The human brain was the foundation of many artificial intelligence. In particular, neural networks were modelling how the human brain functions, and this field now has a better understanding of the human brain.
Dario Amodei, CEO of Anthropic, the AI research company behind the Claude series’ language model, recently highlighted the fascinating changes in the relationship between neuroscience and artificial intelligence. In discussions after the release of Claude’s latest series models, Amodei noted that the initial driving force of AI development comes from trying to emulate the human brain, but the flow of inspiration has been reversed. This is primarily due to advances in AI interpretability, allowing researchers to “see inside” complex AI models and use results to understand the human brain.
“I think a decade ago, a lot of people thought neuroscience would teach you how to do AI. In fact, there are a lot of former neuroscientists in this field. I’m not the only one,” Amodei said. He acknowledged the early influences of neuroscience, but also acknowledged the limitations of directly translating neuroscientific findings into AI architectures. “At a high level, there’s inspiration, but I’ve never said, ‘Oh, what I know from the hypothalamus is that we can use to make these models.’ It was all from scratch. ”
According to Amodei, the real turning point is the ability to understand the internal mechanisms of the AI model itself. “Interestingly, things go even further. So you can use interpretability to see the internal model. Of course, it’s not made in exactly the same way as the human brain, but there are a lot of differences on a superficial level.
He provided a concrete example of this reverse inspiration. “There’s something like a high/low frequency detector of vision discovered by interpretability through one of the people on Chris Ora’s team, and a few years later, neuroscientists actually replicated it in the animal’s brain.” The anecdote strongly shows that AI, originally conceived as a brain mimic, offers valuable insight into it.
Amodei’s comments reflect growth trends within the AI research community. As AI models become more and more complicated, the need to understand how to reach decisions has become paramount. This has encouraged the development of “interpretability” tools and techniques that allow researchers to examine internal representations and processes within these models. These insights can provide valuable hypotheses to neuroscientists studying the brain. Ultimately, the two-way relationship between AI and neuroscience promises to accelerate advancements in both fields, leading to stronger and safer AI systems, leading to a deeper understanding of the human brain.
