Researchers from the University of Copenhagen and University of Helsinki created a machine learning algorithm that can predict a person’s preferences based on how their brain responses match others’ brain responses. This research could be used to provide content tailored to a given person’s brain and reveal more about human behavior.

Online algorithms are used daily to guess user preferences based on past online actions and how those actions compare to others. These algorithms are used in streaming, news, shopping, music and more. These algorithms operate on collaborative filtering, which uses hidden patterns in human behavior to predict what a user may find interesting or appealing.

The team explored how an algorithm could be used to match an individual’s pattern of brain responses with the brain responses of other people. For this study, the team focused on making an algorithm that can predict a person’s attraction to a face they have not seen yet.

In past studies, EEGs were placed on participant’s heads and then participants were shown various faces. These studies demonstrated that machine learning can use electrical activity from the brain to detect which faces participants found the most attractive.

Current collaborative filtering techniques are based on explicit behavior, like ratings, click behavior and content sharing. These are not reliable and do not ultimately reveal the real underlying user preferences. Also, there are other factors that may affect their choices, including social norms.

The team conducted an experiment where participants were shown a large number of images from human faces and asked to look for faces they found the most attractive. While the participants were looking at faces, researchers recorded their brain signals. This data was used to train a machine learning model to distinguish between brain activity when participants saw an attractive person versus when they saw someone they did not find attractive. Then a new machine learning model was trained on the previous algorithm's data and data from a larger number of participants. The new algorithm calculated which new facial images a given participant may find attractive. This prediction was based on individual participant’s brain signals and on how the other participants responded.

Long term, the team’s new method could provide more nuanced information about people’s preferences. It would also be a step toward an era of mindful computing. Mindful computing would allow users to have more access to information about themselves using a combination of computers and neuroscience techniques.

The new tech comes with a new challenge, how to protect brain-based data from misuse. It is important to consider data privacy, ownership and ethical use of the raw data being collected. Currently, there are a variety of companies that misuse their customer’s behavior data. This problem could only become more dangerous with data that is coming from the user’s brain.

There is a long way to go before the team’s technique can be applied beyond the lab. Brain-computer devices need to become cheaper and easier to use before the average person will be using this tech at home. According to the team, this is at least a decade away.

This research was presented at WWW 21 Proceedings of the Web Conference 2021.