What Is Computational Psychiatry?

We often hear that the brain can be thought of as a type of computer. But how seriously should we take this metaphor? And might it tell us anything useful about psychiatric disorders?

It’s not hard to see how the brain-as-computer metaphor can be enticing. The brain gets input from the senses and a computer gets input from a keyboard. For both brains and computers, these inputs are turned into electrical signals that carry information into the system and affect its inner workings. To work correctly, the signals in both cases need to be interpreted correctly. For example, it wouldn’t be very good if my computer interpreted the “enter” button to mean “delete,” and it wouldn’t be good if my brain interpreted a happy face to mean a person was angry. Both brains and computers also need to store information in memory appropriately and select some outputs over others. Selecting the right outputs, such as choosing the right thing to pay attention to or the right action to accomplish a goal — or in the computer’s case, showing the right thing on the screen, running the right program, or printing the right document — are also crucial for proper function.

When thinking about psychiatric disorders from this perspective, we also know that modern computers can malfunction for multiple reasons. One reason is hardware damage: disconnected wires, a cracked screen, a broken hard drive, and so forth. Another reason is problems with software, such as poorly written or corrupted program code, too many programs running at the same time, or malware. Visible brain damage, such as with brain tumors, head injury, or Parkinson’s disease, could be thought of as hardware damage. In turn, some have suggested that psychiatric disorders might be thought of as software issues. For example, perhaps unfortunate early life experiences — such as abuse or neglect — could be thought of as “programming” the brain to interpret, store, or act on information in unhelpful ways. In this metaphor, a part of psychotherapy could then be to adjust the “code” in these programs to improve functioning.

Of course, the brain is also not like modern computers in many other ways. The laptop currently in front of me mainly just responds to input. It’s not capable of the same types of independent internal thought processes as a brain, it doesn’t have anything like human emotions, and it doesn’t need to control a body or seek out its own energy sources. The distinction between hardware and software is also not as clear in the brain. For example, even though emotional disorders don't typically involve brain damage visible on an MRI scan, the "software" differences associated with depression or anxiety symptoms are still encoded by patterns of physical connections in the brain; they are just too microscopic to see. Progress in robotics may improve our understanding of some of these differences — but it’s clear that the kinds of computational problems the brain needs to solve are quite different than those of standard computers, and it needs to operate in different ways to do so. Nonetheless, new approaches in psychiatry research suggest this perspective could be valuable for understanding how problems with mental health could arise and be maintained.

Computational Psychiatry

Over the past several years, a new field of research has emerged called computational psychiatry, which takes this perspective seriously. By thinking about the computations necessary for things like perception, learning, and action, it suggests ways in which brain networks (the “hardware”) can give rise to psychological processes (the “software”). (For more academic introductions, see [1, 2]). It might also help identify new factors that contribute to mental illness and, possibly, the development of new treatments. In brief, computational psychiatry studies how the brain processes information. It does this by simulating these processes on computers with mathematical models. Using these tools, it can describe how brain processes carry out the computations needed to perform cognitive and emotional functions, which provides a direct link between biology and psychology. For example, by better characterizing specific psychological functions, such as how we make social decisions — and how we perceive and interpret their outcomes — it can help explain why some individuals have more trouble in acquiring social support (e.g., maintaining close relationships) and how this influences the brain and body. Crucially, it offers a new window into how the brain’s computations may go awry in mental illness.

To give you a better idea of how this works, I will provide a more specific example below. It illustrates one type of information processing that has been studied and how this may help us better understand psychiatric disorders.

Perceptual processes as a concrete example:

It’s not uncommon for two people to witness the same event and yet perceive/interpret that event in different ways. This can happen based on differences in what a person pays attention to and in what they expected might happen beforehand. In emotional disorders, this may contribute to various perceptual biases. For example, individuals with anxiety might perceive things as being more threatening, or individuals with depression may interpret something more pessimistically. Neuroscience research suggests this is because, in perception, the brain combines prior expectations with sensory signals. Basically, if the brain trusts its expectations more than it trusts the new sensory signals, then the perception of those signals (i.e., the interpretation of their meaning) will be biased toward those expectations. For example, if you strongly expect that a person does not like you, and then they make a slightly negative facial expression, you’re likely to perceive their face as unfriendly. In contrast, if you think they are generally friendly, you may simply perceive them to be tired or think they are feeling bad for some other reason.

In computational psychiatry, this can be simulated with equations involving probabilities. For example, you might (perhaps unconsciously) believe the probability that someone doesn’t like you is p = .8 (i.e., an 80% chance). And perhaps the slightly negative face itself (i.e., the sensory signal) indicates a probability of p = .6 that they don’t like you. If you combine these probabilities mathematically, the result is that there is an 86% chance that the face means they don’t like you. In contrast, if you think beforehand that there is a probability of p = .3 that they don’t like you, then, if you do the math, the probability that they don’t like you after seeing the face is only p = .39 (i.e., there is a higher [61%] chance they do like you). If so, you may perceive the face as less negative, or perhaps assume they are feeling bad for some other reason. The idea here is that the brain is performing these computations unconsciously, and what you perceive is the interpretation your brain assigns as the highest probability. So, in the first case, you’d perceive that they are unfriendly and don’t like you, while in the second case you wouldn’t. This way of mathematically modeling perception has helped us understand several clinically relevant phenomena.

As one example, recent studies have suggested that individuals with psychosis are more likely to hear tones simply because they learn to expect them, even if no tones are played. This suggests their brains may unconsciously trust prior expectations too much — which could lead to hallucinations [3]. Studies in our lab have also shown that, in individuals with emotional disorders and eating and substance use disorders, the brain may not treat signals from the heart as though they are trustworthy when trying to count their heartbeats [4]. Again, this “lack of trust” is not conscious. But it means they often do not feel a heartbeat when it happens, and they may also think they feel a heartbeat at times when it doesn’t happen. With these kinds of studies, we can identify how such biases differ from person to person. If these biases are causing problems, we might also start thinking of ways to help them (or, rather, help their unconscious brain processes) “trust” the signals from their body a bit more and improve the brain-body interactions that are important for emotional functioning.

Summary

This is just one example of how understanding computational processes in the brain, and how they are affected in different psychiatric disorders, might advance our understanding of these disorders and perhaps point the way to more specific treatments. Many other examples could be given that allow scientists to test, for example, how quickly people learn from rewards and punishments, how much they value gaining information to reduce uncertainty, or how many steps in the future they consider when making decisions. Only time will tell how successful this new approach may be. But I look forward to the insights it could offer about how the brain works and how we might use this information to improve mental health and well-being.