Knut and Alice Wallenberg Foundation has awarded SEK 28 million to a five-year project that will use artificial intelligence to determine whether brain activity reveals which odour a person is smelling.
The project is led by Danica Kragic, Professor of Computer Science at KTH Royal Institute of Technology, with Johan Lundström, Professor of Psychology at the Department of Clinical Neuroscience , as co-applicant. SEK 14 million of the grant goes to Karolinska Institutet. The project starts 2027.
Brain activity can already be used to decode what a person sees or hears, but smell has lagged far behind. A single odour molecule activates many types of receptors, the brain represents odours as dispersed patterns rather than orderly maps, and the same odour is perceived differently from one person to the next.
The project will adapt transformer models, the type of AI behind today's language and image models, to smell. The models will be trained not only on the chemical structure of odour molecules but also on how people rate odours and how their brains respond to them. By comparing the models' internal representations with human perception and brain activity, the researchers will test how far an artificial system can develop a human-like understanding of smell.
"Smell is still the sense we understand least. We do not know how the brain turns a mixture of molecules into the experience of coffee or a rose. AI lets us find patterns in brain signals that we would never find on our own," says Johan Lundström .
