Artificial Intelligence’s Winners and Losers
This post has been viewed 3,673 times.
People, get ready.
“We’ve crossed a critical threshold,” says Ken Sena, a Wells Fargo analyst focused on A.I. Artificial intelligence now has the programming strategies, processor power, and abundant data to transform many industries. Sena got Wells Fargo to build a machine learning system called Aiera to assist in his equity research.
A superhuman Go program has been one of A.I.’s grand challenges, so DeepMind chose the ancient Chinese game as a test for its “neural network” software, modeled on the structure of our brains. Earlier versions of the AlphaGo software learned the strategies through a carefully supervised study of more than 100,000 recorded games played by human experts. The latest version, AlphaGo Zero, started from a blank slate. With no human guidance or knowledge, beyond the basic rules, AlphaGo Zero played 30 million simulated games over 40 days—discovering for itself the strategies learned by humans over millennia and finding new ones.
This has thrilled computer scientists, because training A.I. systems for any task has been an expensive, time-consuming bottleneck. Systems like AlphaGo may soon exhibit superhuman performance on other tasks, if appropriately well-defined and predictable. “Machine learning can’t do everything that you and I can do,” says Sena. “It can do some things better and some things worse.”
Read original articleTags:
