while learning about new technologies. As such, while
regulatory innovation had started primarily serving a learning purpose, it can also become an
influencing force.
In conclusion, the idea of regulators as "market makers" is particularly relevant in emerging
technologies, such as AI, blockchain, and fintech. These technologies are rapidly transforming
the financial industry and creating new opportunities for innovation, but they also raise
important regulatory challenges, such as data privacy, cybersecurity, and consumer protection.
As Ramos and Mazzucato (2022) note, while AI applications can improve the world, with no
effective rules, they may create new inequalities or reproduce unfair social biases. Similarly,
market concentration may be another concern, with AI development being dominated by a
selected
underpinned by sound regulation
In doing so, they note that the key is to equip the policymakers to manage how AI systems are
deployed rather than always playing catch up.
As the technological landscape evolves, the role of the regulator and the parameters within
which it operates will also become increasingly blurry. There will be a need for guidance from
the regulator around best practices, standards, and ethical considerations concerning new

30
technologies. Consequently, regulatory experimentation will only increase in the future and
become more collaborative and data-led. Ultimately, if a regulation has to become more
forward-
helping create the rules and parameters of the game.

7 Recommendations
7.1 Academia
Academia has a strong role to play in supporting the regulation of AI and the research and
development of AI to support financial regulation. Key to this role will be the active
engagement with regulators and industry, and there are many good examples to build upon.
We recommend further initiatives to support collaboration, including cross regulatory-
academic secondments to understand the ways of working and share learnings, as has been the
case in the project that supported this report. Other recommendations for academia are:
Develop models, frameworks and recommendations for Responsible AI, which address
issues around fairness and accountability.
Propose how we can integrate AI with blockchain and De-Fi, which can improve the
efficiency of both technologies and help utilise their potential better.
Behavioural and experimental finance researchers need to investigate how AI results
and descriptions must be presented so that customers develop trust and finally perceive
the product as attractive.
Development of explainable AI and interpretable AI: the current status of Explainable
AI (XAI) requires significant time to run and is expensive. The Defense Advanced
Research Projects Agency (DARPA) has invested 50 Million USD and launched a 5-
-
-in-the-
Focus on developing AI-based models combining different AI techniques while
factoring in human intelligence. Scholars have agreed that combining AI techniques
can create more accurate models, strengthening trust in AI systems.
Developing AI models which adequately address issues concerning explainability and
transparency (see Milana and Ashta, 2021, for example)

7.2 Industry
Despite the outlined benefits of AI for the finance sector, industry reports, and academic studies
as cost reduction and
process optimization (PwC, 2020). More opportunities lie ahead, which can be pursued more
productively if the challenges outlined above are overcome. In particular, we strongly
recommend stronger engagement with academia and regulatory bodies, especially regarding
emerging technologies, projects and applications and their uses. Knowledge sharing (with
appropriate commercial and regulatory safeguards) will advance the market and society. We
also make the following recommendations to financial organisations:
Be aware of data privacy challenges when developing and deploying AI models. Be
aware of the unintended consequences and potential pitfalls associated with using AI.
Bring human-in-the-loop (intervention): this is vital for several reasons 1) Human
(that is, False Positives and
False Negatives, respectively). 2) Builds trust in machine learning models. 3) Ensures

31
 accountability for decisions. 4) Ensures adequate evidence exists to deliver
consequential regulatory actions. 5) Process privileged information and decisions safely
and securely.
Ensure effective governance framework within organisations and on industry level
(e.g., Model Risk Management and data quality validation): include effective
assessment and reviews of ML models from development to deployment. Ensure
technical skills training of employees, i.e., train employees how to use AI-based
systems and be aware of AI ethics.
Understand better the threats AI can bring regarding systemic risk to the financial
systems.
Ensure accountability, verifiability, and algorithms, data, and design process
evaluation.

7.3 Regulators
Regulators should move from a reactive to a more proactive approach to understanding
emerging technologies such as AI in terms of both opportunities and challenges. Such a
proactive approach can help regulators understand how best to regulate them. Regulatory
intervention can address the threats and challenges associated with using AI in the finance
sector.
Correct the most salient unintended consequences of the use of AI based on a risk-based
approach (regulate strictly only high risks).
Promote fair competition between FinTech using AI and traditional financial
institutions.
Strike a balance between AI overregulation and promoting AI development and use in
finance.
Understand better the opportunities and threats of AI through regulatory experimenting.
Assess the opportunities and challenges of regulating through AI.
Ensure customer protection: regulate both financial institutions and algorithm
providers.
Address ethical concerns surrounding the use of AI in the finance sector and consider
customer perception and trust when developing regulations for AI use in finance.
Foster collaboration between regulators and AI developers. This could build upon
existing mechanisms, -Private Forum
(AIPPF) or the Veritas initiative bringing together MAS and the financial industry in
Singapore to strengthen internal governance of the application of AI.
Develop a regulatory framework for data sharing that balances privacy concerns with
the need for data sharing.
Ensure international coordination and consistency of regulations for AI in finance.

32
References

Abbeel, P., Quigley, M., Ng, A.Y., 2006. Using inaccurate models in reinforcement learning,

Association for Computing Machinery, New York, NY, USA, pp. 1 8.

[Link]
Adadi, A., Berrada, M., 2018. Peeking Inside the Black-Box: A Survey on Explainable
Artificial Intelligence (XAI). IEEE Access 6, 52138 52160.
[Link]
Adadi, A., Berrada, M., 2018. Peeking Inside the Black-Box: A Survey on Explainable
Artificial Intelligence (XAI). IEEE Access 6, 52138 52160.
[Link]
Adamopoulos, P., Ghose, A., Todri, V., 2018. The Impact of User Personality Traits on Word
of Mouth: Text-Mining Social Media Platforms. Inf. Syst. Res. 29, 612 640.
[Link]
Agrawal, A., Gans, J.S., Goldfarb, A., 2019. Artificial Intelligence: The Ambiguous Labor
Market Impact of Automating Prediction. Journal of Economic Perspectives 33, 31 50.
[Link]
Ahmed, S., Alshater, M. M., Ammari, A. E., & Hammami, H., 2022. Artificial intelligence and
machine learning in finance: A bibliometric review. Research in International Business
and Finance, 61, 101646. [Link]

