July 28 - According to Xinhua News Agency, Nobel Prize winner and godfather of artificial intelligence Geoffrey Hinton In 2025 World Artificial Intelligence ConferenceThe lecture, titled "Will Digital Intelligence Replace Biological Intelligence", is also Hinton's first lecture in China.
In his talk, Hinton reviews two paradigms that have characterized the development of AI over the past 60 years.One is the logical type, arguing that the essence of intelligence is reasoning, which can be achieved by manipulating symbolic representations with rules;Another biologically based, believes that the key to intelligence lies in the connectivity of neural networks.
Hinton said that according to symbolic AI theory, sentences can be transformed into precise logical expressions, but he argued that this is not how people actually understand sentences. He used Legos as a metaphor to illustrate the dynamic combination of words:
The ability to build any shape with Legos in thousands of combinational dimensions is essentially a form of modeling, as is language.
It is worth mentioning that Hinton said in his speech that there is a difference between hard and soft computer science, and even if the hardware is destroyed, the knowledge in the software will always exist and be easy to share. The efficiency of human knowledge dissemination is low, for example, about 100 bits of information can be transmitted through speech per second. If energy is cheap, the advantage of digital computing is more obvious.
"This raises deep concerns for me. Humans are so accustomed to existing as the top intelligence that it's hard to imagine being dominated by a higher intelligence. It's like keeping a tiger cub as a pet; we have to make sure it doesn't turn on us as an adult."Hinton said the value of AI in areas such as healthcare, education, and climate change is irreplaceable, and that it is unrealistic to unilaterally ban AI, suggesting that countries develop AI technology on their own but share the results of research that 'keeps AI benign'.
Hinton said.When training a super AI, how to make it both super intelligent and willing to assist humans is a major issue that deserves global cooperation.