Review

This post is a slightly-adapted summary of two twitter threads, here and here.

The t-AGI framework

As we get closer to AGI, it becomes less appropriate to treat it as a binary threshold. Instead, I prefer to treat it as a continuous spectrum defined by comparison to time-limited humans. I call a system a t-AGI if, on most cognitive tasks, it beats most human experts who are given time t to perform the task.

What does that mean in practice?

  • A 1-second AGI would need to beat humans at tasks like quickly answering trivia questions, basic intuitions about physics (e.g. "what happens if I push a string?"), recognizing objects in images, recognizing whether sentences are grammatical, etc.
  • A 1-minute AGI would need to beat humans at tasks like answering questions about short text passages or videos, common-sense reasoning (e.g. Yann LeCun's gears problems), simple computer tasks (e.g. use photoshop to blur an image), justifying an opinion, looking up facts, etc.
  • A 1-hour AGI would need to beat humans at tasks like doing problem sets/exams, writing short articles or blog posts, most tasks in white-collar jobs (e.g. diagnosing patients, giving legal opinions), doing therapy, doing online errands, learning rules of new games, etc.
  • A 1-day AGI would need to beat humans at tasks like writing insightful essays, negotiating business deals, becoming proficient at playing new games or using new software, developing new apps, running scientific experiments, reviewing scientific papers, summarizing books, etc.
  • A 1-month AGI would need to beat humans at coherently carrying out medium-term plans (e.g. founding a startup), supervising large projects, becoming proficient in new fields, writing large software applications (e.g. a new OS), making novel scientific discoveries, etc.
  • A 1-year AGI would need to beat humans at... basically everything. Some projects take humans much longer (e.g. proving Fermat's last theorem) but they can almost always be decomposed into subtasks that don't require full global context (even tho that's often helpful for humans).

Some clarifications:

  • I'm abstracting away from the question of how much test-time compute AIs get (i.e. how many copies are run, for how long). A principled way to think about this is probably something like: "what fraction of the world's compute is needed?". But in most cases I expect that the bottleneck is being able to perform a task *at all*; if they can then they'll almost always be able to do it with a negligible proportion of the world's compute.
  • Similarly, I doubt the specific "expert" theshold will make much difference. But it does seem important that we use experts not laypeople, because the amount of experience that laypeople have with most tasks is so small. It's not really well-defined to talk about beating "most humans" at coding or chess; and it's not particularly relevant either.
  • I expect that, for any t, the first 100t-AGIs will be *way* better than any human on tasks which only take time t. To reason about superhuman performance we can extend this framework to talk about (t,n)-AGIs which beat any group of n humans working together on tasks for time t. When I think about superintelligence I'm typically thinking about (1 year, 8 billion)-AGIs.
  • The value of this framework is ultimately an empirical matter. But it seems useful so far: I think existing systems are 1-second AGIs, are close to 1-minute AGIs, and are a couple of years off from 1-hour AGIs. (FWIW I formulated this framework 2 years ago, but never shared it widely. From your perspective there's selection bias—I wouldn't have shared it if I'd changed my mind. But at least from my perspective, it gets points for being useful for describing events since then.)

And very briefly, some of the intuitions behind this framework:

  • I think coherence over time is a very difficult problem, and one humans still struggle at, even though (I assume) evolution optimized us hard for this.
  • It's also been a major bottleneck for LLMs, for the principled reason that the longer the episode, the further off the training distribution they go.
  • Training NNs to perform tasks over long time periods takes much more compute (as modelled in Ajeya Cotra's timelines report).
  • Training NNs to perform tasks over long time periods takes more real-world time, so you can't gather as much data.
  • There are some reasons to expect current architectures to be bad at this (though I'm not putting much weight on this; I expect fixes to arise as the frontier advances).

Predictions motivated by this framework

Here are some predictions—mostly just based on my intuitions, but informed by the framework above. I predict with >50% credence that by the end of 2025 neural nets will:

  • Have human-level situational awareness (understand that they're NNs, how their actions interface with the world, etc; see definition here)
  • Beat any human at writing down effective multi-step real-world plans. This one proved controversial; some clarifications:
    • I think writing down plans doesn't get you very far, the best plans are often things like "try X, see what happens, iterate".
    • It's about beating any human (across many domains) not beating the best human in each domain.
    • By "many domains" I don't mean literally all of them, but a pretty wide range. E.g. averaged across all businesses that McKinsey has been hired to consult for, AI will make better business plans than any individual human could.
  • Do better than most peer reviewers
  • Autonomously design, code and distribute whole apps (but not the most complex ones)
  • Beat any human on any computer task a typical white-collar worker can do in 10 minutes
  • Write award-winning short stories and publishable 50k-word books
  • Generate coherent 5-min films (note: I originally said 20 minutes, and changed my mind, but have been going back and forth a bit after seeing some recent AI videos)
  • Pass the current version of the ARC autonomous replication evals (see section 2.9 of the GPT-4 system card; page 55). But they won't be able to self-exfiltrate from secure servers, or avoid detection if cloud providers try.
  • 5% of adult Americans will report having had multiple romantic/sexual interactions with a chat AI, and 1% having had a strong emotional attachment to one.
  • We'll see clear examples of emergent cooperation: AIs given a complex task (e.g. write a 1000-line function) in a shared environment cooperate without any multi-agent training.

The best humans will still be better (though much slower) at:

  • Writing novels
  • Robustly pursuing a plan over multiple days
  • Generating scientific breakthroughs, including novel theorems (though NNs will have proved at least 1)
  • Typical manual labor tasks (vs NNs controlling robots)

FWIW my actual predictions are mostly more like 2 years, but others will apply different evaluation standards, so 2.75 (as of when the thread was posted) seems more robust. Also, they're not based on any OpenAI-specific information.

Lots to disagree with here ofc. I'd be particularly interested in:

  • People giving median dates they expect these to be achieved 
  • People generating other specific predictions about what NNs will and won't be able to do in a few years' time
New Comment
15 comments, sorted by Click to highlight new comments since:

(expanding on my reply to you on twitter)

For the t-AGI framework, maybe you should also specify that the human starts the task only knowing things that are written multiple times on the internet. For example, Ed Witten could give snap (1-second) responses to lots of string theory questions that are WAY beyond current AI, using idiosyncratic intuitions he built up over many years. Likewise a chess grandmaster thinking about a board state for 1 second could crush GPT-4 or any other AI that wasn’t specifically and extensively trained on chess by humans.

A starting point I currently like better than “t-AGI” is inspired the following passage in Cal Newport’s book Deep Work:

[Deciding whether an activity is “deep work” versus “shallow work”] can be more ambiguous. Consider the following tasks:

  • Example 1: Editing a draft of an academic article that you and a collaborator will soon submit to a journal.
  • Example 2: Building a PowerPoint presentation about this quarter’s sales figures.
  • Example 3: Attending a meeting to discuss the current status of an important project and to agree on the next steps. It’s not obvious at first how to categorize these examples.

The first two describe tasks that can be quite demanding, and the final example seems important to advance a key work objective. The purpose of this strategy is to give you an accurate metric for resolving such ambiguity—providing you with a way to make clear and consistent decisions about where given work tasks fall on the shallow-to-deep scale. To do so, it asks that you evaluate activities by asking a simple (but surprisingly illuminating) question:

How long would it take (in months) to train a smart recent college graduate with no specialized training in my field to complete this task?

In the case of LLM-like systems, we would replace “smart recent college graduate” with “person who has read the entire internet”.

This is kinda related to my belief that knowledge-encoded-in-weights can do things that knowledge-encoded-in-the-context-window can’t. There is no possible context window that turns GPT-2 into GPT-3, right?

So when I try to think of things that I don’t expect LLM-like-systems to be able to do, I imagine, for example, finding a person adept at tinkering, and giving them a new machine to play with, a very strange machine unlike anything on the internet. I ask the person to spend the next weeks or months understanding that machine. So the person starts disassembling it and reassembling it, and they futz with one of the mechanism and see how it affects the other mechanisms, and they try replacing things and tightening or loosening things and so on. It might take a few weeks or months, but they’ll eventually build for themselves an exquisite mental model of this machine, and they’ll be able to answer questions about it and suggest improvements to it, even in only 1 second of thought, that far exceed what an LLM-like AI could ever do.

Maybe you’ll say that example is unfair because that’s a tangible object and robotics is hard. But I think there are intangible examples that are analogous, like people building up new fields of math. As an example from my own life, I was involved in early-stage design for a new weird type of lidar, including figuring out basic design principles and running trade studies. Over the course of a month or two of puzzling over how best to think about its operation and estimate its performance, I wound up with a big set of idiosyncratic concepts, with rich relationships between them, tailored to this particular weird new kind of lidar. That allowed me to have all the tradeoffs and interrelationships at the tip of my tongue. If someone suggested to use a lower-peak-power laser, I could immediately start listing off all the positive and negative consequences on its performance metrics, and then start listing possible approaches to mitigating the new problems, etc. Even if that particular question hadn’t come up before. The key capability here is not what I’m doing in the one second of thought before responding to that question, rather it’s what I was doing in the previous month or two, as I was learning, exploring, building concepts, etc., all specific to this particular gadget for which no remotely close analogue existed on the internet.

