The mouse navigates through a corridor in a natural experimental environment using its whiskers. Their brain activity is measured. © The Grainger College of Engineering at the University of Illinois Urbana-Champaign
Scientists were able to gain new insights into their decision-making through experiments on mice. The researchers hope that these insights can help in the development of more efficient AI systems.
New research from the Grainger College of Engineering at the University of Illinois Urbana-Champaign provides insights into brain processes that challenge traditional assumptions in neuroscience. The Working group led by Yurii Vlasov showed that decision-making processes do not, as previously assumed, only begin in higher-order brain areas that are responsible for complex processing. Instead, a fundamental, early developed brain region – the primary somatosensory cortex – was observed to be significantly involved in decision-making.
During the experiments, the mice were fixed on their heads in order to measure brain activity using electrode probes. They ran on a treadmill through an artificial passageway in complete darkness, using their whiskers to navigate. The tasks alternated between navigation in a straight aisle and environments that required left-right turns. This setup simulates natural conditions in which mice use their whiskers to make directional decisions in their burrows.
The scientists wanted to learn more about natural processes in decision-making that could serve as a model for more efficient artificial systems. Since current AI models perform poorly, especially in decision-making, research here could contribute to further development. Natural intelligence is more computationally powerful than current forms of artificial intelligence and requires significantly less energy, making it an attractive model for future AI.
“By studying the dynamics of neuronal activity, we may be able to better understand how they play a role in decision-making,” says Vlasov. “Perhaps this is an approach that uncovers previously unknown mechanisms – how these feedback loops are dynamically organized and how they shape and shape different levels of processing. Perhaps this can be integrated into new AI models.”