es in
Competitive Markets. Contributions to Finance and Accounting, Springer, Cham, pp. 327-
340.
Anagnoste, S., 2018. Setting Up a Robotic Process Automation Center of Excellence. Manag.
Dyn. Knowl. Econ. 6, 307 332.
Arulkumaran, K., Deisenroth, M.P., Brundage, M., Bharath, A.A., 2017. Deep Reinforcement
Learning: A Brief Survey. IEEE Signal Process. Mag. 34, 26 38.
[Link]
Ashta, A. and Herrmann, H., 2021. Artificial intelligence and fintech: An overview of
opportunities and risks for banking, investments, and microfinance, Strategic Change,
Volume 30, Issue 3, pp. 211-222.
Ashta, A., Herrmann, H., 2021. Artificial intelligence and fintech: An overview of
opportunities and risks for banking, investments, and microfinance. Strategic Change 30,
211 222. [Link]
Ashta, A., Herrmann, H., 2021. Artificial intelligence and fintech: An overview of
opportunities and risks for banking, investments, and microfinance. Strategic Change 30,
211 222. [Link]
Aziz, S., Dowling, M., Hammami, H., and Piepenbrink, A., 2022. Machine learning in finance:
A topic modelling approach. European Financial Management, Volume 28, Issue 2, Pages
744-770.
Baghdasaryan, V., Davtyan, H., Grigoryan, A., and Khachatryan, K., 2021. Comparison of

33
 econometric and deep learning approaches for credit default classification. Strategic
Change, Volume 3, Issue 3, pp. 257-268
Bahrammirzaee, A., 2010. A comparative survey of artificial intelligence applications in
finance: artificial neural networks, expert system and hybrid intelligent systems. Neural
Comput & Applic 19, 1165 1195. [Link]
Bahrammirzaee, A., 2010. A comparative survey of artificial intelligence applications in
finance: artificial neural networks, expert system and hybrid intelligent systems. Neural
Comput & Applic 19, 1165 1195. [Link]
Bank for International Settlements, 2021. SupTech tools for prudential supervision and their
use during the pandemic ailable at: [Link]
Bank of England (2022) DP5/22 - Artificial Intelligence and Machine Learning. Available at:
[Link]
regulation/publication/2022/october/artificial-intelligence
Bao, W., Lianju, N., Yue, K., 2019. Integration of unsupervised and supervised machine
learning algorithms for credit risk assessment. Expert Syst. Appl. 128, 301 315.
[Link]
Barclays, 2019. Artificial Intelligence AI payments: Barclays Corporate, Artificial Intelligence
AI Payments | Barclays Corporate. Available at:
[Link]
revolution/#speedandefficiencyofpayments.
Bazarbash, M., 2019. Fintech in financial inclusion: machine learning applications in assessing
credit risk. International Monetary Fund.
BEIS (2020) The Use of Emerging Technologies For Regulation, [Link]. Available at:
[Link]
data/file/926585/[Link].
BEIS, 2021. The Potential Impact of Artificial Intelligence on UK Employment and the
Demand for Skills. BEIS.
[Link]
data/file/974907/EYFS_framework_-_March_2021.pdf.
BEIS, 2022 Regulatory Horizons Council publishes New report on unlocking uk innovation,
[Link]. Available at: [Link]
council-publishes-new-report-on-unlocking-uk-innovation
Bennett, D., Niv, Y., Langdon, A.J., 2021. Value-free reinforcement learning: policy
optimization as a minimal model of operant behavior. Curr. Opin. Behav. Sci., Value based
decision-making 41, 114 121. [Link]
Berkey, R., Douglass, G. and Reilly, A., 2019.
[Link]/gb-en/insights/artificial-intelligence/ai-investments
Bezerra, P.C.S., Albuquerque, P.H.M., 2017. Volatility forecasting via SVR GARCH with
mixture of Gaussian kernels. Comput Manag Sci 14, 179 196.
[Link]
, F., 2020. Artificial Intelligence Innovation in Financial Services.
[Link]
Bisht, G., Kumar, S., 2022. Fuzzy Rule-Based Expert System for Multi Assets Portfolio
Optimization, in: Rushi Kumar, B., Ponnusamy, S., Giri, D., Thuraisingham, B., Clifton,
34
 C.W., Carminati, B. (Eds.), Mathematics and Computing, Springer Proceedings in
Mathematics & Statistics. Springer Nature, Singapore, pp. 319 333.
[Link]
Black, J., 2014. LSE Legal Studies Working Paper No.
24/2014. DOI: 10.2139/ssrn.2519934
Blake, K., Gharbawi, M., Thew, O., Visavadia, S., Gosland, L., Mueller, H., 2022. Machine
learning in UK financial services [WWW Document]. URL
[Link]
Blei, D.M., 2003. Latent Dirichlet Allocation. J. Mach. Learn. 3, 993 1022.
BoE and FCA ,2022, The AI public-private forum: Final report, Bank of England. Available
at: [Link]

Journal of Financial Stability 53, 100836. [Link]

Broby, D. et al., 2022.
Technology & Regulation, 2022. DOI: 10.26116/techreg.2022.009
Brynjolfsson, E., Rock, D., Syverson, C., 2021. The Productivity J-Curve: How Intangibles
Complement General Purpose Technologies. American Economic Journal:
Macroeconomics 13, 333 372. [Link]
Buckley, R.P., Zetzsche, D.A., Arner, D.W., Tang, B.W., 2021. Regulating artificial
intelligence in finance: Putting the human in the loop. The Sydney Law Review 43, 43
81. [Link]
Buckley, Ross P. and Zetzsche, Dirk Andreas and Arner, Douglas W. and Tang, Brian,
Regulating Artificial Intelligence in Finance: Putting the Human in the Loop, 2021. 43
Sydney Law Journal 43 (2021) , University of Hong Kong Faculty of Law Research Paper
2021/016, Available at SSRN: [Link]
Cambridge Centre for Alternative Finance, 2022. State of SupTech Report 2022
at: [Link]
[Link]
Canhoto, A. I., 2020. Leveraging machine learning in the global fight against money laundering
and terrorism financing: An affordances perspective. Journal of Business Research, in
press.
Cao, L., 2020. AI in Finance: A Review. [Link]
Cao, L., 2022. AI in Finance: Challenges, Techniques, and Opportunities. ACM Comput. Surv.
55, 64:1-64:38. [Link]
Cao, L., 2022. AI in Finance: Challenges, Techniques, and Opportunities. ACM Comput. Surv.
55, 64:1-64:38. [Link]
Cao, LB., 2021. AI in Finance: Challenges, Techniques and Opportunities. Volume 1, Issue 1
40 pages. [Link]
Chen, K., Lin, H.-Y., Yu, C., Chen, Y.-C., 2009. The prediction of Taiwan government bond
yield by neural networks, in: Proceedings of the 13th WSEAS International Conference

Stevens Point, Wisconsin, USA, pp. 491 497.