I think a similar thing is true in programming, and that the recent success of coding assistants is just because a whole lot of coding tasks are just not too deeply different from something-or-other on the internet. If a human had hypothetically read every open-source codebase on the internet, I think they’d more-or-less be able to do all the things that Copilot can do without having to think too hard about it. But when we get to more unusual programming tasks, where the hypothetical person would need to spend a few weeks puzzling over what’s going on and what’s the best approach, even if that person has previously read the whole internet, then we’re in territory beyond the capabilities of LLM programming assistants, current and future, I think. And if we’re talking about doing original science & tech R&D, then we get into that territory even faster.

How long would it take (in months) to train a smart recent college graduate with no specialized training in my field to complete this task?


This doesn't seem like a great metric because there are many tasks that a college grad can do with 0 training that current AI can't do, including:

  • Download and play a long video game to completion
  • Read and summarize a whole book
  • Spend a month planning an event

I do think that there's something important about this metric, but I think it's basically subsumed by my metric: if the task is "spend a month doing novel R&D for lidar", then my framework predicts that we'll need 1-month AGI for that. If the task is instead "answer the specific questions about lidar which this expert has been studying", then I claim that this is overfitting and therefore not a fair comparison; even if you expand it to "questions about lidar in general" there's probably a bunch of stuff that GPT-4 will know that the expert won't.

For the t-AGI framework, maybe you should also specify that the human starts the task only knowing things that are written multiple times on the internet. For example, Ed Witten could give snap (1-second) responses to lots of string theory questions that are WAY beyond current AI, using idiosyncratic intuitions he built up over many years. Likewise a chess grandmaster thinking about a board state for 1 second could crush GPT-4 or any other AI that wasn’t specifically and extensively trained on chess by humans.

I feel pretty uncertain about this, actually. Sure, there are some questions that don't appear at all on the internet, but most human knowledge is, so you'd have to cherry-pick questions. And presumably GPT-4 has also inferred a bunch of intuitions from internet data which weren't explicitly written down there. In other words: even if this is true, it doesn't feel centrally relevant.

Ah, that’s helpful, thanks.

Sure, there are some questions that don't appear at all on the internet, but most human knowledge is, so you'd have to cherry-pick questions.

I think you’re saying “there are questions about string theory whose answers are obvious to Ed Witten because he happened to have thought about them in the course of some unpublished project, but these questions are hyper-specific, so bringing them up at all would be unfair cherry-picking.”

But then we could just ask the question: “Can you please pose a question about string theory that no AI would have any prayer of answering, and then answer it yourself?” That’s not cherry-picking, or at least not in the same way.

And it points to an important human capability, namely, figuring out which areas are promising and tractable to explore, and then exploring them. Like, if a human wants to make money or do science or take over the world, then they get to pick, endogenously, which areas or avenues to explore.

But then we could just ask the question: “Can you please pose a question about string theory that no AI would have any prayer of answering, and then answer it yourself?” That’s not cherry-picking, or at least not in the same way.


But can't we equivalently just ask an AI to pose a question that no human would have a prayer of answering in one second? It wouldn't even need to be a trivial memorization thing, it could also be a math problem complex enough that humans can't do it that quickly, or drawing a link between two very different domains of knowledge.

I think the “in one second” would be cheating. The question for Ed Witten didn’t specify “the AI can’t answer it in one second”, but rather “the AI can’t answer it period”. Like, if GPT-4 can’t answer the string theory question in 5 minutes, then it probably can’t answer it in 1000 years either.

(If the AI can get smarter and smarter, and figure out more and more stuff, without bound, in any domain, by just running it longer and longer, then (1) it would be quite disanalogous to current LLMs [btw I’ve been assuming all along that this post is implicitly imagining something vaguely like current LLMs but I guess you didn’t say that explicitly], (2) I would guess that we’re already past end-of-the-world territory.)

Why is it cheating? That seems like the whole point of my framework - that we're comparing what AIs can do in any amount of time to what humans can do in a bounded amount of time.

Whatever. Maybe I was just jumping on an excuse to chit-chat about possible limitations of LLMs :) And maybe I was thread-hijacking by not engaging sufficiently with your post, sorry.

This part you wrote above was the most helpful for me:

if the task is "spend a month doing novel R&D for lidar", then my framework predicts that we'll need 1-month AGI for that

I guess I just want to state my opinion that (1) summarizing a 10,000-page book is a one-month task but could come pretty soon if indeed it’s not already possible, (2) spending a month doing novel R&D for lidar is a one-month task that I think is forever beyond LLMs and would require new algorithmic breakthroughs. That’s not disagreeing with you per se, because you never said in OP that all 1-month human tasks are equally hard for AI and will fall simultaneously! (And I doubt you believe it!) But maybe you conveyed that vibe slightly, from your talk about “coherence over time” etc., and I want to vibe in the opposite direction, by saying that what the human is doing during that month matters a lot, with building-from-scratch and exploring a rich hierarchical interconnected space of novel concepts being a hard-for-AI example, and following a very long fiction plot being an easy-for-AI example (somewhat related to its parallelizability).

Yeah, I agree I convey the implicit prediction that, even though not all one-month tasks will fall at once, they'll be closer than you would otherwise expect not using this framework.

I think I still disagree with your point, as follows: I agree that AI will soon do passably well at summarizing 10k word books, because the task is not very "sharp" - i.e. you get gradual rather than sudden returns to skill differences. But I think it will take significantly longer for AI to beat the quality of summary produced by a median expert in 1 month, because that expert's summary will in fact explore a rich hierarchical interconnected space of concepts from the novel (novel concepts, if you will).

Good post! I think I basically agree with you except I think that I would add that the stuff that can't be done by the end of 2025 will be doable by the end of 2027 (with the possible exception of manual labor, that might take another year or two). Whereas I take it you think it'll take longer than that for e.g. robustly pursuing a plan over multiple days to happen. 

Care to say what you think there -- how long until e.g. AI R&D has been dramatically accelerated by AIs doing much of the cognitive labor? How long until e.g. a souped-up version of AutoGPT can carry out a wide range of tasks on the internet/computers, stringing them together to coherently act towards goals on timespans of multiple days? (At least as coherently as a typical human professional, let's say?)

My default (very haphazard) answer: 10,000 seconds in a day; we're at 1-second AGI now; I'm speculating 1 OOM every 1.5 years, which suggests that coherence over multiple days is 6-7 years away.

The 1.5 years thing is just a very rough ballpark though, could probably be convinced to double or halve it by doing some more careful case studies.

Thanks. For the record, my position is that we won't see progress that looks like "For t-AGI, t increases by +1 OOM every X years" but rather that the rate of OOMs per year will start off slow and then accelerate. So e.g. here's what I think t will look like as a function of years:

YearRichard (?) guessDaniel guess
202315
2024515
202525100
20261002000
2027500Infinity (singularity)
20282,500 
202910,000 
203050,000 
2031250,000 
20321,000,000 

I think this partly because of the way I think generalization works (I think e.g. once AIs have gotten really good at all 2000-second tasks, they'll generalize quickly with just a bit more scaling, tuning, etc. to much longer tasks) and partly because of R&D acceleration effects where e.g. once AIs have gotten really good at all 2000-second tasks, AI research can be partially automated and will go substantially faster, getting us to 10,000-second tasks quicker, which then causes further speedup in R&D, etc.

@Richard_Ngo Seems like we should revisit these predictions now in light of the METR report https://metr.org/AI_R_D_Evaluation_Report.pdf 

This also is related to the crux between me and Ajeya Cotra, between me and Paul Christiano, between me and Rohin Shah... I think their view is that the "2020 AGI/TAI training requirements" variable is a lot higher than I think (they are thinking something like 1e36 FLOP, I'm thinking something like 1e29) because they are thinking you'll need to do lots and lots of long-horizon training to get systems that are good at long-horizon tasks, whereas I'm thinking you'll be able to get away with mostly training on shorter tasks and then a bit of fine-tuning on longer tasks. 

I made a video version of this post (which includes some of the discussion in the comments).
 

Hello,

Your viewpoint is very interesting. I have two questions.

If we use an "expert threshold" to define advanced artificial intelligence, what would the threshold for superintelligence be? There are exceptional individuals, such as Leonardo da Vinci or Einstein, who could be considered superintelligent. 

I believe, in order to be a fair comparison, superintelligence should be compared to these rare cases. Yet, we lack sufficient data on such individuals, and even if we had the data, our current systems might not comprehend these unique forms of intelligence. What's your perspective on this?

Also, you characterized superintelligence as the equivalent of 8 billion AI systems working for one year. However, it's impractical to have 8 billion people collaborate on a task for a year. How should this be measured appropriately?

Thank you.