Poker Playing Bot Adapted for Use by the Pentagon

A poker bot created by researchers at Carnegie Mellon University has now been adapted for an unlikely project – to serve the US military.  The bot, named Libratus (meaning ‘balanced’ in Latin), attracted world attention when it soundly beat four leading poker players at no-limit Texas Hold ‘em games in 2017. At the time, this was seen as a milestone in Artificial Intelligence development, mainly because the bot needed to be able to calculate some information what wasn’t visible. The bot won some $1.8 million in play money from the poker players by calculating how they may respond to its decisions, and was even able to bluff and develop betting strategies.

Now, the technology used for Libratus to beat the poker players – known as computational game theory – has been adapted for government use. The professor who led the Carnegie Mellon U project, Tuomas Sandholm, created a startup called Strategy Robot with the objective of using the same technology for the Pentagon to use in war games and simulations to explore military strategy. Strategy Robot inked a two-year $10 million deal with the US Army last year. The will support a relatively new military unit called the Defense Innovation Unit.

The developers are keeping mum on the military projects that Libratus will work on, for obvious security reasons. However, it is known that the bot can make suggestions regarding where to place military units for example.

According to an article in Wired, “pro pokers who took on the bot found that it flipped unnervingly between tame and hyper-aggressive tactics, all the while relentlessly notching up wins as it calculated paths to victory.”

The article quoted Prof. Sandholm as saying: “That opens yourself up to a lot of exploitation, because the real adversary may not play according to your assumptions.”

A second startup, called Strategic Machine, has been founded – also based on the bot’s technology – which is being used in commercial settings.


Comments are closed.
Google Analytics Alternative