Silicon wafer manufacturing contributes to many remarkable achievements in technology. One of these is artificial intelligence, a self-learning entity powered by chips built on wafer substrates. AI has hit numerous milestones already, the most recent of which is a bot that succeeded in playing poker against real people.
Software developers and mathematicians have always found poker to be a massive enigma for programming. This is due to its nature as an “imperfect-information” environment. Unlike chess, which allows a bot to study pre-determined moves, game information is usually withheld from all participants in poker.
The two AI systems discussed below were developed by Noam Brown and Tuomas Sandholm. Both worked at Carnegie Mellon University, where Brown took graduate studies with Sandholm as his professor.
Libratus first came out five years ago when it beat top poker players, including Dong Kim, in a three-week competition at a Pittsburgh casino. The bot became notable for not using neural networks. Instead, it utilized three built-in systems that worked together to help the AI learn from past games and grow proficient.
Libratus essentially learned from trial-and-error. The leading software trained itself by dealing trillions of poker hands until it became proficient enough to challenge humans. The second system acted like a memory; it went through past games during play and recommended them to the central program while also analyzing and learning new patterns.
Lastly, Libratus employed a third program to study the day’s games and remove repeated patterns that human players can take advantage of.
Libratus’ success led to the development of Pluribus, a collaboration between the CMU developers and Facebook AI. While Libratus excelled at one-on-one poker games, the new robot took on six players in Texas Hold ‘Em.
Pluribus built on the knowledge and development gained from Libratus. Still, it had to start learning from scratch like its predecessor. Its main distinction from Libratus, however, is it looked forward to only a few moves ahead instead of planning toward the game’s end.
Facebook also claimed to have made Pluribus consume less than $150 worth of cloud computing resources in training. It requires less processing power and silicon wafers than other AI projects.
Our stringent manufacturing and quality control protocols ensure wafers with high production yields. Call us now if you’d like a free quote.