Top video game players are currently playing for a pool of almost $25 million’s worth in prizes in a game called DOTA 2, but one event that will take place on the main stage will be the most public display yet of a growing technology that will likely re-shape warfare in the way that the maturation of gunpowder or military aviation once did.
A team of artificial intelligence bots from OpenAI will take on a team of five professional human players and the AI will likely win. It will likely be the closest thing AI gets to a Billy Mitchell bombing the Ostfriesland moment.
The details of the game aren’t super important for this discussion, but you can read about it here if you care. Basically, the game is much more complex than the board games that AIs have been taking on in the last few years, and requires a much more complex system of evaluations and executions to win against humans, especially in team play.
Last year, the AI from OpenAI trained on one-vs-one matches and went from barely being able to control units in game to defeating the single-best human player in the world in less than six months. This year, it’s targeting team battles and has already defeated a team made up of players in the top 0.5 percent of the world.
Fans watch the proceedings during the 2014 DOTA 2 invitational.
So, what does this all have to do with war? OpenAI doesn’t exist to win video games. It’s a nonprofit started by big names like Elon Musk in order to advance safe AI (Artificial intelligence restricted to working for the benefit and safety of humans). In fact, most of OpenAI’s projects have nothing to do with video games. They just use DOTA 2 as a flashy way to get and keep people interested in their AI work.
What OpenAI really cares about is fueling breakthroughs in AI research and development for use in everything from managing cities to controlling factories. And while they don’t pursue military research, it’s not hard to see how a computer that can control a mage throwing fireballs across a digital battlefield might be taught how to control cruisers firing artillery shells across the water.
Other researchers have already created an AI that can outperform humans in small aerial dogfights. If an AI created with OpenAI’s deep-learning was aimed at that milestone, it could be expected to take on human opponents within a few months of creation, then win against teams within another year or less, and be able to dominate most human teams soon thereafter.
The U.S. Navy’s unmanned X-47B jet aircraft.
And that’s while we make the computer fly jets designed for humans and if it’s forced to treat its planes as assets it can’t sacrifice. But jets flown by humans don’t need to be constrained by the limits of the human body, meaning they can take tighter turns at higher speeds. And we don’t have to treat losing jets the same as we would losing jets with humans on board. The computer could treat them like DOTA 2 heroes: valuable, but ultimately disposable for the right gain.
And the U.S. and Chinese militaries, among others, know about these advantages of AI, and are pursuing AI technology for just that reason. And it won’t just apply to jets, but also submarines, armored vehicles, and potentially even infantry. After all, OpenAI has helped AIs train each other for controlling human-like bodies in everything from digital sumo matches to high winds.
So it’s easy to imagine that, in the next war, China and America will start turning more and more to their robot partners for help against their enemies, potentially each other.
For at least the next few decades, larger ships will still need human crews, which means that hundreds or thousands of sailors will still be at risk while fighting.
(U.S. Navy photo by Mass Communication Specialist 2nd Class Jacques-Laurent Jean-Gilles)
But this will create an even more frightening change in warfare than aerial bombing did. Sailors and soldiers will be asked to go forward either knowing that the decision has been made by an unfeeling machine or knowing that there was a chance the decision was made by a machine, and that they will be fighting a mix of machines and humans.
And AIs will likely be better at strategic decisions eventually, but it will still carry an added moral weight for troops knowing that they aren’t executing the will of a senior human, but a robot.
But, of course, it won’t be all bad. In isolated areas with little need for humans to safeguard against collateral damage, entire battles could be fought with little or no human losses.
A U.S. Marine leads a robot on a simulated patrol.
(U.S. Marine Corps photo by Kyle J. O. Olson)
Imagine a few decades into the future, one where robots can control warships and planes, submarines, and anti-aircraft guns. Now imagine the historic Battle of Midway where Japan lost five ships and 292 aircraft while suffering 2,500 casualties. The U.S. lost two ships and 145 aircraft while suffering 307 casualties.
Now, combine those two visions, a Battle of Midway where two of the world’s largest navies clash with almost no sailors having to fight at the front. Yes, the ships and planes would still be lost, and control of vast swaths of the world would still be decided by violent clashes, but the human sacrifice would be in the single or double digits.
And the victory for the winner will still be complete. After all, if the U.S. fleet survives at Midway, whether it is crewed by robots or humans, that’s still a physical fleet that can move towards the Japanese home islands.
So, yes, AI will almost certainly revolutionize warfare, and it will happen in the secrecy of classified labs until exploding into the open in a large war.