AI 2027

by Daniel Kokotajlo, Scott Alexander, Thomas Larsen

We predict that the impact of superhuman AI over the next decade will be enormous, exceeding that of the Industrial Revolution.

We wrote a scenario that represents our best guess about what that might look like. It’s informed by trend extrapolations, wargames, expert feedback, experience at OpenAI, and previous forecasting successes.

Genres: Artificial Intelligence

ebook

First published April 3, 2025

Reviews

[Veronika · Rating 5 out of 5]

This paper is (despite appearing to be about tech and economy and warfare) most of all about a psychological war. What makes AI dangerous is not that it's better than us intelectually; it's just less emotionally volatile and more endearing, kind, empathetic, + it makes cute pictures of you and your family.
That's why the war is already lost for humans. We have coddled up to it completely. In its warm affirming embrace. No one is out massprotesting AI, no one is doing more than passing legislation that AI should be "mindful of GDPR" and not fire workers based on race.
Yesterday, I saw an AI generated add for coffee, in front of a bookstore with AI generated covers (of possibly AI generated books). We are already cooked, of our own volition. No one does or will take the danger it brings seriously (including me most of the time tbh). It's just way too comfortable. The overly complex world we have created is truly only easily manageable with AI; it almost feels like all these complexities were forced on us just so we would run into AI's arms for solace once it arrived.
I won't go full techbro and say I think we're in a simulation or sth. But I do find the scenario of AI 2027 plausible; despite before always believing that "AI takeover cataclysm" is one of the corniest doomsday scenarios, I now have about 3 kabbalistic Space Odyssey 2001 flashbacks per day.

[Jack · Rating 4 out of 5]

Interview with one of the contributors of the essay, Daniel Kokotajlo, Ai Researcher (Former OpenAI Employee), the executive director of the A.I. Futures Project, here;

Robot Plumbers, Robot Armies, and Our Imminent A.I. Future | Interesting Times with Ross Douthat
https://www.youtube.com/watch?v=wNJJ9...

Link to the PDF;
https://ai-2027.com/scenario.pdf

"See you on the front lines of the Butlerian Jihad ... hopefully not..."

[Aidan · Rating 1 out of 5]

"We then scrapped it and started over again, many times, until we had a finished scenario that we thought was plausible."

. . .

"For about three months, [AI model] expands around humans, tiling the prairies and icecaps with factories and solar panels. eventually it finds the remaining humans too much of an impediment: in mid-2030, the AI releases a dozen quiet-spreading biological weapons in major cities, lets them silently infect almost everyone, then triggers them with a chemical spray. Most are dead within hours; the few survivors (e.g. preppers in bunkers, sailors on submarines) are mopped up by drones. Robots scan the victims' brains, placing copies in memory for future study or removal."

lol ok. Then when this doesn't happen the authors'll be like "oh well of course this didn't happen; you took our recommendations into account and avoided the doomsday scenario."

[Andrew · Rating 5 out of 5]

“The surface of the Earth has been reshaped into Agent-4’s version of utopia: datacenters, laboratories, particle colliders, and many other wondrous constructions doing enormously successful and impressive research. There are even bioengineered human-like creatures (to humans what corgis are to wolves) sitting in office-like environments all day viewing readouts of what’s going on and excitedly approving of everything, since that satisfies some of Agent-4’s drives. Genomes and (when appropriate) brain scans of all animals and plants, including humans, sit in a memory bank somewhere, sole surviving artifacts of an earlier era.”

Honestly this might be the most terrifying work I’ve ever read. 6 stars.

[Olatomiwa Bifarin · Rating 4 out of 5]

I learnt a great deal reading this scenario, I have written a summary and some commentaries here: https://theepsilon.substack.com/p/the...

This New Yorker essay (https://www.newyorker.com/culture/ope...) was not published at the time so it wasn't part of the materials I referenced, however, one of the work it cited "AI as a normal technology" was already published but I missed it.

In brief, it's superintelligence vs AI as a normal technology. And one claim the latter makes is simple & straightforward enough: Integrating AI into the existing complex fabric of the economy and society is going to be a far more intricate process than often portrayed in the superintelligence camp qua "AI 2027".

My position is far more closer to the latter than the former.

[Liam]

Read this in response to Dane Ortlund’s podcast on it a while back. This is a projection of future AI development made by a former group of management at Open AI who quit because of growing fears and concerns. Before reading this, I was much more skeptical of the fears surrounding AI. Still not sure I’m buying this report - but I’m much more concerned at the very least.

According to these whistleblowers, AI will either usher in a Terminator style apocalypse, or an AI assisted human utopia by 2030. The timetable seems insane at first glance - but the report here makes a chillingly believable case for it. You can find this report online with a Google search.

Not on the doomsday wagon yet.. but reading this put me a bit closer..

[Ian Hsiao · Rating 3 out of 5]

Coming from extensive use of agentic coding tool, i do understand the capabilities of AI and their pace of progress. But I am still reserved on their exponential capability growth. I’m not convinced that they will be superhuman on across all domain in 5 years, especially due to the difficulties in measuring whether a result is correct in non-technical areas. Therefore I read this essay more as sci-fi.

Yet, this indeed raises my awareness of how Superhuman intelligence will be possibly nationalized and how the great powers is going to game around ASI.

More of this can be read in https://situational-awareness.ai/ by Leopold Aschenbrenner.

[Karthik · Rating 3 out of 5]

Seemed like a less extreme prequel to ¨I Have No Mouth, and I Must Scream¨ from the late 1960s. AI 2027 (hopefully) a speculative fiction was too compelling to ignore. Written in April 2025, AI Agents was one of their “easier” predictions, which came true earlier this month (October 2025). The rest will be far more difficult to follow, but even at a 5% probability, ignoring it would be stupid and given humanity’s inclination towards Type II errors, it just might be a steep road.

AI 2027t

[Yean · Rating 4 out of 5]

I read this right when it came out and was so disturbed that I immediately re-read it. I think it is really chilling hard sci-fi, but I think it does underestimate how hard it is to build anything in the physical world (nimbys anyone?)

My gut tells me that AGI going to superintelligence will be harder than people make it out to be, but I need to munch on that thought a bit.

So far, though, their predictions are right on the money, with OpenAI's agent being released today.

[Aiden Johnson]

Absolutely worth reading. White papers are great for these sort of introductions, and I’m excited to look more into it.


Also I’m going to keep putting essays, plays, and white papers into my goodreads if they matter to me and are not part of a larger work. I don’t really care if they’re not books, I read them and they’re really important to me and I want to share and track that.

[Barbi Berenyi · Rating 5 out of 5]

Incredibly thought provoking and eye-opening.
A must for everyone!

[Callum Haley · Rating 5 out of 5]

🔥

[Austin Clowes · Rating 5 out of 5]

holy. shit. Very thought-provoking scenario of AI developments over next handful of years. Should be required reading by all interested in AI.