AI doesn’t need to replace humans to weaken human capability

TL;DR

AI boosts productivity (25% less email time, 26% more completed tasks), but MIT EEG research shows LLM-assisted writers had lower brain engagement and worse performance when the tool was removed. Dr. Vishal Kapoor calls this cognitive erosion and proposes an “AI-fed, human-led” discipline: think before prompting, don’t outsource the first question or the final decision. He argues AI adoption should be rationed by task type, especially in education and original research.

Productivity has become the easiest metric in the AI conversation because it is visible and immediately attractive to businesses. Cognitive capacity is harder to count. A faster document tells a company how quickly work was completed; it does not reveal whether the person producing it has become less capable of forming an argument without assistance.

Researchers at MIT’s Media Lab found something similar using EEG monitors on subjects writing essays with and without a chatbot’s help. Those who used an LLM produced faster drafts but showed markedly lower brain engagement while writing, and performed worse than their unaided peers once the tool was removed.

The authors gave the pattern a name: cognitive debt, a deficit that accrues slowly and comes due only once the assistance disappears.

Cognitive offloading, or the act of handing a mental task to an external tool, isn’t new. Calculators and search engines did versions of the same thing. What distinguishes generative AI, researchers argue, is that it doesn’t just retrieve information or compute a sum; it reasons on a person’s behalf, delivering conclusions instead of raw material for a mind to work through. Repeated over months and years, that substitution can risk producing a workflow fluent in prompting but underdeveloped in the reasoning, memory, imagination, and judgment the prompting was meant to support.

Dr. Vishal Kapoor, a researcher and strategist at major multinational banks, believes the productivity conversation needs another variable: the condition of the human mind doing the work. His concern is cognitive erosion from excessive dependence on AI, a risk he argues is still difficult to measure conclusively even as evidence around cognitive offloading accumulates.

“I think it is important that we bring this up at this time,” Dr. Kapoor argues. “We are using so much of AI without any consideration for cognitive erosion.”

Dr. Kapoor is not arguing for a retreat from AI. He uses it extensively and has developed a working philosophy he calls “AI-fed, human-led,” placing the human at the front of the reasoning process while using AI to accelerate or strengthen the work. His own experience sharpened the concern. After using AI for reasoning, he observed a decline in his cognitive profile on a brain training tool, a result he measured at roughly 40 to 50 points on a 1,000-point scale.

He perceives that observation as a personal warning about what can happen when reasoning becomes routinely outsourced.

“When AI becomes a dominant part of the process, reasoning is often the first to go,” Dr. Kapoor says. He also identifies memory and imagination as faculties that could become vulnerable as people increasingly rely on machines to retrieve information and generate possibilities.

His proposed discipline is a practice he fears is becoming obsolete: think before prompting. Dr. Kapoor gives the example of research. A passive user might ask AI to research a country and wait for the answer. A human-led user arrives with observations, a hypothesis, and a defined gap, then asks AI to investigate the missing evidence and challenge the reasoning. “Do not outsource the first question or the final decision,” he argues.

According to Dr. Kapoor, this approach is significant within organizations because AI can amplify weak judgment just as efficiently as strong judgment. A workforce trained to operate AI systems without maintaining independent reasoning could become highly productive while losing the expertise needed to challenge an output, recognize a flawed assumption, or make a decision under unfamiliar conditions.

Dr. Kapoor believes policy has a role in preventing that trade-off from becoming structural. AI adoption, he argues, should be rationed according to the nature of the task, with particular caution around education and original research. “The purpose of education is to develop the human brain. If the human brain is not developed as part of that education, then what’s the use of that education?” he argues.

His argument places responsibility at the center of AI governance. Machines can process information at extraordinary speed, yet human beings bring responsibilities, values, discernment, and philosophies to decisions. Those qualities become especially important in policymaking and high-stakes business judgment, where an efficient answer can still be the wrong answer.

The next stage of AI strategy therefore requires a different investment thesis. Organizations will need to measure what their systems can automate while paying attention to whether their people remain capable of reasoning without them.

Dr. Kapoor envisions future work advising sovereigns, policymakers, and business leaders around precisely this question: what does a society look like when humans live and work alongside increasingly capable AI?

The AI race, in his view, has been framed almost entirely around building more capable machines. He’d rather see it framed around a second, more human debate: whether the people using those machines are still capable of thinking without them.

Get the TNW newsletter

Get the most important tech news in your inbox each week.