[ad_1]
Google DeepMind is revealed to be effective enough to master various games over the years. He is officially the best Go player in the world after defeating the sitting human world champion and has since turned his figurative hand to Starcraft 2 while also being able to guess when the patients are going to die. It's really a jack of all trades.
The problem is that DeepMind is a lone wolf. It works alone and does not necessarily know how to play with humans – or other AIs. You might not consider the best cure for arming DeepMind, but the good news is that the weapons and the battlefield are purely virtual: the AI has learned to play Capture the multiplayer mode of Quake III and, predictably, has reached the point where he can teach humans a thing or two.
For those who do not know about Capture the Flag, like "Pin the Tail on the donkey" the name really leaves little to the imagination. Two armed teams have a flag and must bring their opponent back to his base to score a point. If they are pulled en route, the flag is returned to the base.
READ MORE: DeepMind Uses Street View to Teach AI to Navigate Cities
You're probably aware of it now. Give him a controller, and you will make a fair fist – but it's a little harder for artificial intelligence, who not only has to learn the rules and tactics, but even the basics of the game – move, change weapon, shoot, and even learn to spot what an enemy looks like. He learns all this from the field by experimenting, which means that you will probably have to make several hundred thousand draws before anyone knows how to score a point. This is not an exaggeration: it took DeepMind nearly half a million five-minute games on randomly generated cards to get in tune.
Once there, it turned out to be huge. DeepMind's AI agents have learned the basics, but have also learned human strategies, like keeping their own flag, camping on the opponent's base and teaming up to make sure that they are safe. they were more numerous than their enemies.
These tactics worked. The researchers organized a mini-tournament in the rather approximate name of the job, mixing 40 human helpers with DeepMind in a combination of purely human teams, teams reserved for AI and one-man teams. mixture of both. The best combination was the bot-only team, which managed to win with a 74% win probability. Middling humans had a 43% chance, while strong human players only managed 52%.
READ NEXT: 10 Things You Need to Know About the Future of AI
The AI of DeepMind is Better Than Humans, She Also Knows How To Cooperate with the others. Oh.
However, there is a fairly important warning: that the probability of victory of 74% is for 2v2 matches. When teams of four were introduced, the effectiveness of DeepMind dropped to 65%. This is even more effective than humans, but suggests that some of the lessons do not seem to evolve towards larger teams.
Of course, beating humanity to its own game is a side effect, rather than the intention of the research. Teaching the AI to cooperate is a noble goal, and if there is one thing that it can learn from humans as a species historically, it is that we are definitely more than the sum of our parts.
Source link
