The federal government has been the first to jump into artificial intelligence recently, testing it for use in everything from healthcare to HR. This is because technology companies continue to lobby against the increasing deployment of advanced AI tools in governments, such as agents or generation AI, and lawmakers are trying to weigh the potential negative consequences of doing so. The government is also tackling the question of whether to place guardrails around technology. And in this maelstrom of potential AI advantages and disadvantages, there is new research from MIT. This may provide a unique new reason to pump at least a bit of AI-based brakes.
A study by MIT's Media Lab tested cognitive function in a variety of student groups. I used AI tools such as ChatGpt to split key tasks like writing essays according to the amount I would accomplish over several months. The study sought to find out how people's brains responded to using AI tools over time. And the answer to that question is, according to research: Participants who used AI exclusively to help write essays showed weaker brain connections, reduced memory retention, and a decline in their sense of ownership towards work. Essentially, their brains have become lazy. And even if I stopped using AI tools later, the effect still persisted.
The study involved 54 students from five Boston-area universities, all wired with EEG headsets to monitor brainwave activity. After obtaining their thinking patterns and brain wave baselines, they were assigned a task to investigate and write various essays over four months.
To achieve the essay writing goals, students were divided into three groups. The first group was told to write essays using a larger language model, like ChatGpt. They interacted with the AI to help them create each essay, but they let all the heavy lifts. The second group wrote their own essays, but they were able to use Google and other search engines to conduct research and find sources for citations. The last group was to go to completely old school, and simply wrote their own papers and conducted practical research as they did in the library. The EEG headset monitored brain activity over time in all three groups to see how different technologies and writing methods affected them compared to baseline.
The group that used Google in the study showed moderate amounts of cognitive activity, while the group who worked alone without technical help showed the highest. Perhaps not so surprising, the group who wrote papers using ChatGPT-4 alone showed the lowest amount of EEG activity. In fact, cognitive function has decreased over time in key areas of the brain. Incidentally, the ChatGpt group had the weakest connection to their work. 83% of students were unable to remember the key points of the essay and were unable to provide accurate citations from the paper.
According to the authors of the MIT experiment, “This study illustrates the pressing problem that learning skills are likely to be reduced based on the results of the study. Using LLMS [large language models] Although benefits were initially apparent as demonstrated over the course of four months, participants in the LLM group were worse than their counterparts in the brain-only group at all levels of neural, language and scoring. ”
Perhaps it is the fact that cognitive decline in brain wave activity and reduction in brain wave activity measured in the AI-Only group have continued ever since the study was completed. Even after ceasing to use CHATGPT, participants still have slower brain activity. When you start outsourcing your thoughts, you will see that your brain does not jump accurately at the opportunity to regain the wheels.
Reports from the MIT experiment do not suggest that people will stop using AI, which is far from AI. In fact, the ChatGPT group was able to complete the task much faster than the other two. AI tools are absolutely useful for efficiency, especially for time-consuming tasks such as data entry and long document summarization. However, the use of these tools suggests that it is important not only to ensure accuracy, but also to keep up with cognitive health. This advice will be doveTailed in another important study conducted by Stanford University and detailed in the newly released 2025 AI Index. It emphasizes the importance of keeping people in a loop all the time as a way to improve AI governance.
In government, the solution might be to develop guidelines that encourage employees to overcome problems before turning their eyes to AI. It's as easy as writing down a rough draft before asking for a rewrite, or manually outline your ideas and then manually depicting them before polishing your AI. There may also be a bit of cognitive hygiene to ensure that humans stay in loops and balls, like the cybersecurity playbooks and data governance frameworks already in place at most agencies.
Although AI can help us make us faster, it's because MIT research warns us that we should probably never forget how to do things perfectly.
John Breeden II is an award-winning journalist and reviewer with over 20 years of experience covering technology. He is the CEO of High-tech writersa group that creates technical thinking leadership content for organizations of all sizes. Twitter: @labguys
