AI, toy or disaster?
AlphaGo beat KE JIE who was the best Go player in the world in May 2017. This is a symbol that computer can beat human in the most complex games which are played by human. So the problem comes, if one day in the future, will the computer or robots rule human? Intelligent systems now provide lots of convenient for us, but when their intelligence is far beyond human, will they enslave human?
To analyze this problem, we should know the mechanisms of intelligent systems like AlphaGo. To build a AlphaGo, we need three elements: Algorithms, Machine and Data. Let me describe these three elements in detail.
Algorithm: The model used by AlphaGo is called Deep Learning or Deep Neural Network. Neural network means this model is quite similar with the brain structure in our brain. This model can find and model very complex patterns. However is should trained with data and the more neurons it has, the more powerful it is.
Machine is the physical fundament of the system, it should be powerful, cheap and fast. The first neural network has been invented about 30 years ago, but it is not powerful at all is because the power of machines at that time is very low. It is like that, human and other animals all have brains. The reason human is smarter than them is because we have heavier brain and more neurons in our brain. So it is in the intelligence systems.
Data: To train a model, we need huge number of labeled data. Labeled data means manually label the dataset, and tell which category every piece of data belongs to. It is like we teach a child which image is bird and which is a tree. AlphaGo found the pattern of Go in hundreds of thousands of competition information before became a master in Go. Although there are some algorithms now can learn by themselves, but the process is quite slow and with low efficiency. So the limit is, if we want a powerful intelligent model, we should provide enough labeled data. However, the process of labelling is slow and expensive.
Although AlphaGo beat KE JIE, we have to admit that the intelligence system is still in the early childhood. These systems look very smart; however they can only be used in scenario. What’s more, they have no cognitive ability. So, if machine want to rule human, they have very long way to go.
In the future? Maybe…
Maybe with the development of technology, these limits are not limits any more. Toys may become disaster.