Chen, S., and Ge, L., 2021. A learning-based strategy for portfolio selection. Int. Rev. Econ.
35
 Financ. 71, 936 942. [Link]
Chrupala, G., 2014. Normalizing tweets with edit scripts and recurrent neural embeddings, in:
Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics
(Volume 2: Short Papers). Presented at the ACL 2014, Association for Computational
Linguistics, Baltimore, Maryland, pp. 680 686. [Link]
Clavera, I., Rothfuss, J., Schulman, J., Fujita, Y., Asfour, T., Abbeel, P., 2018. Model-Based
Reinforcement Learning via Meta-Policy Optimization, in: Proceedings of The 2nd
Conference on Robot Learning. Presented at the Conference on Robot Learning, PMLR,
pp. 617 629.
Cooper, L., Holderness, K., Sorensen, T., Wood, D.A., 2019. Robotic Process Automation in
Public Accounting. [Link]
Credit scoring for good: Enhancing financial inclusion with smartphone-based microlending,
Thirty Ninth International Conference on Information Systems, 2018
Culbertson, D., 2018. Demand for AI Talent on the Rise [WWW Document]. [Link]. URL
[Link]
Culkin, R., Das, S.R., 2017. Machine Learning in Finance: The Case of Deep Learning for
Option Pricing. J. Invest. Manag. 15, 92 100.
Daníelsson, J., Macrae, R., Uthemann, A., 2022. Artificial intelligence and systemic risk.
Journal of Banking & Finance 140, 106290.
[Link]
Daníelsson, J., Macrae, R., Uthemann, A., 2022. Artificial intelligence and systemic risk.
Journal of Banking & Finance 140, 106290.
[Link]
Davenport, T., Guha, A., Grewal, D. et al. How artificial intelligence will change the future of
marketing. J. of the Acad. Mark. Sci. 48, 24 42 (2020). [Link]
019-00696-0
Deloitte (2020) Calculating real ROI on intelligent automation (IA), Deloitte. Available at:
[Link]
telecommunications/[Link].
Dougherty, J., Kohavi, R., Sahami, M., 1995. Supervised and Unsupervised Discretization of
Continuous Features, in: Prieditis, A., Russell, S. (Eds.), Machine Learning Proceedings
1995. Morgan Kaufmann, San Francisco (CA), pp. 194 202.
[Link]
DRCF, 2022. Auditing algorithms: The existing landscape, role of regulators and future
outlook, [Link]. Available at: [Link]
from-the-drcf-algorithmic-processing-workstream-spring-2022/auditing-algorithms-the-
existing-landscape-role-of-regulators-and-future-outlook
Driscoll, T., 2018. Value Through Robotic Process Automation: Replacing labor--and
transaction-intensive processes with RPA can reduce costs while improving efficiency and
quality. Strateg. Finance 99, 70 72.
European Commission, 2022. European blockchain regulatory sandbox for Distributed Ledger
Technologies, European Commission. Available at: [Link]
blocks/wikis/display/EBSI/Sandbox+Project#:~:text=The%20European%20Blockchain
%20Regulatory%20Sandbox,for%20innovative%20blockchain%20technology%20soluti

