What it is: Artificial intelligence (AI) programmers are testing whether AI can beat humans at the StarCraft video game.
For decades, artificial intelligence was all hype with little to show for its efforts. The big problem with AI in the early days was that programmers tried to create intelligent computers by programming them to be intelligent. That meant the programmers themselves had to know how intelligence worked and how it could handle unforeseen problems in the future. Not surprisingly, these early attempts at AI failed or had to be scaled back dramatically.
Where AI initially succeeded was in tackling niche area and one of the most successful was game playing. To demonstrate AI, programmers first tackled the problem of beating a grandmaster at chess. Chess programs essentially got stronger simply by being faster and being able to analyze chess moves faster and more comprehensive than any human could possibly do.
This brute-force approach to AI worked, but kept AI limited to narrow realms where the rules were clear and fairly simple. When AI programs started to beat chess masters regularly, that’s when the AI community realized they needed to do more than just tackle a narrow niche like chess.
One of the next areas AI researchers tackled was playing Go, which is a game that’s far more complex than chess. Initially people thought there could never be an AI program that could defeat a Go master, but AlphaGo, an artificially intelligent Go player created by Google, recently defeated Lee Sedol, one of the top Go players in the world.
AI chess programs typically won by brute force, but AlphaGo was one of the first AI programs that could learn from its mistakes without being specifically reprogrammed by a human. With the success of AlphaGo, AI programmers are now tackling StarCraft, a real-time strategy game that requires players to coordinate multiple actions and deal with ever-changing conditions. In other words, unlike chess or Go, StarCraft is far more complex.
Old fashion AI programs that relied on brute force could never master StarCraft (or Go for that matter), but machine learning AI programs can simply play endless rounds of StarCraft and gradually learn from its mistakes each time. Eventually the goal is to see whether an AI program can master StarCraft against human players.
AI programs are continuing to get smarter and create the illusion of intelligence. The question is when does the illusion of intelligence disappear and become accepted as true intelligence? Handling restricted domains like chess, Go, or even StarCraft is impressive, but how can AI programs handle the world outside their speciality?
As comedian Emo Phillips once said, “A computer once beat me at chess, but it was no match for me at kick boxing.”