We’ve built computers that can outplay the finest chess grandmasters in the world, virtual personal assistants that can schedule tasks and control our homes, and algorithms that can predict with increasing accuracy what we’ll want to watch, read, or listen to next.
But true artificial intelligence – a computer that can solve a wide range of problems through reason, planning, abstraction, and learning – hasn’t come about yet. There are machines that are better than humans at specific tasks, but no machine that’s as good as or better than a human at thinking.
We’re getting close to that point, though, Google chairman Eric Schmidt argued in an op-ed for the BBC on Saturday. Mr. Schmidt says artificial intelligence (AI) research has been steadily building since the term was first coined in 1955, and that scientists have made a few big breakthroughs in the past several years.
One of those breakthroughs, Schmidt writes, came from a team led by AI pioneer Geoff Hinton. In 2009, Dr. Hinton’s team beat the state of the art for automated speech recognition and immediately cut the number of errors in Google’s speech recognition engine by 25 percent, “the equivalent of ten years of research all at once.”
At around the same time, AI was boosted by Big Data – thousands of computers all networked together, providing resources from which machines could learn. That sort of networking allowed computers to begin tackling problems such as identifying a written language from a picture or a spoken language from a sound file, or recognizing the subject of an image. (It’s trivial for a person to look at a picture of a dog and recognize that it’s a dog, but that kind of identification was, until recently, very difficult for computers.)
Schmidt says it’s important for AI researchers to direct their efforts toward solving real-world problems such as automatically filtering e-mails and social media posts for users, planning vacations, or even finding the next pop superstar.
“A decade ago, to launch a digital music service, you probably would have enlisted a handful of elite tastemakers to pick the hottest new music,” Schmidt writes. “Today, you're much better off building a smart system that can learn from the real world - what actual listeners are most likely to like next - and help you predict who and where the next Adele might be.”
Schmidt predicts that machine learning, and artificial intelligence more generally, will improve at an increasing rate in the next several years, and that an algorithmic approach to big problems in areas such as genomics, energy, and climate science will wholly replace a traditional approach.
“In the next generation of software, machine learning won't just be an add-on that improves performance a few percentage points,” Schmidt writes.
There’s no telling when, or even whether, true artificial intelligence will be created, but the Google chairman argues that breakthroughs in this area could be instrumental to solving the biggest problems facing humanity.