Working memory is our ability to keep a limited number of memory objects in our consciousness at the same time and manipulate these. It is one of the most basic cognitive processes and is behind widely different things such as planning, reading comprehension and solving mathematical problems. Therefore, it is by nature flexible and can be applied to a variety of abstract or concrete memory objects. It is therefore one of the most studied cognitive phenomena and is strongly correlated with intelligence and mental flexibility.

At the same time, working memory impairments are a central part of many cognitive disorders such as schizophrenia, Alzheimer's and ADHD. However, despite a large amount of research and progress, we still do not know how processes on a neural level can lead to the flexibility of working memory.

The brain is a complex and tightly reconnected system. It consists of billions of neurons that are interconnected in different directions, rather than a system constructed by humans where the flow of information can be followed and understood by looking at the connections. To understand such systems, simulations in computer models are of great help and can provide vital insights. It is possible to train artificial neural networks to hold and manipulate information in seemingly similar ways as in the human brain.

Mikael Lundqvist
Mikael Lundqvist

"A crucial difference, however, is that these artificial systems can only perform the task on objects that they have trained on, while humans and other higher primates (but not mice, for example) can learn a task and then flexibly apply it in other situations or to other objects," says Mikael Lundqvist, researcher at the Department of Psychology, Stockholm University.

He will now study this crucial flexibility within the research project The Hot-Coal hypothesis of working memory (Hot-Coal WM) when he is awarded an ERC Starting Grant of EUR 1.5 million.

Within the project, he will propose a new mathematical model, the Hot-Coal model, for how neural networks could implement flexibility. This model will be developed in artificial networks and at the same time tested in brain activity collected from both animal models and human research subjects at the new brain imaging centre at Stockholm University (SUBIC). Experiment and theory will be alternated to exploit the benefits of both approaches.

If the project is successful, it could entail enormous advances in understanding our mental flexibility, how it breaks down into cognitive disorders and how it can be applied in artificial systems to create a flexible, artificial intelligence.

More information on

Read more about the Lundqvist lab