"How, exactly, could AI take over by 2027?"
Introducing AI 2027: a deeply-researched scenario forecast I wrote alongside @slatestarcodex, @eli_lifland, and @thlarsen
Apr 3, 2025 · 4:04 PM UTC
The scenario: ai-2027.com
PDF version: ai-2027.com/scenario.pdf
Looking forward to engaging with comments!
The @slatestarcodex & @DKokotajlo episode.
Scott and Daniel break down every month from now until the 2027 intelligence explosion.
Misaligned hive minds, Xi and Trump waking up, automated Ilyas accelerating AI progress.
I went in quite skeptical. But I learned a tremendous amount by bouncing different objections against them.
Links below. Enjoy!
Timestamps:
00:00:00 - AI 2027
00:07:45 - Forecasting 2025 and 2026
00:15:30 - Why LLMs aren't making discoveries
00:25:22 - Debating intelligence explosion
00:50:34 - Can superintelligence actually transform science?
01:17:43 - Cultural evolution vs superintelligence
01:24:54 - Mid-2027 branch point
01:33:19 - Race with China
01:45:36 - Nationalization vs private anarchy
02:04:11 - Misalignment
02:15:41 - UBI, AI advisors, & human future
02:23:49 - Factory farming for digital minds
02:27:41 - Daniel leaving OpenAI
02:36:04 - Scott's blogging advice
very well written and an interesting read. your forecast of how the stock market would react to some of these potential outcomes is pretty poor though.
1
22
15,943
would be curious to hear you debate with researchers from @EpochAIResearch who seem to expect AI-automated AI R&D to be far less consequential. What are your key disagreements?
xitter.twitterapi.xyz/EpochAIResearch/status…
Even within our own research team, timelines for transformative AI differ substantially. In this episode, the two Epoch AI researchers with the longest and the shortest timelines for transformative AI candidly examine the roots of their disagreements.
They discuss:
How and why their timelines for specific milestones differ
Current technical challenges for AI progress
Why widespread automation beats geniuses in datacenters
Pitfalls of conventional AI prediction approaches
Critiquing "single AGI" or "utopia vs. doom" narratives
How a world with AGI might look like
And much, much more.
(0:00:00) - Preview
(0:01:08) - Contrasting AGI Timelines
(0:08:30) - Updating Beliefs as Capabilities Advance
(0:17:07) - Moravec’s Paradox and the Agency Challenge
(0:32:40) - Missing Capabilities for AGI
(0:47:43) - Beating benchmarks vs Being Useful
(0:59:20) - AI Excelling in Some Tasks While Struggling with Others
(1:07:33) - Economic Impact of AI vs the Internet
(1:24:08) - Widespread Automation Beats Genius in Datacenters
(1:51:37) - How Stories Shape Our Expectations of AI
(2:03:24) - How AGI Will Impact Culture
(2:10:46) - Beyond Utopia-or-Extinction
(2:16:57) - AI's Impact on Wages and Labor
(2:27:49) - Why Better Preservation of Information Accelerates Change
(2:39:32) - Markets Shaping Cultural Priorities
(2:55:51) - Challenges in Defining What We Want to Preserve
(3:06:47) - Risk Attitudes in AI Decision-Making
(3:12:50) - Historical Lessons for AI Coexistence
(3:21:45) - A Warning Sign in Safety Discourse
(3:30:20) - Revisiting Core Assumptions in AI Alignment
(3:49:46) - Simple Models in Complex Domains
I think my major point of disagreement with your modeling is that thousands of AI agents are unlikely to be equivalent to thousands of human researchers, due to the difficulty of making their creative process sufficiently different from each other.
2
16
4,488
Does this scenario meaningfully change if leaders at OpenBrain fully accept your conclusions today and assume their top competitors do too?
2
7
19,174
Great thought-provoking read along w similar work by @LRudL_. Thought it odd that 2nd place labs and open source AI have very little role to play. I'd think a strong #2 lab should change the behavior of the #1. But I need to ponder more.
1
5
10,164
Yes, I disagree with @LRudL_ about takeoff speeds but found their scenario very well done. It's the sort of thing we want to see more of.lesswrong.com/posts/CCnycGce…
As for 2nd place companies + open source: in our wargames about half the time they make a big difference, though they are irrelevant half the time. It is indeed a bit of a shame that AI 2027 depicts them as being irrelevant because we definitely aren't confident in that.
Can you do this prediction chain again, but this time using xAI Grok4 as the "Open Brain" instead of OpenAI GPT? I believe the scenarios have potential to be played out wildly differently from what was considered a few months ago.
I request you gguys doing this prediction every few months because the pace of AI change is a few years of industrial revolution compressed into a few months.
1
3
1,424
OpenBrain was meant to represent whichever company is in the lead at the time of takeoff, not necessarily OpenAI. Also, I don't think xAI would behave significantly differently from OoenBrain were it in the lead. I think all of the major US companies would probably behave roughly like we depict OpenBrain behaving (i.e. massively irresponsibly) with some variations and chance of divergence of course.
9
1,024
I can't express how glad I am that a work like this exists. I am working on a blog post analyzing some calculations in AI 2027, and I was curious as to how you all came up with the "6% of compute will go to research automation." Would love to chat about it, as the Research Automation section doesn't go into too much detail. ai-2027.com/research/compute…
(I can't DM you it seems, so commenting here instead)
1
2
1,331
How can you speculate with any confidence on what the AI does after it becomes superintelligent? If you could accurately predict the behavior of a superintelligent being, you'd be one yourself.
2
2
1,856
The idea that AI is going to speed up algorithmic progress by 50% next year sounds implausible to me.
Also, won’t their competitors also have capable AIs so the edge over competitors (including Chinese ones) will be much less than 50%, even if you are right?
1
3
1,466
We are uncertain of course. But based on conversations with researchers at frontier AI companies, it seems that AIs are already speeding up algorithmic progress by 10% or so today. And they are barely agents at all; a year and a half from now they will be much better.
Yes, their competitors will also have capable AIs so the edge will be less than 50%.
1
6
1,381