36
 ons
European Parliament, 2022. Artificial Intelligence Act and regulatory sandboxes: Think tank:
European parliament, Think Tank | European Parliament. Available at:
[Link]
Fabri, L., Wenninger, S., Kaymakci, C., Beck, S., Klos, T., and Wetzstein, S. (2022). Potentials
and Challenges of Artificial Intelligence in Financial Technologies, MCIS 2022
Proceedings. 14.
Fabri, L., Wenninger, S., Kaymakci, C., Beck, S., Klos, T., Wetzstein, S., 2022. Potentials and
challenges of artificial intelligence in financial technologies. MCIS 2022 Proceedings.
Fabri, L., Wenninger, S., Kaymakci, C., Beck, S., Klos, T., Wetzstein, S., 2022. Potentials and
challenges of artificial intelligence in financial technologies. MCIS 2022 Proceedings.
Fang, F., Dutta, K., Datta, A., 2014. Domain Adaptation for Sentiment Classification in Light
of Multiple Sources. Inf. J. Comput. 26, 586 598. [Link]
FCA, 2020. Digital Sandbox Pilot: FCA DataSprint, FCA. Available at:
[Link]
FCA, 2021. Supporting innovation in financial services: The Digital Sandbox, FCA. Available
at: [Link]
FCA, 2023. Synthetic Data Call for Input Feedback Statement. Available at:
[Link]
Fenwick, Mark and Kaal, Wulf A. and Vermeulen, Erik P.M., Regulation Tomorrow: What
Happens When Technology is Faster than the Law?, 2017. American University Business
Law Review, Vol. 6, No. 3, 2017, Lex Research Topics in Corporate Law & Economics
Working Paper No. 2016-8, U of St. Thomas (Minnesota) Legal Studies Research Paper
No. 16-23, TILEC Discussion Paper No. 2016-024, Available at SSRN:
[Link] or [Link]
Ferran, E., 2023.
Journal of Financial Regulation, DOI: 10.1093/jfr/fjad001
FINRA, 2020. Artificial Intelligence (AI) in the Securities Industry
[Link]
Fischer, T., and C. Krauss., -term Memory Networks
for Fi
654 669.
Flavián, C., Pérez-Rueda, A., Belanche, D., & Casaló, L. V., 2022. Intention to use analytical
artificial intelligence (AI) in services the effect of technology readiness and awareness.
Journal of Service Management, 33(2), 293-320.
Forst, M., Kaplan, R.M., 2006. The importance of precise tokenizing for deep grammars.
Presented at the LREC, pp. 369 372.
FSB, 2020. The Use of Supervisory and Regulatory Technology by Authorities and Regulated
institutions [Link]
Fu, R., Huang, Y., Singh, PV, 2021. Crowds, Lending, Machine, and Bias. Information
Systems Research, Vol. 32, No. 1.
Furman, J., Seamans, R., 2019. AI and the Economy. Innovation Policy and the Economy 19,
161 191. [Link]
37
 Artificial Neural Networks for improved stock price prediction. Expert Systems with
Applications 44, 320 331. [Link]
Gomes, C., Jin, Z., Yang, H., 2021. Insurance fraud detection with unsupervised deep learning.
J. Risk Insur. 88, 591 624. [Link]
Gómez Martínez, R., Prado Román, M., & Plaza, C. P., 2019. Big data algorithmic trading
systems based on investors' mood. Journal of Behavioral Finance, 20(2), 227 238.
Gorenc Novak, M., Veluscek, D.,, 2016. Prediction of stock price movement based on daily
high prices. Quant. Financ. 16 (5), pp. 793 826.
[Link]
Gotthardt, M., Koivulaakso, D., Paksoy, O., Saramo, C., Martikainen, M., Lehner, O., 2020.
Current State and Challenges in the Implementation of Smart Robotic Process Automation
in Accounting and Auditing. ACRN J. Finance Risk Perspect.
[Link]
Grady, P., 2023. The AI Act Should Be Technology-Neutral. Report. Center for Data
Innovation. Can be retrieved here: [Link]
should-be-technology-neutral/
Greenspan, H., Van Ginneken, B., Summers, R.M., 2016. Guest Editorial Deep Learning in
Medical Imaging: Overview and Future Promise of an Exciting New Technique. IEEE
Transactions on Medical Imaging 35, 1153 1159.
[Link]
Güler, K., & Tepecik, A., 2019. Exchange rates' change by using economic data with artificial
intelligence and forecasting the crisis. Procedia Computer Science, 158, 316 326.
Habert, B., Adda, G., Adda-Decker, M., de Mareuil, P.B., Ferrari, S., Ferret, O., Illouz, G.,
Paroubek, P., 1998. Towards Tokenization Evaluation. Presented at the LREC, pp. 427
432.
Hamann, F., 2021. 5 fintech and bank partnerships that are generating revenue. Subaio. URL
[Link]
revenue
Hasselt, H. van, Guez, A., Silver, D., 2016. Deep Reinforcement Learning with Double Q-
Learning. Proc. AAAI Conf. Artif. Intell. 30. [Link]

Hofmann, T., 2001. Unsupervised Learning by Probabilistic Latent Semantic Analysis. Mach.
Learn. 42, 177 196. [Link]
[Link]
Huang, X., Liu, X., & Ren, Y., 2018. Enterprise credit risk evaluation based on neural network
algorithm. Cognitive Systems Research, 52, pp. 317 324.
Islam, S.R., 2018. A Deep Learning Based Illegal Insider-Trading Detection and Prediction
Technique in Stock Market. CoRR
Jackson, P., 1986. Introduction to expert systems.
Jan, C.-L. Detection of Financial Statement Fraud Using Deep Learning for Sustainable
Development of Capital Markets under Information Asymmetry. Sustainability, 2021, 13,
9879. [Link]
38
Jemli, R., Chtourou, N., Feki, R., 2010. Insurability Challenges Under Uncertainty: An Attempt
to Use the Artificial Neural Network for the Prediction of Losses from Natural Disasters.
Panoeconomicus 57, 43 60. [Link]
Jullum, M., Løland, A., Huseby, R.B., Ånonsen, G., Lorentzen, J., (2020). Detecting money
laundering transactions with machine learning. J. Money Laund. Control 23(1), 173 186.
[Link]
Juneja, P., 2021. Disadvantages of Artificial Intelligence in Commercial Banking.
Management Study Guide. [Link]
[Link]
Jussupow, E. et al., 2022. Business &
Information Systems Engineering, 64, pp. 293-309. DOI: 10.1007/s12599-022-00750-2
Kalyanakrishnan, S., Panicker, R.A., Natarajan, S., Rao, S., 2018. Opportunities and
Challenges for Artificial Intelligence in India, in: Proceedings of the 2018 AAAI/ACM

New York, NY, USA, pp. 164 170. [Link]

Kannan, S., Somasundaram, K., 2017. Autoregressive-based outlier algorithm to detect money
laundering activities. J. Money Laund. Control 20 (2), 190 202.
Kedia, V., Khalid, Z., Goswami, S., Sharma, N., Suryawanshi, K., 2018. Portfolio Generation
for Indian Stock Markets Using Unsupervised Machine Learning, in: 2018 Fourth
International Conference on Computing Communication Control and Automation
(ICCUBEA). Presented at the 2018 Fourth International Conference on Computing
Communication Control and Automation (ICCUBEA), pp. 1 5.
[Link]
Kerkez, N., 2020. Artificial intelligence and machine learning can repurpose humans, not
replace them. ABA Banking Journal, 112(6), pp. 30 32
Kert, K., Vebrova, M. and Schade, S., 2022. Regulatory learning in Experimentation Spaces,
European Commission. Available at:
[Link]
Khurana, D., Koli, A., Khatter, K., Singh, S., 2023. Natural language processing: state of the
art, current trends and challenges. Multimed. Tools Appl. 82, 3713 3744.
[Link]
Khurana, D., Koli, A., Khatter, K., Singh, S., 2023. Natural language processing: state of the
art, current trends and challenges. Multimed Tools Appl 82, 3713 3744.
[Link]
Kobayashi, T., Arai, K., Imai, T., Tanimoto, S., Sato, H., Kanai, A., 2019. Communication
Robot for Elderly Based on Robotic Process Automation, in: 2019 IEEE 43rd Annual
Computer Software and Applications Conference (COMPSAC). Presented at the 2019
IEEE 43rd Annual Computer Software and Applications Conference (COMPSAC), pp.
251 256. [Link]
Königstorfer, F. and Thalmann, S., 2022. Applications of Artificial Intelligence in commercial
banks A research agenda for behavioral finance, Journal of Behavioral and Experimental
Finance, Volume 27.
Königstorfer, F., Thalmann, S., 2020. Applications of Artificial Intelligence in commercial
banks A research agenda for behavioral finance. Journal of Behavioral and Experimental

