If you’re confused, you’re forgiven. After all, how important is a trivial game?
But hear Domingos out. He sees something totally different when he looks at Tetris.
For the University of Washington computer science professor, Tetris as an example of an NP-complete problem, a term computer scientists use. It’s a problem we have no solution for, but if we did, we could verify quickly that we had found it.
“If you can solve Tetris, you can solve thousands of the hardest and most important problems in science, technology, and management — all in one fell swoop. That’s because at heart they are all the same problem,” he writes.
While algorithms are increasingly common in our lives — they do everything from detect credit card fraud, to set prices online and order Facebook’s newsfeed, those algorithms are all narrowly focused. Domingos’s hunch, one shared by many, is that there is one learning algorithm that can derive all knowledge from data. The trick is uncovering it. One popular method is reverse engineering the human brain. Others try to simulate evolution on a computer because of the diverse and intelligent life that the process created in nature.
Domingos says the invention of this master algorithm would be “one of the greatest scientific achievements of all time.” It would be worth trillions to a company, and invent everything we need to have invented.
Beating Tetris would be the smallest of the small potatoes. There’s the potential to cure cancers, give everyone a robot butler, and end the need to hold jobs.
But how we figure out the master algorithm is unclear. Domingos gives an overview of what he calls the five tribes of machine learning: symbolists, connectionists, evolutionaries, Bayesians and analogizers. As technology books on complex subjects go, it’s accessible for the general reader.
The connectionists have been hot lately, with advances in deep learning showing how computers can paint like the greatest artists, recognize our faces in photos and teach cars can drive themselves. Domingos, who has studied machine learning for decades, has seen paradigms go in and out of style. The advances arrive in occasional bursts rather than a steady drip.
He sees Google as the front-runner of the established players, but the race is far from over. To ultimately crack intelligence, Domingos thinks more people outside the field of machine intelligence need to be thinking about it.
“It’s completely up for grabs who exactly is going to solve this problem,” Domingos told me. “My guess is that it is probably someone who doesn’t come from any of these schools [of thought]. It’s going to be someone who is 20-years-old and just has a new idea.”
If we discover the master algorithm, society will change quickly. Many jobs will disappear. Domingos, while acknowledging some downsides and a potentially rocky adjustment period, ultimately sees the master algorithm as a huge positive.
He expects some people will become extremely wealthy — especially whoever discovers the master algorithm. But gains will be realized across society. With almost infinitely productive machines, we’ll be living in an era of abundance. Domingos is confident future generations will be far better off than current generations, even if they don’t have jobs. A basic income, guaranteed to all humans, would ensure a certain standard of living.
“If people’s lives are getting better they won’t actually mind inequality all that much,” said Domingos, which doesn’t share the concerns of some tech leaders that artificial intelligence is a massive threat to us.
He may sound like a blissful optimist, but in the grand scheme of things, technology has indeed been a tremendous positive for humanity. While it has brought new risks and annoyances, humans are generally wealthier, safer and healthier than at any point in history.
Now we’ll get to see if we can make the next leap.
“We’re at the point where we’re pretty close to doing it,” Domingos said. “But I also have the feeling that at the end of the day we are going to need some ideas no one has had yet.”