Summary of H-arc: a Robust Estimate Of Human Performance on the Abstraction and Reasoning Corpus Benchmark, by Solim Legris et al.
H-ARC: A Robust Estimate of Human Performance on the Abstraction and Reasoning Corpus Benchmark
by Solim LeGris, Wai Keen Vong, Brenden M. Lake, Todd M. Gureckis
First submitted to arxiv on: 2 Sep 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- Secondary: None
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The Abstraction and Reasoning Corpus (ARC) is a visual program synthesis benchmark designed to test challenging out-of-distribution generalization in humans and machines. Despite limited progress on this challenge using existing AI methods, understanding human performance is crucial for the benchmark’s validity. This paper provides a more robust estimate of human performance by evaluating 1729 individuals on the full ARC problem set, with average performance ranging from 73.3% to 77.2% correct on the training set and 55.9% to 68.9% correct on the public evaluation set. The study also finds that most tasks are solvable by typical crowd-workers recruited online, exceeding current AI approaches. To facilitate research, the paper publicly releases its dataset, H-ARC (human-ARC), which includes human submissions and action traces. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The Abstraction and Reasoning Corpus is a test for artificial intelligence to see how well it can solve problems that are different from what it was trained on. The goal is to make AI better at generalizing and learning new things. To do this, researchers need to know how humans perform on these types of challenges. This paper tested 1729 people on the same set of tasks used to train AI models, and found that most people were able to solve many of the problems correctly. This is important because it shows that even though AI may not be very good at this type of problem yet, humans are still much better. |
Keywords
* Artificial intelligence * Generalization