39
 Finance 27, 100352. [Link]
Kotsiantis, S.B., 2007. Supervised Machine Learning: A Review of Classification Techniques.
Informatica 31.
Krauss, C., Do, X.A., Huck, N., 2017. Deep neural networks, gradient-boosted trees, random
forests: Statistical arbitrage on the S&P 500. Eur. J. Oper. Res. 259, 689 702.
[Link]
Kruse, L., Wunderlich, N., Beck, R., 2019. Artificial Intelligence for the Financial Services
Industry: What Challenges Organizations to Succeed.
Kruse, L., Wunderlich, N., Beck, R., 2019. Artificial Intelligence for the Financial Services
Industry: What Challenges Organizations to Succeed.
Kuiper, Ouren & van der Burgt, Joost & Leijnen, Stefan & van den Berg, Martin., 2021.
Exploring Explainable AI in the Financial Sector: Perspectives of Banks and Supervisory
Authorities.
Kumar, A., Zhou, A., Tucker, G., Levine, S., 2020. Conservative Q-Learning for Offline
Reinforcement Learning, in: Advances in Neural Information Processing Systems. Curran
Associates, Inc., pp. 1179 1191.
Kumar, M., Thenmozhi, M., 2014. Forecasting stock index returns using ARIMA-SVM,
ARIMA-ANN, and ARIMA-random forest hybrid models. International Journal of
Banking, Accounting and Finance 5, 284 308.
[Link]
Kvamme, H., Sellereite, N., Aas, K., Sjursen, S., 2018. Predicting mortgage default using
convolutional neural networks. Expert Syst. Appl. 102, 207 217.
[Link]
Lahmiri, S., and Bekiros, S., 2019. Can machine learning approaches predict corporate
bankruptcy? Evidence from a qualitative experimental design. Quantitative Finance, 19(9),
pp. 1569 1577.
Lamberton, C., Brigo, D., Hoy, D., 2017. Impact of Robotics, RPA and AI on the Insurance
Industry: Challenges and Opportunities.
Lamberton, C., Brigo, D., Hoy, D., 2017. Impact of Robotics, RPA and AI on the Insurance
Industry: Challenges and Opportunities. SSRN Scholarly Paper.
Lee, I., 2017. Big data: Dimensions, evolution, impacts, and challenges. Business Horizons 60,
293 303. [Link]
Lee, I., Shin, Y.J., 2020. Machine learning for enterprises: Applications, algorithm selection,
and challenges. Bus. Horiz., Artificial Intelligence and Machine Learning 63, 157 170.
[Link]
Lee, I., Shin, Y.J., 2020. Machine learning for enterprises: Applications, algorithm selection,
and challenges. Business Horizons, Artificial Intelligence and Machine Learning 63, 157
170. [Link]
Legg, M. & Bell, F., 2019. -
The University of Tasmania Law Review, 38(2). Available at:
[Link]
Lepenioti, K. et al., 2020.
International Journal of Information Management, 50, pp. 57-70. DOI:

40
 10.1016/[Link].2019.04.003
Levine, S., Abbeel, P., 2014. Learning Neural Network Policies with Guided Policy Search
under Unknown Dynamics, in: Advances in Neural Information Processing Systems.
Curran Associates, Inc.
Levine, S., Finn, C., Darrell, T., Abbeel, P., 2016. End-to-end training of deep visuomotor
policies. J. Mach. Learn. Res. 17, 1334 1373.
Li, W., & Mei, F., 2020. Asset returns in deep learning methods: An empirical analysis on SSE
50 and CSI 300. Research in International Business and Finance, 54, 101291.
Li, Y., Jiang, W., Yang, L., and Wu, T., 2018. On neural networks and learning systems for
business computing. Neurocomputing, 275, pp. 1150 1159
Liang, L., Wu, D., 2005. An application of pattern recognition on scoring Chinese corporations
financial conditions based on backpropagation neural network. Computers & Operations
Research 32, 1115 1129. [Link]
port [WWW Document]. URL
[Link] Report
Luo, S., Lin, X., and Zheng, Z., 2019. A novel CNN-DDPG based AI-trader: Performance and
roles in business operations. Transportation Research Part E: Logistics and Transportation
Review, 131, 68 79.
Mah, P.M., Skalna, I., Muzam, J., Song, L., 2022. Analysis of Natural Language Processing in
the FinTech Models of Mid-21st Century. J. Inf. Technol. Digit. World 4, 183 211.
[Link]
Mahmoud, M., Algadi, N., Ali, A., 2008. Expert System for Banking Credit Decision, in: 2008
International Conference on Computer Science and Information Technology. Presented at
the 2008 International Conference on Computer Science and Information Technology, pp.
813 819. [Link]
Mann, M., & Matzner, T., 2019. Challenging algorithmic profiling: The limits of data
protection and anti-discrimination in responding to emergent discrimination. Big Data &
Society, 6(2), 2053951719895805.
Mansour, K., 2020. 4 ways NLP technology can be leveraged for insurance > early metrics,
Early Metrics. Available at: [Link]
leveraged-for-insurance/
Mazzarisi, P., Ravagnani, A., Deriu, P., Lillo, F., Medda, F. and Russo, A., 2022. A machine
learning approach to support decision in insider trading detection (Metodi sperimentali di
machine learning per supportare le decisioni nella detection degli abusi di mercato)
CONSOB-Scuola Normale Superiore di Pisa. CONSOB Fintech Series,
McCallum, A., Li, W., 2003. Early results for named entity recognition with conditional
random fields, feature induction and web-enhanced lexicons, in: Proceedings of the
Seventh Conference on Natural Language Learning at HLT-NAACL 2003. Presented at
the the seventh conference, Association for Computational Linguistics, pp. 188 191.
[Link]
Mckinsey, 2020. Reimagining customer engagement for the AI bank of the future (2020).
Available at:
[Link]
0Insights/Reimagining%20customer%20engagement%20for%20the%20AI%20bank%20

