Artificial Intelligence (noun): the capability of a machine to imitate intelligent human behaviour.
As a child, I owned an electronic chess board that was programmed to play against a human. That chess board housed a computer that calculated the best move against you, the opponent. Unknowingly, I was playing with a machine that was acting like human - albeit the fact that it was programmed to respond in that manner.
Computers have replaced humans far and wide. However, most of these machines only perform pre-programmed tasks. Basically, they don’t have the capacity to “think”, i.e. they aren’t “intelligent”. Both ‘thought' and ‘intelligence' are inherent features of us humans, and by no means do today’s computers compete with our capabilities… or really?
You must have read in recent news about AlphaGo, the artificial intelligence system that defeated World Champion Lee Sedol over a series of five Go matches. AlphaGo, that is designed by Google-owned DeepMind, won four out of the five matches and surprised the world with its triumph. ‘Go’, an ancient Chinese board game, is generally regarded as an impossible game for computers thanks to its exponential number of board configurations: more configurations than the number of atoms in the universe. AlphaGo has, however, proved this wrong.
The abilities of artificially intelligent systems are remarkable. At times, these systems can perform a task in a manner so different from human perception, that it would leave us flabbergasted. This was displayed in AlphaGo’s second encounter with Lee Sedol, where the AI played an unconventional move that stunned the world champion, who later said, “I am speechless.” That was the winning move for AlphaGo in that match.
What happened during that move is the result of deep neural networks. All AI systems work on the concept of neural networks, very similar to how human brains work. By feeding a computer with thousands of images of a particular object, it can be made to recognise the object. That is the basis of search engines and voice assistants like Google and Siri which use these neural networks to make sense of a particular situation. In fact, that is how Facebook recognises your face in a picture!
After understanding a situation, the AI must be able to take the best possible next step to respond to it. With a million different ways to approach a problem, it would be impossible to singularly analyse each solution through brute force. However, an AI system never does that - it responds according to what it has learnt from previous experiences. This bridges the gap between humans and computer AI systems.
The AlphaGo displays all these qualities. The move that I discussed earlier is a result of the multitude of games it played with both humans and other versions of itself. The AlphaGo consists of two deep neural networks - the ‘policy’ network and the ‘value’ network, as explained by David Silver, a member of the Google DeepMind team. When the AI is faced with a particular configuration of the board, the policy network determines the handful of best locations for the next move. The value network then determines which one of the handful might actually lead to a win - by calculating the chances of winning after predicting, say, the next 20 moves in the game.
Before the tournament with the World Champion, AlphaGo played a lot of Go. Not only did it learn from its creator’s Go skills, but also from other prominent Go players. Then it was pitted against a tweaked version of itself - a thousand games a day, more than any human could play within that timespan. This training made the AI more ‘experienced' with the moves and board configurations, giving it the intelligence and intuition required to win the tournament.
Of the five games played, four were won my AlphaGo. However, the only game Lee Sedol won had a ‘winning move’ too, a move that only human intuition could predict. Undoubtedly, Sedol’s move 78 in game four was the turning point in the tournament, proving that AlphaGo, too, could be stumped. This move probably surprised the computer, which eventually lost that game.
AlphaGo is just a small example of how artificial intelligence could parallel human intelligence. Complex AI systems in the future would greatly enhance human capabilities. Perhaps at a certain point in time, an AI system would be able to predict and judge a situation far better than a human mind. This would allow for development at an exponentially faster pace, leading to a revolution in the history of mankind.
At the same time, there is a risk. What if we manage to create an AI system that learns to multiply itself, by building newer, better versions of itself? That would allow machines to dominate humans - a war that humans would have no opportunity to win. Prof Stephen Hawking famously stated to BBC: “The development of full artificial intelligence could spell the end of the human race. It [AI] would take off on its own, and re-design itself at an ever increasing rate. Humans, who are limited by slow biological evolution, couldn’t compete, and would be superseded."
As of today, the risk of robots taking over earth is too small to stop us from building better artificial intelligence systems. AI systems have the capability to solve problems that humans face today, and are of tremendous value in the development of technology, medicine, design and everything beyond.
Whatever be the case, the race between human beings - products of a million years of evolution - and intelligent computer systems developed by us human beings has begun.
Follow Technonerds on Facebook & Twitter to catch all of my content.
Nature Video (YouTube)