The Power Problem Is the Next Major Breakthrough Point in AI
The artificial intelligence (AI) represented by the competition between AlphaGo and human go players brings great "frustration" to human beings. Now people have rationally realized that AI has unparalleled advantages over human beings in terms of search, calculation, storage, optimization, etc. However, AI cannot match human intelligence in perception, reasoning, induction and learning. In the world-focused man-computer war --AlphaGo VS Li Shishi, despite the results of human losing the games, it is noteworthy that AlphaGo actual used 1000 CPUs and 200 GPUs in the game,withelectricity bill up to 300 US dollars per minute,and its network size only 1/1000 of the human brain. In contrast, Li Shishi's "power consumption" should be only a few dozen watts, equivalent to an incandescent light bulb. If AI is to reach the human brain scale, it may require a small hydropower station to power one machine. According to Chen Yunji, a researcher at the Institute of Computing Technology at the Chinese Academy of Sciences, "If you want to use general processors to build a human's brain - scale neural networkof synapses, you may need to build a power station to power it." So it seems that AI will always surpass human beings in terms of computing, storage capabilities, but far less than the brain intelligence in power consumption. This gap also means that reducing power consumption will be a difficult point to break through in the future for AI. If AI is to be applied in large-scale, power consumption problems must be solved.