This strikes me as a sensible way of thinking about machine learning: it’s like when we got relational databases—suddenly we could do more, quicker, and easier …but it doesn’t require us to treat the technology like it’s magic.
An important parallel here is that though relational databases had economy of scale effects, there were limited network or ‘winner takes all’ effects. The database being used by company A doesn’t get better if company B buys the same database software from the same vendor: Safeway’s database doesn’t get better if Caterpillar buys the same one. Much the same actually applies to machine learning: machine learning is all about data, but data is highly specific to particular applications. More handwriting data will make a handwriting recognizer better, and more gas turbine data will make a system that predicts failures in gas turbines better, but the one doesn’t help with the other. Data isn’t fungible.
I don’t think it’s controversial to suggest that LLMs haven’t measured up to any of the lofty promises made by their vendors. But in more concrete terms, consumers dislike “AI” when it shows up in products, and it makes them actively mistrust the brands that employ it. In other words, we’re some three years into the hype cycle, and LLMs haven’t met any markers of success we’d apply to, well, literally any other technology.
Today’s AI promoters are trying to have it both ways: They insist that AI is crossing a profound boundary into untrodden territory with unfathomable risks. But they also define AI so broadly as to include almost any large-scale, statistically-driven computer program.
Under this definition, everything from the Google search engine to the iPhone’s face-recognition unlocking tool to the Facebook newsfeed algorithm is already “AI-driven” — and has been for years.
LLMs have never experienced anything. They are just programs that have ingested unimaginable amounts of text. LLMs might do a great job at describing the sensation of being drunk, but this is only because they have read a lot of descriptions of being drunk. They have not, and cannot, experience it themselves. They have no purpose other than to produce the best response to the prompt you give them.
This doesn’t mean they aren’t impressive (they are) or that they can’t be useful (they are). And I truly believe we are at a watershed moment in technology. But let’s not confuse these genuine achievements with “true AI.”
I feel like there’s a connection here between what Kevin Kelly is describing and what I wrote about guessing (though I think he might be conflating consciousness with intelligence).
This, by the way, is also true of immersive “virtual reality” environments. Instead of trying to accurately recreate real-world places like meeting rooms, we should be leaning into the hallucinatory power of a technology that can generate dream-like situations where the pleasure comes from relinquishing control.
I am not a believer in the AI singularity — the rapture of the nerds — that is, in the possibility of building a brain-in-a-box that will self-improve its own capabilities until it outstrips our ability to keep up. What CS professor and fellow SF author Vernor Vinge described as “the last invention humans will ever need to make”. But I do think we’re going to keep building more and more complicated, systems that are opaque rather than transparent, and that launder our unspoken prejudices and encode them in our social environment. As our widely-deployed neural processors get more powerful, the decisions they take will become harder and harder to question or oppose. And that’s the real threat of AI — not killer robots, but “computer says no” without recourse to appeal.
We’ve taught machines to hallucinate so let’s be honest about their hallucinations.