The author is concerned about the growing role of robots in warfare. Since 1991, since the Gulf War, they have been helping the generals. And now a "killer robot" has been created - a machine that independently makes a decision to destroy enemies, if this is consistent with the general policy of the "control center".
Artificial intelligence developer DeepMind has announced a novelty - a bot called AlphaStar, which will play the popular online strategy Starcraft-2 at the "grandmaster" level.
This is not the first time a bot has beaten people in a strategic war game. In 1981, a program called Eurisko, developed by one of the pioneers of artificial intelligence, Doug Lenat, won the US Traveler Championship, a highly complex strategic war game in which the player must complete his fleet and fight the enemy fleet. Eurisko was eventually awarded the title of Honorary Admiral Traveler. In 1982, the tournament rules were changed in an attempt to thwart the computer. But Eurisko won everyone for two years in a row. After sponsors promised to cancel the tournament if the computer wins again, Lenat retreated.
DeepMind PR would like you to believe that StarCraft "emerged by consensus as the next big challenge (in PC games)" and "has been a challenge for AI developers for over 15 years."
In the last Starcraft Championship, only 4 applications were submitted from academic or industrial research labs. The other nine bots were written alone by individuals, and not as part of the mainstream of research on artificial intelligence (AI).
In fact, the number of authors (42) of the DeepMind article, published Oct 31 in Nature, significantly outnumbers the number of people working on Starcraft bots in the rest of the world. I do not want to overshadow the success of this engineering collaboration, but if you invest enough resources in solving the problem, success is guaranteed.
Unlike recent victories in computer chess, AlphaStar has not been trained by playing against itself. He learned this by imitating the best episodes from nearly a million games played by the very best human players. Prior to this, AlphaStar suffered an undeniable defeat in 19 of 20 fights, playing with real people on the StarCraft game server. AlphaStar played anonymously to prevent people from exploiting weaknesses that might have been found in previous games.
Last December AlphaStar defeated one of the world's best professional players - Grzegorz MaNa Komincz. But this was a version of AlphaStar with reflexes that were significantly faster than human ones, and also with unlimited vision of the playing field (unlike human players who see only one part of it at a time). It can hardly be called a level playing field.
Promotional video:
However, StarCraft has some features that make AlphaStar an impressive development, if not a real breakthrough. Unlike chess or the eastern game of go (a strategy game that uses a 361-field board - ed.), StarCraft players have an imperfect understanding of the state of the game, but the number of possible actions in it is much greater. And StarCraft combat happens in real time, but requires long-term planning.
Robot Wars
The question arises - will we see in the future how robots not only participate in wars, but also plan them? In fact, both are already happening.
Despite many warnings from artificial intelligence researchers (like myself), founders of robotics companies, Nobel laureates and church leaders, a fully autonomous weapon known as a "killer robot" has already been created and will soon be used.
In 2020, Turkey will deploy kamikaze drones on the border with Syria. These drones will use computer vision to identify, track and kill people without human intervention.
This is a terrible development. Computers do not have the moral ability to make a morally balanced decision about who will live and who will die. They lack empathy or compassion. Killer robots will change the very nature of conflict for the worse.
What about "robotic generals"? Computers have helped generals plan war for decades. Back in the early 1990s, during Operation Desert Storm during the Gulf War, artificial intelligence (AI) tools were used to plan a build-up in the Middle East before the conflict. A US general told me shortly thereafter that the amount of money saved was equal to the sum of everything that had been spent on AI development before.
Generals also use computers to develop multivariate military strategies that may suddenly be needed. But we do not leave all decisions on the battlefield in the hands of one soldier, therefore, shifting all responsibility from the general's brain to the computer processor is going too far.
The machine cannot be held accountable for its decisions. Only people are responsible. And this is the cornerstone of international humanitarian law. However, generals will increasingly rely on computer support in their decisions to break through the fog of war and deal with the vast amount of information coming from the front.
If this results in fewer civilian casualties, less friendly fire, and greater respect for international humanitarian law, we should welcome such computer assistance. But the final decision must be made by man, not machine.
Here is one final question to ponder. If tech companies like Google really don't want us to fear a machine uprising, why are they building bots to win virtual wars instead of focusing on something more peaceful like peaceful electronic games? With all due respect to sports fans, we have to admit that the stakes on non-war games come out much lower.
Toby Walsh is a professor of artificial intelligence at the University of New South Wales and the Data61 research community.