41
 of%20the%20future/[Link].
Mei, L., 2022. Fintech Fundamentals: Big Data / Cloud Computing / Digital Economy. Stylus
Publishing, LLC.
Metaxiotis, K., Psarras, J., 2003. Expert systems in business: applications and future directions
for the operations researcher. Ind. Manag. Data Syst. 103, 361 368.
[Link]
Milana, C. and Ashta, A., 2022. Artificial intelligence techniques in finance and financial
markets: A survey of the literature. Strategic Change
Milana, C., Ashta, A., 2021. Artificial intelligence techniques in finance and financial markets:
A survey of the literature. Strategic Change 30, 189 209. [Link]
Milana, C., Ashta, A., 2021. Artificial intelligence techniques in finance and financial markets:
A survey of the literature. Strategic Change 30, 189 209. [Link]
Milana, C., Ashta, A., 2021. Artificial intelligence techniques in finance and financial markets:
A survey of the literature. Strategic Change 30, 189 209. [Link]
Milana, C., Ashta, A., 2021. Artificial intelligence techniques in finance and financial markets:
A survey of the literature. Strategic Change 30, 189 209.
[Link] S., Tyagi, S., 2019. Performance Evaluation of
Machine Learning Algorithms for Credit Card Fraud Detection, in: 2019 9th International
Conference on Cloud Computing, Data Science & Engineering (Confluence). Presented at
the 2019 9th International Conference on Cloud Computing, Data Science & Engineering
(Confluence), pp. 320 324. [Link]
Mokhatab Rafiei, F., Manzari, S.M., Bostanian, S., 2011. Financial health prediction models
using artificial neural networks, genetic algorithm and multivariate discriminant analysis:
Iranian evidence. Expert Systems with Applications 38, 10210 10217.
[Link]
Mor, S., and Gupta, G., 2021. Artificial intelligence and technical efficiency: The case of
Indian commercial banks. Strategic Change, 30(3), pp. 235 245.
Muscoloni, A., Thomas, J.M., Ciucci, S., Bianconi, G., Cannistraci, C.V., 2017. Machine
learning meets complex networks via coalescent embedding in the hyperbolic space. Nat.
Commun. 8, 1615. [Link]
Nesta, 2020 Anticipatory regulation, nesta. Available at:
[Link]
Nevmyvaka, Y., Feng, Y., Kearns, M., 2006. Reinforcement learning for optimized trade
execution, in: Proceedings of the 23rd International Conference on Machine Learning,
680.
[Link]
Nikolopoulos, K., Petropoulos, F., Assimakopoulos, V., 2008. An expert system for forecasting
mutual funds in Greece. Int. J. Electron. Finance 2, 404 418.
[Link]
Nori, H. et al -
10.48550/arXiv.2303.13375
Nwogugu, M., 2006. Decision-making, risk and corporate governance: New dynamic
models/algorithms and optimization for bankruptcy decisions. Applied Mathematics and
Computation 179, 386 401. [Link]
42
OECD, 2019. The role of sandboxes in promoting flexibility and innovation in the digital age
OECD, 2020 Anticipatory innovation governance: Shaping the future through proactive policy
making, OECD. Available at: [Link]
[Link]
Óskarsdóttir, M., Bravo, C., Sarraute, C., Vanthienen, J. and Baesens, B., 2019. The value of
big data for credit scoring: Enhancing financial inclusion using mobile phone data and
social network analytics. Applied Soft Computing, 74, [Link], Joerg, A
Primer on Natural Language Processing for Finance. Available at SSRN:
[Link] or [Link]
Oza, D., Padhiyar, D., Doshi, V., Patil, S., 2020. Insurance Claim Processing Using RPA Along
With Chatbot. SSRN Sch. Pap. [Link]
Patil, K., and Kulkarni, M. (2019). Artificial intelligence in financial services: Customer
chatbot advisor adoption. International Journal of Innovative Technology and Exploring
Engineering, 9(1), pp. 4296 4303
Pendharkar, P.C., 2005. A threshold-varying artificial neural network approach for
classification and its application to bankruptcy prediction problem. Computers &
Operations Research, Applications of Neural Networks 32, 2561 2582.
[Link]
Petrelli, D., and Cesarini, F. (2021) Artificial intelligence methods applied to financial assets
price forecasting in trading contexts with low (intraday) and very low (high-frequency)
time frames, Strategic Change, Volume 30, Issue 3, pp. 247-256
Pirilä, T., Salminen, J., Osburg, V.S., Yoganathan, V. and Jansen, B.J., 2022, January. The
Role of Technical and Process Quality of Chatbots: A Case Study from the Insurance
Industry. In Proceedings of the 55th Hawaii International Conference on System Sciences.
Prince, A. E., & Schwarcz, D., 2019. Proxy discrimination in the age of artificial intelligence
and big data. Iowa L. Rev., 105, 1257.
Prove., 2021. How Banks & Regulators Are Applying Machine Learning [WWW Document].
URL [Link]
PwC, 2020. How mature is AI adoption in financial services? Study. Can be retrieved here:
[Link]
[Link]
Radke, A.M., Dang, M.T., Tan, A., 2020. Using robotic process automation (RPA) to enhance
Item master data maintenance process. LogForum 16.
[Link]
Ramos, G. and Mazzucato, M., 2022. AI in the common interest, Project Syndicate. Available
at: [Link]
frameworks-capacity-building-by-gabriela-ramos-and-mariana-mazzucato-2022-12
Rawat, S., Rawat, A., Kumar, D., & Sabitha, A. S., 2021. Application of machine learning and
data visualization techniques for decision support in the insurance sector. International
Journal of Information Management Data Insights, 1(2), 100012.
Reed, Chris., 2007. Taking Sides on Technology Neutrality. SCRIPT-ed. 4.
10.2966/scrip.040307.263.
Riikkinen, M., Saarijärvi, H., Sarlin, P. and Lähteenmäki, I., 2018, "Using artificial intelligence
to create value in insurance", International Journal of Bank Marketing, Vol. 36 No. 6, pp.
43
 1145-1168. [Link]
Romao, M., Costa, J., Costa, C.J., 2019. Robotic Process Automation: A Case Study in the
Banking Industry, in: 2019 14th Iberian Conference on Information Systems and
Technologies (CISTI). Presented at the 2019 14th Iberian Conference on Information
Systems and Technologies (CISTI), pp. 1 6.
[Link]
Ruan, Q., Wang, Z., Zhou, Y., & Lv, D., 2020. A new investor sentiment indicator (ISI) based
on artificial intelligence: A powerful return predictor in China. Economic Modelling, 88,
27 58.
Ruffolo, M., 2022. The Role of Ethical AI in Fostering Harmonic Innovations that Support a
Human-Centric Digital Transformation of Economy and Society, in: Cicione, F., Filice,
L., Marino, D. (Eds.), Harmonic Innovation: Super Smart Society 5.0 and Technological
Humanism, Lecture Notes in Networks and Systems. Springer International Publishing,
Cham, pp. 139 143. [Link]
Ryll, L., Barton, M.E., Zhang, B.Z., McWaters, R.J., Schizas, E., Hao, R., Bear, K., Preziuso,
M., Seger, E., Wardrop, R., Rau, P.R., Debata, P., Rowan, P., Adams, N., Gray, M.,
Yerolemou, N., 2020. Transforming Paradigms: A Global AI in Financial Services Survey.
[Link]
Ryll, L., Barton, M.E., Zhang, B.Z., McWaters, R.J., Schizas, E., Hao, R., Bear, K., Preziuso,
M., Seger, E., Wardrop, R., Rau, P.R., Debata, P., Rowan, P., Adams, N., Gray, M.,
Yerolemou, N., 2020. Transforming Paradigms: A Global AI in Financial Services Survey.
[Link]
Saeys, Y., Van Gassen, S., Lambrecht, B.N., 2016. Computational flow cytometry: helping to
make sense of high-dimensional immunology data. Nat. Rev. Immunol. 16, 449 462.
[Link]
Sang, E.F.T.K., De Meulder, F., 2003. Introduction to the CoNLL-2003 Shared Task:
Language-Independent Named Entity Recognition.
[Link]
Sazali, S.S., Rahman, N.A., Bakar, Z.A., 2016. Information extraction: Evaluating named
entity recognition from classical Malay documents, in: 2016 Third International
Conference on Information Retrieval and Knowledge Management (CAMP). Presented at
the 2016 Third International Conference on Information Retrieval and Knowledge
Management (CAMP), pp. 48 53. [Link]
Shamima, A, Muneer M. Alshater, Anis El Ammari, Helmi Hammami, 2022. Artificial
intelligence and machine learning in finance: A bibliometric review, Research in
International Business and Finance, Volume 61.
Shiue, W., Li, S.-T., Chen, K.-J., 2008. A frame knowledge system for managing financial
decision knowledge. Expert Syst. Appl. 35, 1068 1079.
[Link]

