Now a machine can beat a human at Go, what next?
AlphaGo’s recent defeat of world Go champion Lee Sedol by four games to one (the image here shows the board at the end of the only match that Sedol was able to beat the Google creation) raises a number of questions about where machine learning is taking us.
The first is pretty obvious: beating expert humans at chess, Jeopardy, or the old Atari games is undoubtedly amusing, and raises the profile of a discipline that tends to be overlooked. The AlphaGo win makes it very clear that robots are soon going to be beating us at just about any task you can imagine, whether it’s working or playing, but that’s not all. We’re going to see robots making big business decisions, such as fixing the European Central Bank’s interest rates, tax policies, or how much of a pension we receive; they may even become our sex partners — but hopefully not all at the same time, however much the idea might turn us on :-)
The real challenge here is not designing and programming robots to beat us at things, but instead to work with us. The idea of us all being replaced by robots is disturbing to most people, and would require a total redesign of society without precedent. We are moving toward a society where robots will play a bigger role than ever, there’s no doubt about that: the issue here is how it happens and what comes later. Those are questions that will take time to answer. What kinds of society we live in, wealth distribution, the role of humans, the development of society, these are the key questions. But the simple truth is that technological development cannot be stopped.
Second question: the AlphaGo win has given machine learning a big impetus following the long AI Winter that the discipline suffered after the 1970s. We’re not just talking about million dollar prizes to promote sciences such as STEM, but something that has captured the public’s imagination, with greater numbers of people deciding to dedicate their careers to AI. These are the factors that will speed up the development of machine learning, even if they are, in reality, the tip of a very large iceberg. Either way, the future is here, and it’s arrived much more quickly than anybody imagined.
Thirdly, the series of Go matches between AlphaGo and Fan Hui and Lee Sedol have all been analyzed microscopically, producing many interesting results. The first is that the machine is much better at learning from experience than humans are. In the same way that a self-driving car that covers one meter of road shares that experience with every other self-driving car on the planet, the feedback from AlphaGo will help create a faster, better “player”. Some of AlphaGo’s moves have been described not just as “beautiful”,but that have a one in ten thousand chance of being played by a human, moves that no human could understand, and much less anticipate. The combination of deep learning with reinforcement learning, pitting a machine against itself and thus creating new moves from others and that are then fed back into the system have been brilliantly demonstrated.
Looking at the question from another perspective, given that the only match Lee Sedol won was the fourth out of the five, and when he stood no chance of winning overall, can we talk about artificial intelligence with the ability to be emotionally intelligent? That’s to say, can we postulate that the machine allowed Sedol to win so as not to completely humiliate him? What would such a scenario suggest for the future of man and machine? Can machines be condescending? And would we want them to be?
I wrote an article a while back advising people not to tweet when drunk: now somebody has created an algorithm that can detect with a tweet has been written by somebody under the influence. And what about avoiding the many deaths each year caused by drunk drivers. How should machines deal with these kinds of issues? Can a machine related to a human in a human way? I can think of any number of games or tasks in which the objective of a machine is not necessarily to win.
Needless to say, winning isn’t everything. If the aim is therefore to improve cooperation between humans and machines, we could use this combination to resolve many of the most difficult problems the world faces.
Meanwhile, we’re still playing at something that long ago ceased to be a game. Anybody fancy a round of poker?
(En español, aquí)
