The speaker delves into the interplay between theory and practice in computational game theory, mentioning examples like Deep Blue and AlphaGo as successes achieved through a combination of scientific and engineering efforts.
Researchers are finding a beautiful moment in using games to advance technology and popularize AI. As AI becomes more mature, games serve as an obvious way to explain what it is to a larger community.
The guest, Dr. Michael Jordan, reflects on the "AI winter" and the lack of progress in the field for 30 years due to limited computer capability, and shares his views on the principles necessary for sustainable AI platforms in the future. He credits the success of AI today to the invention of deep learning models by mathematicians such as Jeff Hinton.
Using AI for game balancing could save game developers thousands of hours of human testers to balance a game, as demonstrated by the success of AlphaGo in self-balancing during millions of games played against itself.
Self-play can quickly develop reasonable strategies in StarCraft despite long sequences of actions in games. This is possible with at least 10 minutes per game on average, making it quicker than relying on supervised learning for large LSTMs or transformers.
Machine learning is proving to be an important tool for the analysis of astronomy data, such as that provided by the Gaia satellite, which has already cataloged over 2 billion Milky Way stars. These methods allow for simulations to be made and initial conditions to be tweaked in order to better understand phenomena such as galaxy size and shape, and it can also lead to progress in medical fields.