Opportunities. Int. J. Adv. Qual. 46, 34 39.

Sigrist F., Hirnschall, C., 2019. Grabit: Gradient tree-boosted tobit models for default
prediction, J. Bank. Financ., 102 (2019), pp. 177-192
Singh, C., Lin, W., 2020. Can artificial intelligence, RegTech and CharityTech provide

44
 effective solutions for anti-money laundering and counter-terror financing initiatives in
charitable fundraising. J. Money Laund. Control. [Link]
0100.
Sotnik, S., Deineko, Z., Lyashenko, V., 2022. Key Directions for Development of Modern
Expert Systems.
Sridhar, D., Getoor, L., 2019. Estimating Causal Effects of Tone in Online Debates.
[Link]
Stone, M., Aravopoulou, E., Ekinci, Y., Evans, G., Hobbs, M., Labib, A., ... & Machtynger, L.,
2020. Artificial intelligence (AI) in strategic marketing decision-making: a research
agenda. The Bottom Line, 33(2), 183-200.
Svetlova, E. AI ethics and systemic risks in finance. AI Ethics 2, 713 725, 2022.
[Link] Common
Regulatory Capacity for AI
[Link]
07/common_regulatory_capacity_for_ai_the_alan_turing_institute.pdf
Thekkethil, M.S., Shukla, V.K., Beena, F., Chopra, A., 2021. Robotic Process Automation in
Banking and Finance Sector for Loan Processing and Fraud Detection, in: 2021 9th
International Conference on Reliability, Infocom Technologies and Optimization (Trends
and Future Directions) (ICRITO). Presented at the 2021 9th International Conference on
Reliability, Infocom Technologies and Optimization (Trends and Future Directions)
(ICRITO), pp. 1 6. [Link]
Thennakoon, A., Bhagyani, C., Premadasa, S., Mihiranga, S., Kuruwitaarachchi, N., 2019.
Real-time Credit Card Fraud Detection Using Machine Learning, in: 2019 9th
International Conference on Cloud Computing, Data Science & Engineering (Confluence).
Presented at the 2019 9th International Conference on Cloud Computing, Data Science &
Engineering (Confluence), pp. 488 493.
[Link]
Thowfeek, M.H., Samsudeen, S.N., Sanjeetha, M.B.F., 2020. Drivers of Artificial Intelligence
in Banking Service Sectors. Solid State [Link], H., Zheng, H., Zhao, K., Liu,
MW., Zeng, DD (2022). Inductive Representation Learning on Dynamic Stock Co-
Movement Graphs for Stock Predictions. INFORMS Journal on Computing, Volume 34,
Issue 4, pp. 1841-2382
Tiwari, R., Srivastava, S., & Gera, R., 2020. Investigation of artificial intelligence techniques
in finance and marketing. Procedia Computer Science, 173, pp. 149 157.
Toronto Centre, 2022. Supervisory implications of artificial intelligence and machine
learning
[Link]
[Link]
Torshin, [Link]., Rudakov, K.V., 2015. On the theoretical basis of metric analysis of poorly
formalized problems of recognition and classification. Pattern Recognit. Image Anal. 25,
577 587. [Link]
Tsai, C.-F., Wu, J.-W., 2008. Using neural network ensembles for bankruptcy prediction and
credit scoring. Expert Systems with Applications 34, 2639 2649.
[Link]
Turek, M., n.d. DARPA - Explainable Artificial Intelligence (XAI) Program, 2017. [WWW

