An astonishing study from MIT has revealed surprising insights about how our brains interact with AI, and the results could radically alter the way we learn and think. However, you should not be swayed by false viral headlines claiming that its use makes us “dumber.” The researchers themselves have sought to focus the debate on how our brains change in response to the use of AI resources.
The research indicates that, although using AI can lighten the cognitive load when performing certain tasks, this may entail an adjustment in the depth of information processing and the formation of long-term memory. The brain adapts in accordance with how it is exercised; if AI takes control of high-level cognitive tasks, the brain may reallocate its resources to be utilized in other areas.
A team of MIT researchers, led by Nataliya Kosmyna, set out to discover the cognitive impact of using an AI system (such as ChatGPT) for educational tasks. To this end, they recruited 54 participants and divided them into three groups:
All participants wrote content across three sessions, and a subset of 18 participants completed a fourth session, where roles were reversed: those from the LLM group switched to “Brain Only” (LLM-to-Brain) and vice versa (Brain-to-LLM). To measure brain activity, the researchers used electroencephalography (EEG), in addition to natural language processing (NLP) analysis and interviews.
Spoiler: Over the course of four months, users who relied on LLMs consistently exhibited lower performance at the neural, linguistic, and behavioral levels.
Participants in the LLM group demonstrated a significantly lower ability to accurately cite their own content, even just minutes after having written it. Whereas 83.3% of LLM users failed to cite correctly in the first session, only 11.1% of the Search and Brain Only groups faced the same difficulty. In the fourth session, when LLM users did not use the tool, 78% failed to cite their studies, compared to only 11% from the group previously operating as “brain only.”
These findings, which have already gone viral, have been viewed by many as a trigger for deterioration in memory encoding due to the externalization of the cognitive process.
In essence, relying on AI for these processes could lead to a decline in your own abilities to perform them.
According to the researchers’ conclusions:
Within the alpha band, associated with internal attention and semantic processing, the Brain Only group demonstrated significantly stronger connectivity. Connectivity in the theta band, linked to working memory load and executive control, was much higher in the Brain Only group, indicating a greater cognitive load. According to the MIT researchers, this suggests that AI reduces the brain’s need to internally generate and manipulate information.
The study suggests that the use of LLMs may influence our retention capacity. Participants who used an LLM displayed a markedly reduced ability to accurately cite their own essays, even shortly after completing them. That is to say, delegating content writing functions to AI appears to affect certain cognitive skills.
Moreover, the study found that AI systems can exacerbate “echo chambers,” limiting exposure to diverse perspectives and reinforcing existing beliefs. This echo chamber or resonance effect has long been debated in the context of social media and the way dependence on algorithms that supply the content we are known to enjoy ends up diminishing the diversity of viewpoints and radicalizing positions.
In addition, essays generated with the assistance of LLMs demonstrated excessive uniformity, with less variability than in other groups. In other words: AI systems produce more predictable language, with fewer variations. In fact, the Brain Only group stood out by producing essays that were distinguishable from those written with AI or search engines.
Indeed, in a twist characteristic of today’s debate surrounding the impact of AI on education, human teachers, exposed daily to LLM-assisted essays, were much more skeptical regarding content uniqueness and structure, rating AI-assisted essays lower in these metrics compared to the ratings of an AI judge. They could even recognize a distinctive writing style associated with the LLM group. In contrast, the AI judge often rated the AI-generated essays as higher in uniqueness and quality.
The authors of the study themselves explain that AI can reduce cognitive load and free up resources for other tasks, but at the expense of reduced long-term memory formation. In this context, the brain adapts to the way we train it: if AI performs high-level planning, the brain will allocate fewer resources to those functions.
In any case, the authors stress that this is a preliminary study, and request that those referencing it avoid using terms such as “stupid,” “dumb,” “brainless,” “harm,” “damage,” “passivity,” “reduction,” and similar words when describing how these tools affect humans.
Let us simply leave it at, “untrained.”
Image: ChatGPT
Your email address will not be published. Required fields are marked *
Δ