You probably have heard that our brains make up only about 2% of the body’s total weight, yet consume roughly 20% of our energy (Raichle & Gusnard, 2002). That disproportionate energy demand hints at the immense processing power locked inside the human mind—and why our species has always been strategic in managing mental and physical effort. I emphasized those strategies in my Don’t Rest in Peace book (McCusker, 2016).
From an evolutionary perspective, early humans were constantly , unconsciously calculating efficiency. Accordingly, anthropologists and evolutionary biologists argue that our ancestors often sought foods and resources that offered the greatest nutritional return for the least effort (Kaplan et al., 2000). That logic of minimizing effort while maximizing reward didn’t end with the Stone Age. Modern humans extended it into the way we manage our time, as well as energy.
Enter artificial intelligence. Many of us embrace AI because it saves time, reduces mental strain, and increases efficiency. Just as our ancestors preferred calorie-rich foods to fuel their bodies and brains, we are now drawn to digital tools that fuel productivity with minimal effort. Yet this evolutionary impulse raises an important question: when should we rely on AI, and when should we rely on our own cognition?
The answer may lie in a cost-benefit analysis. Using AI comes with obvious gains: speed, convenience, and access to information. But there are also hidden costs. If we allow AI to handle too much of our mental workload, we may weaken our memory, problem-solving skills, and even creativity over time (Mitchum & Kelley, 2023). On the other hand, strategic use of AI—such as delegating repetitive or low-level tasks—can free us to focus on higher-order thinking and creative work.
The gains versus costs issue was empirically studied by Nataliya Kosmyna, et al.(2025) at MIT’s Media Lab, with collaborators from Wellesley College and Massachusetts College of Art and Design. The research deserves our careful attention.
Let’s begin with the methodology: 54 participants were asked to write SAT-style essays under three different conditions: Brain-only: Write without any external aid; Search engine: Use Google to assist; and LLM (ChatGPT): Use ChatGPT to assist. This setup obtained across three sessions. In a fourth session, participants switched: those who had used AI changed to writing unaided (LLM-to-Brain), and vice versa (Brain-to-LLM).
Measurements were as follows: Participants wore EEG headsets to monitor brain activity (cognitive neural connectivity across alpha, beta, theta bands, etc.). Researchers also analyzed the writing for originality, linguistic patterns, and had humans and AI evaluate the essays. Post-task interviews assessed recall and ownership.
Neural engagement assessment found that the brain-only group showed the highest and most widespread neural connectivity, indicating deep cognitive engagement. The search engine group fell in the middle—more engaged than the AI group, but less than brain-only. And The LLM (AI) group had the weakest neural engagement, suggesting cognitive offloading and diminished mental processing.
Regarding memory and ownership, over 83% of the AI users (LLM group) couldn't quote their own essays, versus only 11% in the other groups. AI user essays appeared more formulaic and less original, and they reported feeling less ownership of the content.
The most surprising research finding was the persistence of the cognitive effects. For instance, in the final session, participants who had initially used AI and switched to writing unaided did not recover their earlier neural engagement—they remained under-engaged. By contrast, those who began writing unaided and then used AI (Brain-to-LLM) showed increased neural connectivity, almost matching the search engine group. The researchers coined “cognitive debt” to describe the long-term cost of over-relying on AI. Thus, while AI could ease immediate effort, it appeared to erode critical thinking, creativity, memory retention, and essay ownership.
I must underscore something obvious: The MIT study involved only 54 subjects. That is hardly a ringing endorsement to its reliability and validity. Its findings may or may not be replicated in the future. Most important is that the study found exactly what I expected it to find. Maybe this is just one example of my confirmation bias.
Regardless, for me, the best way forward is not to use AI as a replacement for human effort, but as a partner. Just as the body regulates how much energy goes to the brain and other systems, we can regulate how much work we give to AI versus how much we keep for ourselves. The challenge is to strike a balance: gaining the efficiency AI provides without losing the unique cognitive strengths that make us human. In this sense, using AI should not merely be an evolutionary continuation of our search for maximum return on minimal effort. We must mindfully balance not only calories, but our time, attention, and intellectual engagement.
References
Kaplan, H., Hill, K., Lancaster, J., & Hurtado, A. M. (2000). A theory of human life history evolution: Diet, intelligence, and longevity. Evolutionary Anthropology: Issues, News, and Reviews, 9(4), 156–185. https://doi.org/10.1002/1520-6505(2000)9:4<156::AID-EVAN5>3.0.CO;2-7
Kosmyna, N., Hauptmann, E., Yuan, Y. T., Situ, J., Liao, X.-H., Beresnitzky, A. V., Braunstein, I., & Maes, P. (2025). Your brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing task. MIT Media Lab. https://arxiv.org/abs/2506.08872
McCusker, P. J. (2016). Don't Rest in Peace: Activity-Oriented, Integrated Physical and Mental Health (New York: Amazon).
Mitchum, A. L., & Kelley, C. M. (2023). The “Google effect” in the age of artificial intelligence: How reliance on external memory systems may impact learning and cognition. Memory & Cognition, 51(6), 1249–1264. https://doi.org/10.3758/s13421-023-01485-9
Raichle, M. E., & Gusnard, D. A. (2002). Appraising the brain’s energy budget. Proceedings of the National Academy of Sciences, 99(16), 10237–10239. https://doi.org/10.1073/pnas.172399499
No comments:
Post a Comment