The hidden computational role of dopamine in the brain
Neuromodulators in the brain shape our ability to learn much better than previously thought
Our brains may be running on neural networks similar to that of powerful artificial intelligence, enabling dopamine to trigger much more complex and sophisticated learning than previously thought.
This is the conclusion from a study by Emil Wärnberg, a joint doctoral student at KTH and Karolinska Institutet, and Arvind Kumar, an associate professor at KTH.
The study was published in Proceedings of the National Academy of Sciences and challenges the current understanding of how dopamine shapes our ability to learn.
Wärnberg and Kumar believe the findings can help the treatment of brain and mental disorders such as Parkinson’s disease, ADHD and obsessive-compulsive disorder.
Faster and more detailed learning
The popular view is that dopamine drives learning in the brain through bursts or “kicks” of dopamine that more or less uniformly flood the brain.
This is too simplistic, according to the authors. The study presents a more powerful way for dopamine to send feedback to brain cells when we are learning a new task.
Wärnberg and Kumar demonstrate how the cells that release dopamine can create a microscopic landscape with some patches getting more and some patches less dopamine.
“Dopamine acts a bit like the waves in the ocean," Wärnberg explains.
“This way dopamine cells can send much more detailed information to the other brain cells than what is previously known, thus allowing for faster learning.”
Insights from artificial intelligence
The study shows how complicated but coordinated learning can occur in our brains. Deep artificial neural networks, which are behind the recent explosion in artificial intelligence capabilities, are also made of simulated brain cells but are trained with an algorithm called “backpropagation”.
Although incredibly powerful, backpropagation has been thought to be highly artificial and biologically implausible. However, Wärnberg and Kumar show that the dopamine landscape can be used to run a version of ‘backpropagation’ while respecting everything we know about how the brain is built.
“It is very exciting when we find these types of connections between deep neural networks and real brains,” says Wärnberg.
“This will help us understand both how the brain works and at the same time better understand how to build better artificial networks.”
By combining insights from artificial intelligence with a simulation of parts of the brain, researchers can discover new ways to understand how neuromodulators such as dopamine work.
“The fact that dopamine is involved in all these very different disorders demonstrates that it plays a key role in how the brain as a whole functions,” Kumar says.
Read the study "Feasibility of dopamine as a vector-valued feedback signal in the basal ganglia"
For more information, contact:
Text: Sturle Hauge Simonsen
Related news
The hidden computational role of dopamine in the brain
Neuromodulators in the brain shape our ability to learn much better than previously thought
Read the article