In March 2023 a group of 27 neuroscientists and AI researchers, led by Anthony Zador and including Yoshua Bengio, Yann LeCun, James DiCarlo, Jeff Hawkins, Terrence Sejnowski, and Doris Tsao, published “Catalyzing next-generation Artificial Intelligence through NeuroAI” in Nature Communications (Vol. 14, article 1597).
The piece is a position paper for the field its authors call NeuroAI: research at the intersection of neuroscience and artificial intelligence. Its argument is that today’s AI excels at tasks like language and image generation that humans find effortful, yet remains far behind even simple animals at the basic sensorimotor skills of moving through and interacting with a physical world. The authors contend that those animal capabilities, refined over hundreds of millions of years of evolution, encode priors that modern AI lacks and that neuroscience can help recover.
To make the goal concrete, the paper proposes an embodied Turing test: rather than judging a machine by conversation, judge an artificial agent by whether it can interact with the sensorimotor world at the level of a real animal such as a mouse or a bird. The authors call for sustained investment in understanding the neural basis of these abilities and importing the principles into AI.
For a general reader, this paper is a useful snapshot of a serious argument among leading scientists that the path to more capable, robust AI may run back through the biology of brains rather than away from it.