The Lessons of AlphaGo

So Elon Musk is the go-to AI doomsday guy, but if an AI machine predicted to lose a challenge astounds pros when put up against the decade’s best Go player, it’s safe to say that might raise a little bit of concern.

I recently watched the documentary, directed by Gary Kohs, that chronicles the journey of Google’s DeepMind testing both the capabilities and limits of its self-learning AI technology.

The test? To master the abstract strategy board game of Go and compete against the best human players in the world, among them Go champion Lee Sedol. AlphaGo had previously demonstrated great skill in chess, thanks in part to programming from multiple high-ranked grand masters. But what about “Go”, a centuries-old game considered the most complex game ever developed by man? And what would result when AlphaGo was left to learn based off its neural networks, rather than from human programming?

Specialists had predicted that no AI program would be able to accomplish this feat for at least another ten years, and yet, AlphaGo presented itself as a strong representative for machine learning.

SPOILER ALERT: The machine wins. But what are the implications of such a victory?

Was It Just A Game?

On the one hand, the film tries to add levity by having AlphaGo demonstrate its knowledge in a board game. While Go ultimately is based on strategic aptitude and is often attributed to a certain level of brilliance and intelligence, this is a lower level feat than, say, performing heart surgery. Most, if not all, Go aficionados and champions initially doubted that AlphaGo could outsmart “human intuition”, and would therefore be beat either 5-0, or 4-1, on a bad day. But we soon learned the capabilities of the machine’s neural network learning.

AlphaGo’s game strategy came from multiple levels of training. The first tier was to study the matches of high-ranking Go players, and mimic human behavior and board movement based on those matches. Then AlphaGo would practice non-stop, competing against itself in millions of matches. AlphaGo had the unique ability to calculate all possible moves that could take place on the board, up to 50 or 60 moves ahead. This behavior offered AlphaGo a clear advantage against its human opponents. Some commentators noted that AlphaGo, unlike your typical Go player, would take longer periods of time to contemplate the next move, time they considered was “too long”. But human tradition of playing the game, clearly seems to have underestimated the machine’s exponential learning capacity.

After Lee Sedol loses to AlphaGo in the first match, people take notice. The film becomes a battle of humankind versus machine after Sedol loses three straight matches, opening a door into the deep culture associated with the game of Go and heightening the need for humans to prove themselves. Sedol seems to have lost hope, frustrated between not being able to interact with or read his opponent, as would be the norm playing with another human.

The competition becomes a metaphor for something much bigger. I found myself personally rooting for Team Human, and the film’s score really emphasized this emotional reaction. As European Go champion Fan Hui states early in the film, Go isn’t just a game. For players like Lee Sedol, “Go is a lifestyle”. Through the interviews of both Sedol and Fan Hui, you can see the clear amount of faith that they have put into this game. In many ways, it has defined their careers.

Especially for Sedol, it was clear that being on a world stage added stress to perform and beat AlphaGo not only for himself, or for his country, but for all of humanity and as a representative of human intelligence.

No pressure, right?

Who Actually Won?

When Sedol finally comes out victorious in the fourth game, the energy and excitement is palpable. While members of the DeepMind team try to figure out what mistakes AlphaGo made, it’s a great moment that makes you feel like humanity still has a fighting chance. Although AlphaGo goes on to win the fifth game, Sedol holds game four as an unforgettable victory, and a moment of redemption and hope for humans.

Now that AlphaGo has demonstrated the power of AI, the big question is: where do we go from here? Lately, questions and worries about regulation of AI have risen because of the notion that AI has the power to out-think an individual human brain, and the ability to compute on a much quicker level. On one extreme, some fear that these advancements will eventually lead us down the rabbit hole to a dystopian future with machine authoritarian powers.

To a degree, AlphaGo’s victory might be seen as highlighting humanity’s shortfalls when pitted against artificial intelligence. However, another perspective is to view this as a great scientific discovery: the power of humans when leveraging technology.

This point is underscored when lead programmer David Silver comments that AlphaGo’s victory is ultimately a human victory. It took multiple efforts and many scientific advancements for the DeepMind team to conceptualize and build from an idea that started twenty years prior to create AlphaGo. If anything, it goes to show that the power, and the limits of AI ultimately lies within the human mind.

Implications for Bot Training in the Future

The knowledge gained from AlphaGo’s feat will no longer be expended in board games; DeepMind has retired the algorithm and shifted focus to research scientific breakthroughs. There is no longer a question about what is more powerful, man or machine. It’s the combination which is the most powerful. For this reason, it is important that people become educated in this field and start experimenting with artificial intelligence in their daily lives. More important is how we apply these technologies. Can we improve healthcare, modify criminal justice, or solve global warming?—I believe when using artificial intelligence, the solutions are within our reach.

To Stream or Skip?

Definitely watch this film; the competition featured in the documentary is legendary, and a rare occasion in which we can see a scientific hypothesis play out in a real world scenario. There’s discovery and suspense, moments in which I personally got frustrated and felt a bit of fear; it covers the whole gambit of emotions.

I will say that the film could have presented more balanced viewpoints in the interviews during the big competition. Fan Hui’s input was valued as a Go professional, but seeing the opinions of the DeepMind programmers after each match would have offered greater insight to the two sides of the competition, and would drive home the discussion of the perceived man/machine dichotomy.

Overall, the film inspires humans to push ourselves further (I, for example, was inspired to learn how to play the game of Go), and it reminds us of the important question of the ethics of artificial intelligence. What it lacks in cinematic beauty, it more than makes up for in interesting perspectives.