Prof. Dr. John-Dylan Haynes, Director of the Berlin Center for Advanced Neuroimaging (BCAN) and Professor at the Bernstein Center for Computational Neuroscience (BCCN) at Charité in Berlin: When the term “mind reading” is used in everyday language, it often implies looking at a person and, through a sort of magical intuition, being able to see what that person is thinking or feeling. We can indeed assess the emotional states of others to a certain extent based on body language, facial expressions, gestures, and prosody—that is, the way a person speaks. But the body’s exterior only gives us a very rough idea of a person’s inner world—such as what someone is feeling or whether they are awake. So in everyday life, we rely on people to tell us what’s going on in their minds.

In recent years, it has become clear that there is a very close connection between our thoughts and our brain activity. This raises the question of whether technical measurements of brain activity can be used to determine what a person is thinking at any given moment. First, we should briefly explain what is meant by “thoughts” in this field of research, since it is not limited to sentence-like, linguistic thoughts (“I need to go grocery shopping soon”) . It also includes sensory experiences and perceptions, memories, feelings, plans of action, and much more. Thoughts thus encompass the entire spectrum of what one can experience. In principle, all these types of thoughts are accessible to brain researchers.

So, can we determine a person’s thoughts by measuring their brain activity? The answer is “yes and no.” Mind reading in the strict sense would mean reading brain activity like letters in a book. However, this would require an understanding of the system of the brain’s “language of activity”—namely, its syntax (the forms of language) as well as the meaning of individual linguistic symbols, i.e., its semantics. Brain research is still a long way from reaching that point.

In the field of research known as “brain reading,” however, a different approach is taken. The brain’s activity patterns are not read directly but are cracked like an incomprehensible code using mathematical methods. This is similar to the code of the Enigma, the German encryption machine from World War II, which was deciphered using statistical methods. In a similar way, researchers are attempting to use mathematical and statistical methods to determine what a person is thinking at any given moment based on brain activity. In brain reading, therefore, brain patterns are not “read” in the strict sense, but rather thoughts are decoded through code-cracking and deciphering. Thanks in particular to the capabilities of machine learning and artificial intelligence, significant progress has been made in this area in recent years.

However, it is currently only possible to approximate thoughts to a certain degree. Thus, there is still no perfect reconstruction available today. Furthermore, one must first learn how specific thoughts are reflected in the individual activity patterns of a particular person’s brain. This means that an individual model must first be established to determine how thoughts are encoded in a specific brain. The following applies: Extremely vivid thoughts—such as images, sounds, and emotions—can be decoded relatively well. Even very broad categories can be easily recognized. However, the resolution of today’s MRI machines is insufficient for decoding the finer details of thoughts.

One factor makes this work particularly difficult: The activity patterns associated with a specific thought vary from person to person. On the one hand, individual brains differ anatomically. On the other hand, the learning experiences and associations that each person has also differ. My favorite example: A person who had a dog as a child—one that was a loyal friend—has a different set of associations with dogs than someone who was bitten by a dog as a child. These different associations are reflected in the activity patterns. That’s why, in brain reading, the analysis must be tailored to the individual brain. There are various ways to do this. One can take the laborious route and learn separately for each person which activity patterns reflect their thoughts. Or one can use a trick: one learns to transform the patterns from one brain to another. This would require only a quick calibration measurement, but that is still the subject of ongoing research.

It’s also important to approach this topic from an ethical perspective: Is it even acceptable to learn about a person’s thoughts by indirectly analyzing their brain activity? After all, this could make it possible to decode a person’s thoughts without their consent or even their knowledge. This raises an ethical problem. For the past 16 years, we have been hosting the Winter School on the Ethics of Neuroscience and AI in Berlin every year, where we engage in in-depth discussions with students, doctoral candidates, postdocs, and researchers about the ethical dimensions of brain reading. In these discussions, certain applications—especially so-called neuromarketing—are repeatedly viewed with skepticism. This is because most people agree that it is ethically questionable to use brain scanners and AI techniques to optimize products in a way that triggers an irresistible urge to buy. After all, that would constitute a dramatic encroachment on freedom of choice.

Recorded by Stefanie Flunkert

Originally published on November 20, 2011

Question reanswered on June 21, 2026