Cognitive Map-The Evolution of Artificial Intelligence from Sensory to Cognition

On December 16, 2020, the “MEET 2021 Smart Future Conference” was held in Beijing. Professor Tang Jie of the Department of Computer Science at Tsinghua University gave a speech entitled “Cognitive Map, the Next Treasure of Artificial Intelligence” at the meeting. In his speech, Professor Tang first briefly introduced the three eras of artificial intelligence: symbolic intelligence-perceptual intelligence-cognitive intelligence.

Artificial Intelligence-Cognitive AI

There have been three waves in the development of artificial intelligence. We call artificial intelligence the three eras, namely symbolic AI, perceptual AI, and cognitive AI. Cognitive AI has not been realized until now. So, what needs to be done to realize cognitive AI? Professor Tang Jie pointed out that what is urgently needed to build is AI infrastructure, such as how to build a cognitive map.

Why emphasize the logic of cognition? Professor Tang Jie gave the example of GPT-3. GPT-3 has a parameter scale of 175 billion, an order of magnitude close to human neurons. And, in terms of expressive ability, such a large-scale language model has brought AI closer to humans. This brings an enlightenment: Can we directly obtain a general artificial intelligence that surpasses humans through large-scale, large-scale, large-scale calculation methods? Not to mention the training cost of hundreds of millions of RMB, a language model like GPT-3 has now exposed an obvious problem: there is no common sense. For example, if you ask it how many eyes a piece of grass has, it will answer “one eye”.
To solve such problems and use computational methods for cognition, Professor Tang Jie talked about combining two methods to achieve this: First, data-driven. Model all data, learn the relationship between the data, and learn the memory model of the data. Second, knowledge-driven, to build a knowledge graph.

The field of cognitive intelligence is still in its infancy

Artificial intelligence has reached or surpassed human standards in the field of perceptual intelligence such as “listening, speaking, and seeing”, but it is still in its infancy in the field of…