Nobel Laureate Daron Acemoglu: Don't Believe the AI Hype

[1] Logic, Optimization, and Constraint Programming: A Fruitful Collaboration - John Hooker - CMU (2023) [video]:

https://www.youtube.com/live/TknN8fCQvRk

[2] "We Really Don't Know How to Compute!" - Gerald Sussman - MIT (2011) [video]:

https://youtube.com/watch?v=HB5TrK7A4pI

[3] Google OR-Tools:

https://developers.google.com/optimization

[4] MiniZinc:

https://www.minizinc.org/

Macroeconomic and productivity forecasts from 10-15 years ago are pretty accurate, and if anything, were too optimistic on the productivity front, but there was certainly nothing wrong with taking them seriously.

I was borne in the seventies. Much of what is science fact today was science fiction then. And much of that was pretty naive and enlightening at the same time.

My point is that nothing has changed when it comes to people's ability to predict the future. The louder people claim to know what it all means or rush to man-splain that to others, the more likely it is that they are completely and utterly missing the point. And probably in ways that will make them look pretty foolish in a few decades. Most people are just flailing around in the dark. And some of the crazier ones might actually be the ones to listen to. But you'd be well advised to filter out their interpretations and attempts to give meaning to it all.

Hal, the Paranoid Android, Kitt, C3PO, R2D2, Skynet, Data, and all the other science fiction AIs from my youth are now pretty much science fact. Some of those actually look a bit slow and retarded in comparison. Are we going to build better versions of these? I'd be very disappointed in the human race if we didn't. And I'd be also disappointed if that ends up resembling the original fantasies of those things. I don't think many people are capable of imagining anything more coherent than versions of themselves dressed up in some glossy exterior. Which is of course what C3PO is. Very relatable, a bit stupid, and clownish. But also, why would you want such a thing? And the angry Austrian body builder version of that of course isn't any better.

I think the raw facts are that we've invented some interesting software that passes the Turing test pretty much with flying colors. For much of my life that was the gold standard of testing AIs. I don't think anyone has bothered to actually deal with the formalities of letting AIs take that test and documenting the results in a scientific way. That test obviously became obsolete before people even thought of doing that. We now worry about abuse of AIs to deceive entire populations with AIs pretending to be humans manipulating people. You might actually have a hard time convincing people that have been abused in such a way that what they saw and heard was actually real. We imagined it would be hard to convince them it AIs are human. We failed to imagine the job of convincing them they are not is much harder.

Economists, businesspeople & their ilk have proven time & time again that 99% of them just throw darts at a board & see what sticks. The only ingredients required are money, connections and an extroversion (height helps too). That's not to say that most scientists don't do the same thing, that is science after all.

I doubt many people at all would have expected even the success of LLMs before Google's attention paper. NLP experienced a huge jump, previous models always seemed to me like handwritten sets of statistical rules stringing together text and now we have trained sets of statistical rules orders of magnitudes more complex...I have no idea what we'll end up with next.

Provocative question for sure, but how much things changed since 2020 ? Or even 2015 ?

I'm talking about changes in real economy. Except of huge system shock that was Covid, not that much.

>Life being what it is, several people came back at me, citing a prediction I made in 1998 that the internet’s growth would soon slow and that “by 2005 or so, it will become clear that the internet’s impact on the economy has been no greater than the fax machine’s.” I did indeed say that, in a throwaway piece I wrote for the magazine The Red Herring — a piece I still don’t remember having written, but I guess I was trying to be provocative.

..

>But how wrong was I, really, about the internet’s economic impact? Or, since this shouldn’t be about me, have the past few decades generally vindicated visionaries who asserted that information technology would change everything? Or have they vindicated techno-skeptics like the economist Robert Gordon, who argued in a 2016 book that the innovations of the late 20th and early 21st century were far less fundamental than those between 1870 and 1940? Well, by the numbers, the skeptics have won the argument, hands down.

>In that last newsletter, we looked at 10-year rates of growth in labor productivity, which suggested that information technology did indeed produce a bump in economic growth between the mid-1990s and the mid-2000s, but one that was relatively modest and short-lived. Today, let me take a slightly different approach. The Bureau of Labor Statistics produces historical estimates, going back to 1948, of both labor productivity and “total factor productivity,” an estimate of the productivity of all inputs, including capital as well as labor, which is widely used by economists as a measure of technological progress. A truly fundamental technological innovation should cause sustained growth in both these measures, especially total factor productivity.

(read the article to see pictures)

>See the great productivity boom that followed the rise of the internet? Neither do I.