Some computational neuroscientists say self-supervised AI learning models have shown a closer correspondence to brain function than supervised-learning models
Context & Ripple Effects
The claim lands at the end of a decade-long search for learning algorithms that resemble biology. Earlier efforts tried hybrid routes — merging unsupervised language models like GPT-3 with labeled vision data to inject common sense, and neurosymbolic AI pairing deep networks with rule-based reasoning — but both kept supervision in the loop.
What changed is that purely label-free training started producing brain-like behavior on its own: researchers applied self-supervised learning to animal communication, and a later study found human brains and general-purpose neural networks process language sounds similarly. If computational neuroscientists now see closer brain correspondence in self-supervised than supervised models, the argument shifts from architecture to the training objective itself.
First-order effects
- Supervised-learning models lose their standing as the default analog for brain function in computational neuroscience, forcing researchers who model perception and language to justify labeled-data assumptions against a self-supervised baseline.
Second-order effects
- AI labs gain a scientific legitimacy argument for label-free pretraining beyond raw performance — and datasets that mimic natural learning conditions, like the infant headcam footage some researchers already train models on, become strategically valuable to both neuroscientists and machine-learning teams.
Third-order effects
- If the pattern holds, brain-model correspondence becomes an evaluation criterion alongside benchmarks, pushing the field toward training objectives that learn from unlabeled experience the way animals do — and making the supervised/self-supervised split a live question in how intelligence is modeled, not just engineered.
The trend: Neuroscience is becoming a yardstick for AI training methods, with self-supervised learning emerging as the first objective that both brains and machines appear to share.