October 18, 2018

Archives for February 15, 2018

Learning to Learn

“I didn’t know robots had advanced so far,” a reader remarked after last week’s post about how computers are displacing knowledge workers. What changed to make that happen? The machines learned how to learn.

This is from Artificial Intelligence Goes Bilingual—Without A Dictionary, Science Magazine, Nov. 28, 2017.

“Imagine that you give one person lots of Chinese books and lots of Arabic books—none of them overlapping—and the person has to learn to translate Chinese to Arabic. That seems impossible, right?” says . . . Mikel Artetxe, a computer scientist at the University of the Basque Country (UPV) in San Sebastiàn, Spain. “But we show that a computer can do that.”

Most machine learning—in which neural networks and other computer algorithms learn from experience—is “supervised.” A computer makes a guess, receives the right answer, and adjusts its process accordingly. That works well when teaching a computer to translate between, say, English and French, because many documents exist in both languages. It doesn’t work so well for rare languages, or for popular ones without many parallel texts.

[This learning technique is called] unsupervised machine learning. [A computer using this technique] constructs bilingual dictionaries without the aid of a human teacher telling them when their guesses are right.

Hmm. . . . I could have used that last year, when my wife and I spent three months visiting our daughter in South Korea. The Korean language is ridiculously complex; I never got much past “good morning.”

Go matches were a standard offering on the gym TV’s where I worked out. (Imagine two guys in black suits staring intently at a game board — not exactly a riveting workout visual.) Go is also ridiculously complex, and mysterious, too: the masters seem to make moves more intuitively than analytically. But the days of human Go supremacy are over. Google wizard and overall overachiever Sebastian Thrun[1] explains why in this conversation with TED Curator Chris Anderson:

Artificial intelligence and machine learning is about 60 years old and has not had a great day in its past until recently. And the reason is that today, we have reached a scale of computing and datasets that was necessary to make machines smart. The new thing now is that computers can find their own rules. So instead of an expert deciphering, step by step, a rule for every contingency, what you do now is you give the computer examples and have it infer its own rules.

A really good example is AlphaGo. Normally, in game playing, you would really write down all the rules, but in AlphaGo’s case, the system looked over a million games and was able to infer its own rules and then beat the world’s residing Go champion. That is exciting, because it relieves the software engineer of the need of being super smart, and pushes the burden towards the data.

20 years ago the computers were as big as a cockroach brain. Now they are powerful enough to really emulate specialized human thinking. And then the computers take advantage of the fact that they can look at much more data than people can. AlphaGo looked at more than a million games. No human expert can ever study a million games. So as a result, the computer can find rules that even people can’t find.

Thrun made those comments in April 2017. AlphaGo’s championship reign was short-lived: it was unseated a mere six months by a new cyber challenger that taught itself without reviewing all that data. This is from “AlphaGo Zero Shows Machines Can Become Superhuman Without Any Help,” MIT Technology Review, October 18, 2017.

AlphaGo wasn’t the best Go player on the planet for very long. A new version of the masterful AI program has emerged, and it’s a monster. In a head-to-head matchup, AlphaGo Zero defeated the original program by 100 games to none.

Whereas the original AlphaGo learned by ingesting data from hundreds of thousands of games played by human experts, AlphaGo Zero started with nothing but a blank board and the rules of the game. It learned simply by playing millions of games against itself, using what it learned in each game to improve.

The new program represents a step forward in the quest to build machines that are truly intelligent. That’s because machines will need to figure out solutions to difficult problems even when there isn’t a large amount of training data to learn from.

“The most striking thing is we don’t need any human data anymore,” says Demis Hassabis, CEO and cofounder of DeepMind [the creators of AlphaGo Zero].

“By not using human data or human expertise, we’ve actually removed the constraints of human knowledge,” says David Silver, the lead researcher at DeepMind and a professor at University College London. “It’s able to create knowledge for itself from first principles.”

Did you catch that? “We’ve removed the constraints of human knowledge.” Wow. No wonder computers are elbowing all those knowledge workers out of the way.

What’s left for human to do? We’ll hear from Sebastian Thrun and others on that topic next time.


[1] Sebastian Thrun’s TED bio describes him as “an educator, entrepreneur and troublemaker. After a long life as a professor at Stanford University, Thrun resigned from tenure to join Google. At Google, he founded Google X, home to self-driving cars and many other moonshot technologies. Thrun also founded Udacity, an online university with worldwide reach, and Kitty Hawk, a ‘flying car’ company. He has authored 11 books, 400 papers, holds 3 doctorates and has won numerous awards.”

 

Kevin Rhodes writes about individual growth and cultural change, drawing on insights from science, technology, disruptive innovation, entrepreneurship, neuroscience, psychology, and personal experience, including his own unique journey to wellness — dealing with primary progressive MS through an aggressive regime of exercise, diet, and mental conditioning.

Check out Kevin’s latest LinkedIn Pulse article: Leadership and Life Lessons From an Elite Athlete and a Dying Man.