1 of 10
12 quotations
2017 to 2026
Artificial intelligence
AI from machine learning and cyborgs to large language models, AGI and institutions, 2017 to 2026.
Summary
Ours.
In 2017 Cowen said AI had developed faster than he predicted, and in 2019 he expected augmented humans to outcompete robots. In 2023 he called AGI doom a fictional story. In 2024 he said AI and biomedical advances had ended the Great Stagnation. In 2025 he said o3 met the old definition of AGI and named slow institutions as a bottleneck.
2017
AI has developed faster than people expected and faster than his own forecast. AI, in the broad sense of that term, seems to have developed more quickly than people thought. And even in 2011, I wrote a book saying, “AI will develop more quickly than we think.” My book, Average Is Over. And since then, it’s developed more quickly than I thought.
Play from about 1:21:07 Patrick Collison has a Few Questions for Tyler (Live at Stripe) · Official transcript
2018
Machine learning will grade exams and papers for professors. So we as professors won’t need to grade exams anymore, and I don’t just mean multiple choice. You run machine learning on papers, you find what correlates with a good paper, you put the paper through the program.
Play from about 30:45 Daniel Kahneman on Cutting Through the Noise · Official transcript
2019
The next breakthroughs may involve the internet and AI but are not here yet. The next set of big breakthroughs may well involve the Internet, artificial intelligence, Internet of things. They are not quite here yet. You see many signs of them.
Play from about 43:40 Rob Wiblin interviews Tyler on *Stubborn Attachments* · Official transcript
Augmented humans will outcompete self-aware robots of the Asimov kind. The notion that you have human-like beings augmented, cyborgs, I think, will outcompete the notion of robots in the Isaac Asimov sense that they walk like a robot, but they are self-aware and have all the smarts of AI.
Play from about 2:24:46 Rob Wiblin interviews Tyler on *Stubborn Attachments* · Official transcript
2023
The claim that AGI will destroy humanity is a kind of fictional story. Today, I consider it a kind of fictional story, Eliezer: well, the AGI is going to destroy us all is obviously beyond dystopia.
Play from about 19:42 Jerusalem Demsas on The Dispossessed, Gulliver's Travels, and Of Boys and Men · Official transcript
Many people will use LLMs instead of reading books, so he should diversify. You don’t have to believe the next GPT will write a better book than person X, but I do think many more people will be playing around with LLMs rather than reading books, so we need to diversify more out of only doing books.
Play from about 35:35 Conversations with Tyler 2023 Retrospective · Official transcript
2024
The Great Stagnation ended a few years ago with AI and biomedical advances. I think it’s over. Not everyone thinks it’s over. I think with AI and advances in biomedical science, it ended a few years ago.
Play from about 21:16 Conversations with Tyler 2024 Retrospective · Official transcript
A Hayekian point: some problems may have no answer in an illegible universe. I don’t think that’s how intelligence works. And this is a Hayekian point. And some of these problems, there just may be no answer. Like, maybe the universe isn’t that legible, and if it’s not that legible, GPT Eleven doesn’t really make sense as a creature or whatever.
Play from 37:27 Tyler Cowen — Hayek, Keynes, & Smith on AI, animal spirits, anarchy, & growth
2025
University committees planning for AI in the curriculum show where the bottlenecks are. Like just try going to a mid-tier state university and sit down with the committee designed to develop a plan for using artificial intelligence in the curriculum. And then come back to me and tell me how that went, and then we'll talk about bottlenecks.
Play from about 6:20 Tyler Cowen — The #1 bottleneck to AI progress is humans
The o3 model largely meets the old definition of AGI as matching human experts. Maybe the biggest surprise in AI was how good the O3 model was, which I called AGI at the time. Maybe tongue-in-cheek, but I also meant it, that it can’t do every job. The old definition of AGI, that it’s just as good as human experts, is largely true.
Play from about 11:35 Conversations with Tyler 2025 Retrospective · Official transcript
2026
People will deal with a friendly AI, which then deals with government staff. You intermediate through the AI, which is nicer, friendlier, easier to deal with, and then it deals with the nasty person in the government.
Play from about 25:40 Annie Lowrey on the Time Tax, Fraud, and Rationing by Paperwork · Official transcript
People may respect experts more when AI delivers an impersonal distillation of expertise. One view is no one respects the experts because the machines are better, but I think an alternative possibility is the machine, by not being tied to a personality, is less disliked. People actually respect the experts more because they get this impersonal distillation of the experts.
Play from about 26:40 Brendan Foody on Teaching AI and the Future of Knowledge Work · Official transcript