Focus on Chongqing Science and Technology Innovation Conference | Data and knowledge drive them to "make artificial intelligence more human-like"
2024-08-21
한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina
When ChatGPT was first released, it ignited the world's generative artificial intelligence technology, but it was also criticized by many people: it often "talks nonsense in a serious manner". When Sora was widely praised by the Internet, some people jumped out to "pick faults": the "creative" and "realistic" videos it generated sometimes had defects that did not conform to common sense, such as when a person ran forward, his body movements were reversed...
"Artificial intelligence technology has developed by leaps and bounds, but there are still many problems that need to be solved. We are engaged in the construction of multi-granular knowledge space, which is aimed at the basic scientific problem of uncertainty knowledge processing in the field of artificial intelligence." Wang Guoyin, president of Chongqing Normal University, said. In the just announced 2023 Chongqing Science and Technology Award, the project "Model Theory and Method for the Construction of Multi-granular Knowledge Space" led by him won the first prize of Chongqing Natural Science Award.
▲ Wang Guoyin gave an academic report. Photo provided by the interviewee
He introduced that AI is usually data-driven, which means obtaining information and knowledge from massive amounts of data to solve problems. The problem is that it processes data in a mathematical and statistical way, which may not be consistent with human knowledge logic, and the results may be biased.
"We want to break the single data-driven approach and integrate human knowledge logic to drive the construction of artificial intelligence systems, making artificial intelligence more intelligent, or making artificial intelligence more like human intelligence." Wang Guoyin said that they call this approach data and knowledge bidirectional drive. Based on this, they created a multi-granularity cognitive computing model.
According to reports, in the field of information science, scientists visualize information data and use granularity to express the relative size or coarseness of information units, just like we use pixels to describe the quality of a photo. Multi-granularity cognitive computing was first proposed internationally by Wang Guoyin's team in 2017. It is a cognitive computing method that combines the cognitive laws of the human brain with existing artificial intelligence computing methods.
"In fact, humans perceive the world at different levels of granularity." He used face recognition as an example to explain that if an old friend of yours walks towards you from a distance, the artificial intelligence system will identify him through simple facial feature data, such as the distance between the eyes, the shape of the earlobes, the proportion of the facial features, etc., to confirm whether he is your friend. But before you can see the facial features of this person clearly, you can often recognize him through comprehensive information such as posture, movement, and temperament. This is multi-granular cognitive computing. "As we often say, 'seeing the forest for the trees', we can see the forest as a whole, as well as the trunk, branches, and leaves of a tree."
In his opinion, it is precisely under the two-way drive of data and knowledge that artificial intelligence not only meets the information calculation and processing mechanism of computer systems, but also meets the human cognitive mechanism of "large-scale priority" and from macro to micro, and will become more human-like in the future.
It is understood that the results of this project have not only been positively cited by well-known scholars in more than 30 countries and regions, but have also been successfully used to solve key technical problems in the fields of national cyberspace security governance, process industry control and smart health.
"Our exploration has only achieved phased results." Wang Guoyin said that at present, his team is continuing to carry out theoretical and methodological research on the one hand, and actively carrying out applied research on the other hand, so that these results can be more widely used in different industry fields.
