How the Brain Learns
Five questions with Roozbeh Kiani
When we think about learning, we often conjure up images of sitting in a classroom, nose buried in a textbook. But is that what neuroscientists study when they investigate learning?
A good portion or maybe even the majority of what we do in our daily lives is not based on knowledge that is acquired in a classroom. Most of the learning that we do, the learning that shapes our behavior, we acquire by living our life and interacting with the environment around us. The associations we make, habits we build, experiences we recall, and skills we acquire—like riding a bike or playing the piano or being able to distinguish one face from another—these are all types of learning. For these fundamental types of learning, we know a fair amount about the underlying neural mechanisms.
Now, when we consider forms of learning that require direct acquisition of knowledge—like how we learn physics or literature or math in a classroom—these higher levels of learning are significantly less understood. It would be rewarding if what we discover about the mechanisms of learning put us in a position to have better curricula and better learning environments.
What is our current level of understanding about how learning takes place in the brain?
The human brain has about a hundred billion neurons. And the number of connections each neuron makes ranges from hundreds to maybe 10,000. So we are talking about networks that are very very vast. These large networks can be broken down into circuits that interact with each other to communicate and store information. Learning is associated with changes in the neural responses within these circuits in different parts of the brain. To change the response of the circuit, you have to modify the strength of the connections—or synapses—between one neuron and another. For many types of learning, for example the acquisition of some skill, new synapses may form or existing synapses may become stronger (or in some cases weaker). As these synapses are modified through learning, the knowledge is ingrained in the brain and influences the future operations of the circuit. We’re now beginning to understand how modification of the activity of one circuit during the learning process can influence other circuits in the brain. The more we discover, the more we realize how much we still have to learn!
You have been studying monkeys as they learn to visually differentiate between different shapes. What has your work revealed about how circuits operate in tandem as an animal acquires a new skill?
We track the responses of neurons in different parts of the brain throughout the learning process to find out where in the brain the learning takes place—and how. What we find is that neurons in the planning areas of the brain—like the frontal cortex—are the first to change their responses in a way that reflects the time course of learning. The information is then transmitted, with a little bit of lag, back to the visual cortex, the area dedicated to processing patterns of light that enter the eyes. So the brain is using neural connections within the visual cortex as a longer-term memory system for the acquired knowledge.
It’s as if, when the training begins, the animal has to actively think about what it’s learning. But as the animal practices more and more, that knowledge is sent to the visual cortex, which is involved in the earlier stages of information processing. When the knowledge is stored there, it becomes a more automatic form of behavior for the animal. So the animal doesn’t have to push itself to see the differences in shapes—the differences just emerge, the same way that a parent can distinguish between identical twins without having to pay a lot of attention to their faces.
A similar thing happens when people are learning to drive. During the initial learning process, drivers are very consciously and actively involved in the task. But as they practice more and more, the information is transferred to regions in the brain that enable automation of the task, so we can have a conversation and drive without thinking about each of our actions.
Do these same processes take place as babies learn?
We have a toddler at home and this past year and a half has been educational for me observing our son learn. As an adult, it’s easy to forget what it actually takes to develop even the most trivial behavior, for example the brain’s motor program required for walking. But then you observe a toddler try to stand up and walk. Our son is at the stage where everything is very conscious and effortful and full of mistakes. He falls all the time! But in a few months or a year from now he will be walking around without even thinking about it. This is how the brain works and it’s amazing to watch. Moving forward there’s going to be a strong motivation for me to design new experiments to pursue lessons I’m learning from our son as he grows.
Over the past 20 years, researchers have developed increasingly sophisticated methods for manipulating the activity of individual neurons. How are these techniques being used to probe the mechanisms of learning and memory?
Some labs are using these techniques to induce new memories or to tamper with existing memories in mice. For example, if you condition mice to associate a certain sound with a mild foot shock, certain circuits of neurons will modify their response properties. Now if you identify those neurons and change their response properties, you can potentially erase the memory that has been stored there. This is a very exciting direction for research because if it could be expanded into a practical tool, that could be used to help patients suffering from conditions like post-traumatic stress disorder. We are not at the stage where we can replace memories in people. But if we continue to refine the tools that we have, maybe at some point it will lead to artificial management of memory in a way that could help patients.