Libratus, the poker playing AI bot, has been drafted for the US military
Libratus, the AI bot that made the news in 2017 for cleaning out four professional poker players in a game of no-limit Texas Hold 'em, has been called up to work for The Pentagon. Tuomas Sandholm's new startup, called Strategy Robot, is using the same technology today for government use. It now has a $10 million contract with the US Army.
Public records revealed the contract, which is described as “in support of” a Pentagon agency known as the Defense Innovation Unit. The AI technology that Sandholm developed (he was project lead on Libratus) is being re-designed for use in simulations and wargames which will help the US military with strategy and planning.
The AI is built on computational game theory, sometimes called algorithmic game theory, which combines elements of game theory and computer science. Its objective is to understand and design algorithms in strategic environments. When Libratus was pitted against top poker players, it won more than $1.8 million in play money. The software was able to outwit the human players, learning to predict behavior and even bluff.
Sandholm is understandably reluctant to go into detail on what the US military is doing with this AI technology, but he did tell our friends at Wired that Strategy Robot has at least one other military contract on top of the $10 million deal already mentioned. The Defense Advanced Research Projects Agency (DARPA) is also looking into using AI in complex military decision making.
We have also seen AI prevail over champion chess players as well as Chinese strategy board game, Go. The aim in Go is to surround more territory than your opponent, with obvious military overlap.
Strategy Robot is not the only AI company the Pentagon is getting into bed with. Just this week, BGR reported that the US Army is working on autonomous weapons. According to the report, Army’s Assistant Secretary for Acquisitions, Logistics and Technology Bruce Jette, said at a Defense Writers Group breakfast in Washington DC that it would dangerous for the US to rely on humans if enemies are using AI-controlled weapons. Basically, we would not be a match for the machines.
The main challenge for Sandholm and Strategy Robot right now is portability. When Libratus crushed those poker players, it was running Pittsburgh Supercomputing Center's Bridges hardware (pictured above). As Sandholm righty points out, some "platforms" can't carry large computers.
What do you think about AI use in the military? Are we in danger of seeing killer robots becoming a reality?
I'm not going to worry about the rise of killer robots until my PC stops crashing; or at least until any company anywhere on earth makes a "troubleshooter" that actually figures out and fixes what caused the crash.
Anybody who's seen the retired military think tankers regularly empaneled on cable news, knows them (male and femaie) to be granite jawed, rock faced impassive and completely assured - regardless of the war at hand. AI, as poker analyst, has no "tells", it will not roll its eyes at another player, or blush and hunch a good hand closer to its vest. Perfect. Not accurate, by any stretch, but perfect.