Exciting new research from the MRC Brain Network Dynamics Unit and Oxford University’s Department of Computer Science has identified a novel means for comparing learning in AI systems and the human brain. By addressing a fundamental issue in both human and machine learning – credit assignment – the team has uncovered a key difference between the two.
AI systems approach this through backpropagation, adjusting parameters to correct errors in output. This works like a feedback loop, tracing back through the network’s layers to identify which parts of the computation contributed to the error and then refining the AI’s decision-making for future predictions.
The study, published in Nature Neuroscience, explains how backpropagation differs significantly from the human brain’s learning method, known as ‘prospective configuration’. This method predicts the ideal pattern of neural activity resulting from learning first, before changes to the neural connections occur. Crucially, this approach offers a more efficient learning mechanism than backpropagation, as humans can rapidly ingest new information without eroding existing knowledge, a skill AI struggles to match.
To illustrate this concept, the team uses an analogy of a bear fishing for salmon. An AI model would incorrectly assume the absence of salmon if the bear suddenly can’t hear the river due to a damaged ear. In contrast, the animal’s brain, operating on prospective configuration, would still rely on the smell to deduce the salmon’s presence.
Computer simulations demonstrate that models using prospective configuration outperform traditional AI neural networks in learning efficiency. Professor Rafal Bogacz, the lead researcher from MRC Brain Network Dynamics Unit and Oxford’s Nuffield Department of Clinical Neurosciences, describes the study as a “big gap between abstract models performing prospective configuration, and our detailed knowledge of anatomy of brain networks.”
The development of bio-inspired AI, or neuromorphic AI, could bridge this gap and lead us toward machines with a degree of autonomy and environmental interaction. Innovations in this field include low-powered chips modeled on synaptic functions (backed by OpenAI CEO Sam Altman last year) and swarm intelligence which mimics the collective decision-making of groups of insects, birds, and fish.
Though there are still fundamental questions for AI to answer before it can match biological brains, exciting research like that from Oxford University is pushing us closer to a future where machines have a greater degree of independence and environmental interaction. There’s still life in the old human brain yet!
