AI research challenges long-held assumptions about covert attention
Covert attention may be an emergent property, researchers find.
Researchers at the University of California, Santa Barbara, have used AI to uncover new insights into covert attention, a cognitive process that allows people to shift focus without moving their eyes.
In a study published in the Proceedings of the National Academy of Sciences, the team showed that convolutional neural networks trained on visual detection tasks can reproduce key behavioural signatures of covert attention, even without built-in attention mechanisms.
By analysing millions of artificial neurons across multiple trained models, the researchers found that attentional behaviour can emerge from distributed neural activity rather than specialised brain modules.
The analysis revealed previously unreported neuron types, including cells whose activity is suppressed by visual cues and others that amplify signals at expected locations while dampening responses elsewhere.
Several of these neuron types were later confirmed using data from mouse brain studies, reinforcing the idea that AI models can help predict and explain real biological mechanisms underlying attention.
Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!
