The other day, we were looking at the fact that a number of trend watchers are predicting that the current generative AI bubble (chatbots, AI art, etc.) is about to burst: For example, “Shares in AI stocks in the US and Asia slid dramatically overnight, showing that Wall Street’s investors have started to run out of patience with the hype, writes James Moore.”

AI expert Gary Marcus offered some thoughts the other day on a factor that that most analysts overlook but in his view explains everything:

The simple fact is that current approaches to machine learning (which underlies most of the AI people talk about today) are lousy at outliers, which is to say that when they encounter unusual circumstances, like the subtly altered word problems that I mentioned a few days ago, they often say and do things that are absurd. (I call these discomprehensions.)

Gary Marcus, “This one important fact about current AI explains almost everything,” Marcus on AI, August 1, 2024

Right. Like the recent claims about bears in space and the jackrabbits everywhere …

Big Tech got rich, he charges, without confronting the problem:

The people who have temporarily gotten rich or famous on AI have done so by pretending that this outlier problem simply doesn’t exist, or that a remedy for it is imminent. When the bubble deflation that I have been predicting comes, as now seems imminent, it will come because so many people have begun to recognize that GenAI can’t live up to expectations.

The reason it can’t meet expectations? Say it in unison, altogether now: GenAI sucks at outliers. If things are far enough from the space of trained examples, the techniques of generative AI will fail.

Marcus, “Explains almost everything”

In other words, as Robert J. Marks often points out, AI is not creative. It finds solutions if they already exist online but woe betide us if they don’t. That’s when the hallucinations and crazy reasoning begin. If the chatbot answers are crazy enough, users will notice. But what if the answers are wrong but not obviously crazy?

Marcus warns that once you grasp this issue, “almost everything that people like Altman and Musk and Kurzweil are currently saying about AGI being nigh seems like sheer fantasy, on par with imagining that really tall ladders will soon make it to the moon.”

He notes that he and Steven Pinker had warned of this problem a while back. Maybe the big guys figured that it didn’t pay to listen then. If so, maybe their shareholders will pay later.

Where are HAL and David when we need them to prove a point?

Marcus enlarged on this theme in another post at his Substack today. He offered four reasons for thinking the bubble will soon burst. Here are the first two:

1. In 2012, in the New Yorker, I pointed out a series of problems with deep learning, including troubles with reasoning and abstraction that were often ignored (or denied) for years, but that continue to plague deep learning to this day – and that now, at last, have come to be very widely recognized.

In December 2022, at the height of ChatGPT’s popularity I made a series of seven predictions about GPT-4 and its limits, such as hallucinations and making stupid errors, in an essay called What to Expect When You Are Expecting GPT-4]. Essentially all have proven correct, and held true for every other LLM that has come come since.

Gary Marcus, “Why the collapse of the Generative AI bubble may be imminent,” Marcus on AI, August 3, 2024

The other two reasons and the source links are here.

We should remember all this when we hear Ray Kurzweil tell us that AI will think like humans in 2029 or when Sam Altman forecasts a super-competent AI colleague.

Artificial intelligences HAL 9000 and David are murderous and otherwise unlikeable but their crazy is sociopathic, not demented:

HAL:

David:

One wonders how much use HAL or David would have for Altman’s ChatGPT4o. At least we can be reassured that they are just science fiction. Unless, of course, Ray Kurzweil is right … 😉

You may also wish to read: Is Big Tech’s AI starting to run out of other people’s money? People are beginning to wonder if all the AI hype is really going to pay off in substantial improvements. Two things we can be sure of: Things that can’t go on forever won’t. And even artificial intelligence can’t make hype into reality.