Facebook robot beats champions at their own game
Developments in the last week show Facebook and Instagram are leading players in the field of Artificial Intelligence.
On Friday "Pluribus" became the first AI robot to beat five expert Texas Hold'em players at their own game.
Game playing robots use AI to teach themselves how to win through self-play. Pluribus is estimated to have played trillions of games against itself over a period of 8 days in the build up to Friday’s game. In self-play scenarios, bots play multiple copies of themselves and review the success of each hand they put forward. Crucially, they record a “regret factor” which is how much they regret playing a hand based on the outcome it creates. The self-play algorithm is programmed to update the bot’s strategy based on this regret factor. This increases the probability of the bots playing hands that previously had the least regret i.e. were the most successful.
Previously, the success of these robots has been limited to beating just one opponent at a time. The complexities and volume of hidden information presented by multi-player games has previously been too difficult for the bots to process and overcome. This is why Pluribus’ win on Friday is being heralded ground breaking for the advancement of AI, as shows that Facebook, alongside Carnegie Mellon University, have developed technology to overcome multiple pieces of hidden information, at the same time.
The implications of this are extensive. The majority of real world issues, such as cyber security, terrorism and other forms of harmful online content often involve hidden information, which is held by a number of different actors, much like the numerous poker players. The bots now have the power to deal with the multi-facial complexities of these situations in which they were previously limited. Now the algorithm has been cracked, watch this space for impending developments.
In the same week, Instagram, the social media platform which is owned by Facebook, rolled out its new feature which uses AI in different way; to help protect its users against cyber bullying. The software identifies negative comments that people are posting and automatically askes "are you sure you want to post this?". This question appears with an “undo” button which enables the person to undo the post before it is published.
This is hoped to prompt users to reflect on their comments and provide them with the immediate opportunity to remove the post to not cause any harm.
The new technology has been created in the wake of the company coming under under fire for not doing enough to prevent cyber bullying. In the UK it was estimated that over 42% of all 12-20 year olds experienced cyber bullying through social media platforms in 2017.
Head of Instagram Adam Mosseri agrees that the company should do more and aims to harness more AI initiatives to continue to tackle the problem.
To speak to Georgia: groberts@clarkslegal.com