AI Verbal Mantras Trigger Unusual Brain Activity, a Cambridge Study Finds

By enlisting volunteers for functional magnetic resonance imaging (fMRI) experiments, researchers whispered peculiar phrases to them, examining the resulting patterns of brain activity. Astonishingly, they found that these model-selected “out-of-distribution” sentences were capable of driving or suppressing activity in the language areas of the brain. In fact, the study demonstrated an 85.7% increase in activity and a 97.5% decrease compared to diverse naturalistic sentences.

Commenting on their findings, the researchers explained, “By selecting phrases that propel neural activity toward a specified goal, we enter an out-of-distribution mode where the encoding model is tested on phrases that may differ significantly from its training data. These arousal phrases exhibited unusual characteristics and were unlikely to be present in ordinary texts, yet they had the power to elicit reactions in the language network.”

Examples of such intriguing phrases included “Turin loves me not, nor will she” and “People on Instagram Be Like, Gross!” The study highlighted the potential for selecting mantras or generative pictures, known as mandalas, to induce specific states within individuals.

This research holds implications beyond the realm of academia, hinting at the profound impact it may have on industries such as visual content and viewer experiences. With the advent of Large Language Models (LLMs), the process of manipulating neural responses becomes even more intense. As our understanding of the intricate workings of the human brain grows, the potential for creating tailored experiences becomes increasingly apparent.

The study concludes with an evocative image—an individual, adorned with a leather bag and concealed beneath a ski mask, immersing themselves in strange pictures and listening to meaningless texts. This vivid depiction underscores the ever-expanding frontiers of knowledge and the uncharted territories that lie ahead in our exploration of the human mind.

• Recent advances in AI systems have made it possible to decode the brain activity of an individual when they are listening to a story or imagining one themselves. A semantic decoder developed by researchers at the University of Texas in Austin is capable of understanding some of the videos that a participant is watching as they undergo brain scans using an fMRI machine. However, this system is not practical outside of the laboratory due to the need for extensive training and a long session to be conducted on a fMRI machine. If further research is done in the development of this system, it can be used in a variety of applications, including in clinical settings, to reveal stories behind the thinking of individuals and help mental health professionals.

Read more about AI: