Renato Duarte’s analysis of citation patterns in neuroscience and artificial intelligence (AI) offers a fascinating lens through which to view the interaction between these fields! The observation that AI has primarily been loosely inspired by established, abstract concepts from neuroscience, rather than by using specific findings about neural mechanisms to motivate design choices, matches my own experience. That said, many of the design choices present in the current generation of AI tools parallel purported properties of the brain, such as fast activity-based mechanisms for reasoning, geometric codes that expose relative positions, and planning through mental simulation. (Chris Summerfieldand I describe these in apaperpublished earlier this month.)
My favorite plot contrasted the absolutenumberof AI papers citing neuroscience with thefractionof AI papers doing so post-2015. To me, this plot captured the essence of a growing question for people working at the NeuroAI interface about where the most opportunities lie. To recap, at that point, thenumberof cross-field citations began to climb steeply for both the neuro-citing-AI and AI-citing-neuro papers. The fraction of neuro-citing-AI papers also increased; however, the fraction of AI-citing-neuro papers remained low, perhaps dipping.
So is neuroscience having an increasing or decreasing impact on AI? On the one hand, the declining fraction of neuro citations suggests it was not a dominant driver of the AI boom compared with other factors. On the other hand, the rising absolute number of neuro-inspired AI papers still represents a substantial footprint, even if it constitutes a shrinking percentage of the overall AI literature. For researchers who work at the interface of neuroscience and AI and are questioning how to spend their time now, in the most recent post-2022 AI boom, this is the core of the question. Maybe AI is such a phenomenon that even a fringe possibility of influencing AI is a competitive strategy for maximal impact, or maybe the behemoth AI field is on a hard-to-shift path, and lower-hanging fruit can be found in applying AI to the hard problem of understanding the brain. At any rate, I appreciated Duarte’s call for new results and tools to refresh this interdisciplinary relationship.
Read Renato Duarte’s piece:Examining the bridge between neuroscience and AI
Read Tony Zador’s comment:Citation graphs may underestimate neuroscience’s influence on AI