45
 Document]. URL [Link]
-leading position in technologies of Tomorrow,
UKRI. Available at: [Link]
position-in-technologies-of-tomorrow/
Ulbricht, L. and Yeung, K., 2022.
Regulation & Governance, 16(1), pp.
3-22. DOI: 10.1111/rego.12437
Umuhoza, E., Ntirushwamaboko, D., Awuah, J., Birir, B., 2020. Using Unsupervised Machine
Learning Techniques for Behavioral-based Credit Card Users Segmentation in Africa.
SAIEE Afr. Res. J. 111, 95 101. [Link]
Ursachi, O., 2019. Role and applications of NLP in Cybersecurity, Medium. Medium.
Available at: [Link]
cybersecurity-333d9280c737
Van Hove, Leo, and Antoine Dubus. 2019. M-PESA and financial inclusion in Kenya: Of
paying comes saving? Sustainability 11: 568.
Vandrangi, S.K., 2022. PREDICTING THE INSURANCE CLAIM BY EACH USER USING
MACHINE LEARNING ALGORITHMS. Journal of Emerging Strategies in New
Economics, 1(1), pp.1-11.
Varma, V. and Mukherjee, S., 2022. Insider trading: the role of ai as a prevention tool.
[Link]
Vella, V., & Ng, W. L., 2015. A dynamic fuzzy money management approach for controlling
the intraday risk-adjusted performance of AI trading algorithms. Intelligent Systems in
Accounting, Finance and Management, 22(2), 153 178.
Veloso, M., Balch, T., Borrajo, D., Reddy, P., Shah, S., 2021. Artificial intelligence research
in finance: discussion and examples, Oxford Review of Economic Policy, Volume 37,
Issue 3, Pages 564 584, [Link]
Villar, A.S., Khan, N. Robotic process automation in banking industry: a case study on
Deutsche Bank. J BANK FINANC TECHNOL 5, 71 86, 2021.
[Link]
Wang, H.; Li, C.; Gu, B,; and Min, W., 2019. "Does AI-based Credit Scoring Improve Financial
Inclusion? Evidence from Online Payday Lending". ICIS 2019 Proceedings. 20.
[Link]
Webster, J.J., Kit, C., 1992. Tokenization as the Initial Phase in NLP, in: COLING 1992
Volume 4: The 14th International Conference on Computational Linguistics. Presented at
the COLING 1992.
Wei, Y., Yildirim, P., Van den Bulte, C., & Dellarocas, C., 2016. Credit scoring with social
network data. Marketing Science, 35(2), 234-258.
White, 1988. Economic prediction using neural networks: the case of IBM daily stock returns,
in: IEEE 1988 International Conference on Neural Networks. Presented at the IEEE 1988
International Conference on Neural Networks, pp. 451 458 vol.2.
[Link]
WhiteHouse. 2022. The impact of artificial intelligence on the future of workforces in the
european union and the United States of America. [Link]
content/uploads/2022/12/[Link]
46
Willcocks, L., Lacity, M., Craig, A., 2015. Robotic Process Automation at Xchanging.
Wittmann, X., and Lutfiju, F., 2021. Adopting AI in the Banking Sector The Wealth
Management Perspective. In Society 5.0. Springer International Publishing.
WorldBank. 2020. Digital Financial Inclusion. Available online:
[Link]
inclusion
Wu, D., Liang, L., Yang, Z., 2008. Analyzing the financial distress of Chinese public
companies using probabilistic neural networks and multivariate discriminate analysis.
Socio-Economic Planning Sciences 42, 206 220.
[Link] J. (2020). Application of machine
learning models and artificial intelligence to analyze annual financial statements to identify
companies of unfair corporate culture. Procedia Computer Science, 176, 3037 3046.
Xu, F., Uszkoreit, H., Du, Y., Fan, W., Zhao, D., Zhu, J., 2019. Explainable AI: A Brief Survey
on History, Research Areas, Approaches and Challenges. Nat. Lang. Process. Chin.
Comput., Lecture Notes in Computer Science 563 574. [Link]
030-32236-6_51
Xu, J., 2022. Future And Fintech, The: Abcdi And Beyond. FUTURE FINTECH ABCDI
Beyond 1 36.
Yang, Liu, and Youtang Zhang. 2020. Digital Financial Inclusion and Sustainable Growth of
Small and Micro Enterprises
Listed Companies. Sustainability 12: 3733.
Yang, W., Fang, Z., Hui, L., 2016. Study of an Improved Text Filter Algorithm Based on Trie
Tree. Presented at the 2016 International Symposium on Computer, Consumer and Control
(IS3C), pp. 594 597. [Link]
Yu, M., Yang, Z., Kolar, M., Wang, Z., 2019. Convergent Policy Optimization for Safe
Reinforcement Learning, in: Advances in Neural Information Processing Systems. Curran
Associates, Inc.
Yunusoglu, M.G., Selim, H., 2013. A fuzzy rule based expert system for stock evaluation and
portfolio construction: An application to Istanbul Stock Exchange. Expert Syst. Appl. 40,
908 920. [Link]
Zeinalizadeh N, Shojaie AA, Shariatmadari M (2015). Modeling and analysis of bank customer
satisfaction using neural networks approach. International Journal of Bank Marketing 33:
717-732.
Zetzsche, DA, Warner, D., Buckley, R., Tang, Brian, 2020. Regulating AI in Finance: Putting
the Human in the Loop. Oxford Business Law Blog. Can be retrieved at:
[Link]
human-loop
Zhang, B.Z., Ashta, A., and Barton, M.E., 2021. Do FinTech and financial incumbents have
different experiences and perspectives on the adoption of artificial intelligence? Strategic
Change, Volume 30, Issue 3, Pages 223-234.
Zhang, Z., Zohren, S., Roberts, S., 2020. Deep Reinforcement Learning for Trading. J. Financ.
Data Sci. 2, 25 40. [Link]
Zheng, X., Zhu, M., Li, Q., Chen, C., Tan, Y., 2019. FinBrain: when finance meets AI 2.0.
Frontiers Inf Technol Electronic Eng 20, 914 924.

47
 [Link]
Zhong, X., Enke, D., 2019. Predicting the Daily Return Direction of the Stock Market using
Hybrid Machine Learning Algorithms. Financial Innovation 5, 1 20.
[Link]

48