Why current AI fails to learn with the efficiency of a biological brain
In short: Artificial intelligence requires hundreds of repetitions to master a single concept, yet humans can learn from a single glance. This gap suggests that our machines are missing a fundamental principle of intelligence: the way the biological brain balances new information with existing knowledge.
Most modern AI relies on backpropagation, an external algorithm that adjusts parameters to reduce output errors. However, this method often suffers from interference; learning new information can rapidly degrade previously held knowledge. In contrast, researchers from Oxford University and the MRC Brain Network Dynamics Unit have identified a different mechanism called 'prospective configuration.' Instead of immediate adjustment, the brain first settles neuron activity into an optimal, balanced state before modifying synaptic connections. This process preserves existing knowledge, allowing for faster and more robust learning.
The practical difference is stark. Consider a bear fishing for salmon: if the bear loses its hearing due to injury, an artificial neural network might incorrectly conclude the salmon are gone because the lack of sound disrupts the entire learned pattern of the river. A biological brain, using prospective configuration, maintains the connection between the smell of salmon and the river, despite the missing auditory input. This ability to handle sensory changes without losing context is a key advantage of biological systems.
The field is now moving toward creating 'digital twins' of neural structures. Researchers like Andreas Tolias have already used foundation models—similar to those powering ChatGPT—to create a digital model of a mouse's visual cortex. By training on brain activity recorded while mice watched movies, they created a model capable of predicting how neurons respond to new visual inputs. As scholars like Dan Yamins explore, these simulations could eventually bridge the gap between abstract mathematical models and the complex anatomy of the human brain.
Opens the Intellect+ feed with this card first. Swipe up for the next one.
Sources
- Summary by Intellect · source: Should AI Research Try to Model the Human Brain?
- Also: Study shows that the way the brain learns is different from the way that artificial intelligence systems learn
- Also: Building AI simulations of the human brain
- Also: pmc.ncbi.nlm.nih.gov — Artificial Intelligence and Neuroscience: Transformative Synergies in Brain Research and Clinical Applications
To cite this card, link to https://intellectplus.co.uk/app/technology-should-ai-research-try-to-model-the-human-brain.
How we source, check and correct cards: Editorial standards.
General information, not professional advice. Spotted an error? Flag the card or see corrections and complaints.
Intellect Plus is a visual educational discovery platform for exploring fascinating facts and ideas. Brain Food, Not Brain Rot. What is Intellect Plus?
