Chapter
Clips
In order to compare the intelligence of human and artificial systems, it is important to control for priors, or the same set of knowledge, and experience, or training data.
1:06:26 - 1:12:56 (06:29)
Summary
In order to compare the intelligence of human and artificial systems, it is important to control for priors, or the same set of knowledge, and experience, or training data. To create more human-like AIs, a universal test should be developed to establish a comparison between human and artificial systems to determine the level of intelligence.
ChapterTesting Intelligence for Humans and Machines
Episode#120 – François Chollet: Measures of Intelligence
PodcastLex Fridman Podcast
The game of go relies on hard-coded priors about 2D space and winning conditions, and other board games also share similarities.
1:12:56 - 1:13:56 (01:00)
Summary
The game of go relies on hard-coded priors about 2D space and winning conditions, and other board games also share similarities. When using reinforcement learning in self-play, the number of priors may be low, highlighting the importance of making explicit any special priors in the game.