|
Briefing
|
- Poker Playing AI – Researchers from Facebook and Carnegie Mellon University built Pluribus, first artificial intelligence that defeated human experts in six-player no-limit Texas Hold'em poker game
- Humans vs Bot – AI bested human players in both “five AIs versus one human” and “one AI versus five humans” formats
- Improved AI – Pluribus is based on earlier work on Libratus, poker playing bot that defeated human pros in two-player Texas Hold'em game in 2017, and incorporates capabilities of online search and fast self-play algorithms
- More Efficient – Requires less processing power and memory than previous poker algorithms, training on 64-core server, less than 512 GB of RAM, and no GPU, the equivalent of less than $150 worth of cloud computing resources, compared to supercomputer hardware and millions of dollars in resources for prior AIs
- Self-Play Training – Pluribus devised own strategy by playing against itself, with researcher Tuomas Sandholm noting that bot learned to randomize its strategy in varying ways and probabilities for different situations
- Other Applications – Such AI that can work with hidden information and/or multiple agents can also be used in fraud prevention, cybersecurity, taking action on harmful content, drug design, military robotic systems, multi-party negotiation or pricing, optimal media spending, and auction bidding
|
