What Happened With Bio Anchors?

[Original post: Biological Anchors: A Trick That Might Or Might Not Work]

I.

Ajeya Cotra’s Biological Anchors report was the landmark AI timelines forecast of the early 2020s. In many ways, it was incredibly prescient - it nailed the scaling hypothesis, predicted the current AI boom, and introduced concepts like “time horizons” that have entered common parlance. In most cases where its contemporaries challenged it, its assumptions have been borne out, and its challengers proven wrong.

But its headline prediction - an AGI timeline centered around the 2050s - no longer seems plausible. The current state of the discussion ranges from late 2020s to 2040s, with more remote dates relegated to those who expect the current paradigm to prove ultimately fruitless - the opposite of Ajeya’s assumptions. Cotra later shortened her own timelines to 2040 (as of 2022) and they are probably even shorter now.

So, if its premises were impressively correct, but its conclusion twenty years too late, what went wrong in the middle?

II.

First, a refresher. What was Bio Anchors? How did it work?

In 2020, the most advanced AI, GPT-3, had required about 10^23 FLOPs to train.

(FLOPs are a measure of computation: big, powerful computers and data centers can deploy more FLOPs than smaller ones)

Cotra asked: how quickly is the AI industry getting access to more compute / more FLOPs? And how many FLOPs would AGI take? If we can figure out both those things, determining the date of AGI arrival becomes a matter of simple division.

She found that FLOPs had been increasing at a constant rate for many years. And if you looked at planned data center construction, it looked on track to continue increasing at about that rate. New technological advances (algorithmic progress) made each FLOP more valuable in training AIs, but that process also seemed constant and predictable. So there was relatively constant growth in effective FLOPs (amount of computation available, adjusted by ability to use that computation efficiently).

There was no obvious way to know how many FLOPs AGI would take, but there were some intuitively compelling guesses - for example, an AGI that was as smart as humans might need a similar level of computing capacity as the human brain. Cotra picked five intuitively compelling guesses (the namesake Bio Anchors) and turned them into a weighted average.

Then she calculated: given the rate at ...