These findings debunk numerous prominent papers that claimed to be able to read minds using EEG.
Is it possible for someone to read their mind by studying the electrical signals coming from their brain? This may be more complex than most people believe.
Researchers at Purdue University, who work at the intersection between artificial intelligence and neuroscience, claim that a well-known dataset used to answer this question was confounded. This means that many eye-popping results based on this dataset and which received high-profile recognition were false.
Over more than a year, the Purdue team conducted extensive testing on the dataset. These tests examined brain activity in individuals who participated in a study exploring a series of images. Each participant wore a cap containing dozens of electrodes as they viewed the photos.
The work of the Purdue team is published in IEEE Transactions on Pattern Analysis & Machine Intelligence. The National Science Foundation provided funding for the Purdue team.
“This measurement technique, known as electroencephalography or EEG, can provide information about brain activity that could, in principle, be used to read minds,” said Jeffrey Mark Siskind, professor of electrical and computer engineering at Purdue’s College of Engineering. The problem is that EEG was used in a way that contaminated the data. Researchers did not randomize the order of images in the study. This allowed them to determine which image was being seen by reading EEG timing and ordering information. They could not solve the fundamental problem of visual perception from brain waves.
When they couldn’t get similar results from their tests, the Purdue researchers began to question the dataset. They began to analyze the data and discovered that the dataset was not randomized.
Hari Bharadwaj is an assistant professor with a joint appointment at Purdue’s College of Engineering and Health and Human Sciences. “This is one of our challenges when working in cross-disciplinary areas research areas,” she said. Cross-disciplinary research is often required for critical scientific questions. Sometimes, researchers trained in one discipline must learn the common pitfalls when applying their ideas to another. This case shows that the previous work was affected by a disconnect between AI/machine learning scientists and neuroscientists.
The Purdue team reviewed publications that used the dataset for tasks such as object classification, transfer learning, and generation of images depicting human perception and thought using brain-derived representations measured through electroencephalograms (EEGs)
Ronnie Wilbur, a professor who holds a joint appointment in Purdue’s College of Health and Human Sciences and College of Liberal Arts, said that “the question of whether someone could read another person’s minds through electric brain activity was very valid.” “Our research has shown that a better approach to reading minds is required.”
