Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: AI Forecasting: One Year In, published by jsteinhardt on July 4, 2022 on LessWrong.
Last August, my research group created a forecasting contest to predict AI progress on four benchmarks. Forecasts were asked to predict state-of-the-art performance (SOTA) on each benchmark for June 30th 2022, 2023, 2024, and 2025. It’s now past June 30th, so we can evaluate the performance of the forecasters so far.
Forecasters were asked to provide probability distributions, so we can evaluate both their point estimates and their coverage (whether the true result was within their credible intervals). I’ll dive into the data in detail below, but my high-level takeaways were that:
Forecasters’ predictions were not very good in general: two out of four forecasts were outside the 90% credible intervals.
However, they were better than my personal predictions, and I suspect better than the median prediction of ML researchers (if the latter had been preregistered).
Specifically, progress on ML benchmarks happened significantly faster than forecasters expected. But forecasters predicted faster progress than I did personally, and my sense is that I expect somewhat faster progress than the median ML researcher does.
Progress on a robustness benchmark was slower than expected, and was the only benchmark to fall short of forecaster predictions. This is somewhat worrying, as it suggests that machine learning capabilities are progressing quickly, while safety properties are progressing slowly.
Below I’ll review the tasks and competition format, then go through the results.
Forecasting Tasks and Overall Predictions
As a reminder, the four benchmarks were:
MATH, a mathematics problem-solving dataset;
MMLU, a test of specialized subject knowledge using high school, college, and professional multiple choice exams;
Something Something v2, a video recognition dataset; and
CIFAR-10 robust accuracy, a measure of adversarially robust vision performance.
Forecasters were asked to predict performance on each of these. Each forecasting question had a $5000 prize pool (distributed across the four years). There were also two questions about compute usage by different countries and organizations, but I’ll ignore those here.
Forecasters themselves were recruited with the platform Hypermind. You can read more details in the initial blog post from last August, but in brief, professional forecasters make money by providing accurate probabilistic forecasts about future events, and are typically paid according to a proper scoring rule that incentivizes calibration. They apply a wide range of techniques such as base rates, reference classes, trend extrapolation, examining and aggregating different expert views, thinking about possible surprises, etc. (see my class notes for more details).
Here is what the forecasters’ point estimates were for each of the four questions (based on Hypermind's dashboard):
Expert performance is approximated as 90%. The 2021 datapoint represents the SOTA in August 2021, when the predictions were made.
For June 2022, forecasters predicted 12.7% on MATH, 57.1% on MMLU (the multiple-choice dataset), 70.4% on adversarial CIFAR-10, and 73.0% on Something Something v2.
At the time, I described being surprised by the 2025 prediction for the MATH dataset, which predicted over 50% performance, especially given that 2021 accuracy was only 6.9% and most humans would be below 50%.
Here are the actual results, as of today:
MATH: 50.3% (vs. 12.7% predicted)
MMLU: 67.5% (vs. 57.1% predicted)
Adversarial CIFAR-10: 66.6% (vs. 70.4% predicted)
Something Something v2: 75.3% (vs. 73.0% predicted)
MATH and MMLU progressed much faster than predicted. Something Something v2 progressed somewhat faster than predicted. In contrast, Adversarial CIFAR-10 progressed somewhat slower than predicted. Overall, progress on...
