Some chat about AI

AI has been a hot topic for a while and there have been some well known firms and individuals participating in this sector. DeepMind and OpenAI may be the two best known AI firms; one developed AlphaGo and one of the other's founders was Elon Musk. Both companies are developing AI which can play video games that DeepMind is focusing on Starcraft II and OpenAI is focusing on Dota 2. The two firms definitely considered their choices very carefully that I believe the reason for choosing Starcraft II was to design an AI that can work on several tasks under uncertainty and the reason for choosing Dota 2 was to design an AI that can cooperate in a complex environment. If they can succeed (they are both now close to success), it will be a big step in the AI sector.

Based on what we have seen about AI performances (such as AlphaGo), AI often shows its difference from our human beings. Sometimes we feel our intelligence is challenged by AI, for example the Go strategies which have been studied for thousands of years are definitely challenged by AlphaGo. Some people think that we may be able to learn from AI and through learning, we can improve our thinking further. This is definitely a reasonable argument. However, the difficulty is that machine learning does not really tell us what is really in AI's mind. Furthermore, because of the speed of AI's learning process, we are not possible to monitor every stage of its improvement. The most possible and straight way is to study the final outcome and try to draw some reasonable conclusions.

Then such study becomes like empirical studies that we do not exactly what is in the process and try to guess the process by using statistical tools and other tools. The study on AI is like this but with a relatively simple system.