AI is starting to look less like a productivity tool and more like a cognitive lease agreement. The latest warning comes from MIT researchers who tracked 54 students over four months. They measured brain activity with EEG scans and found that the heaviest AI users were the ones most likely to go quiet upstairs when the tool disappeared.
For South African founders, analysts, lawyers, marketers, and product teams, that should land as more than a headline. The same software shaving hours off a workday can also shave off the habit of thinking through a problem from first principles, especially when the output looks polished enough to stop asking questions.
What the MIT study found
The MIT group followed students through writing tasks done with and without AI assistance. The basic pattern was not flattering to the people leaning hardest on the tool. Their neural connectivity dropped sharply, by 47% in the most exposed group, and their recall of AI-produced material was weak. When the AI support was removed, their performance fell off fast.
That combination matters because it points to more than simple convenience. A person can use automation to speed up a task and still remain mentally engaged. This study points to something different. The frequent users appeared to hand over the hard part of the task, then struggle to recover the thread once the machine was taken away.
The researchers used the phrase “cognitive atrophy” to describe the effect. The warning is plain enough: if a person keeps outsourcing drafting, structuring, and problem-solving to a system that always responds quickly, the brain gets fewer chances to practise the work that builds memory, judgment, and independent thought.
Why South African workplaces should care
The local temptation is obvious. AI writes the first draft of a pitch deck. It trims a board pack. It turns a messy client brief into neat bullet points. It drafts code, summarises a legal clause, or rewrites a proposal into cleaner English before anyone has had to sit with the actual problem.
That is useful, until it becomes a habit that weakens the staff member using it.
In Johannesburg and Cape Town, small teams are already doing the work of larger ones, so AI can look like a force multiplier. In reality, it can become a crutch if the team starts accepting the output instead of interrogating it. A consultant who no longer knows how to read raw numbers. A junior lawyer who can produce a draft but not explain the logic behind it. A developer who can prompt well but cannot debug without assistance. These are not theoretical risks. They are the shape of deskilling.
The pressure is sharper in sectors where speed gets rewarded and no one has time for slow thinking. Finance teams need summaries. Agencies need content. Founders need decisions. Fast summaries can flatten nuance, and AI is very good at sounding certain right before it gets something wrong.
A recent conversation around web tech insights makes the point neatly. The web industry has spent years celebrating automation as if fewer human touchpoints automatically meant better outcomes. It does not. The real test is whether the person using the tool is still doing the cognitive lifting that gives the tool value in the first place.
Where the risk shows up first
The MIT findings line up with everyday workplace behaviour. The danger is not one dramatic collapse; it is a slow drift.
A manager starts using AI to summarise reports, then stops reading the source material.
A marketer uses it for every campaign line, then loses the muscle for writing in a voice that sounds like the brand.
A support team leans on scripts and chat prompts, then struggles when a customer asks something unscripted and messy.
A founder keeps asking AI to shape strategy notes, then becomes a collector of plausible options rather than a decision-maker.
That is the real problem behind dependency. The output still looks decent, so the user does not notice the skill leak until the tool is switched off or produces a bad answer. By then, the habit has already settled in.
A memory problem hides inside the convenience. If the user did not have to wrestle with the structure of the answer, they often remember less of it. That should worry any business that expects staff to build institutional knowledge rather than merely move documents around.
How to use AI without going flat
The answer is not to ban AI from the office. That would be a nostalgic and pointless gesture. The better move is to set rules that keep the human brain in the loop.
Start with the tasks that should never be fully automated.
- Final strategy decisions should be reviewed by a person who understands the market, not just the prompt.
- Legal, financial, and client-facing drafts should be checked against the source material, not accepted as clean because they sound polished.
- Technical work should include at least some AI-free debugging or problem-solving, so the team keeps its core skills warm.
- Important proposals should be drafted with AI, then rewritten by a human who can spot weak logic and local context the model missed.
That last point matters in the South African market. A tool can produce fluent English. It cannot automatically understand a municipal payment cycle, a sector-specific compliance issue, or the difference between a broad Johannesburg audience and a tighter SME base in Durban. Humans still have to supply that judgment.
Training should also change. Teams need more than prompt tips. They need practice in spotting bias, checking sources, and pushing back when the machine is confident but shallow. Prompting is not a shortcut around thinking. It is a thinking skill in its own right, because the quality of the question usually determines the quality of the answer.
Businesses can go one step further and create AI-free windows for specific tasks. A weekly memo written without assistance. A strategy session where no one opens the chatbot. A junior analyst who has to produce the first pass before the model gets involved. These small constraints keep the cognitive engine from idling.
The bigger issue is not speed
The debate around AI usually gets framed as a race between people and machines. That framing is lazy. The real question is whether people are still doing the parts of the job that make them worth paying.
If AI handles the grunt work and leaves the human to judge, refine, challenge, and decide, the relationship is healthy. If AI becomes the first and final stop for thought, the user starts losing the very skills that made the tool useful in the first place.
That is why the MIT findings hit a nerve. They do not prove that every AI user is becoming intellectually weaker. They do show that repeated reliance can thin out the mental effort that keeps memory, focus, and critical thinking sharp. In a workplace where founders want speed and teams want less admin, that warning is easy to ignore right up until it shows up in the quality of the work.
The blunt question is whether AI is making people sharper or merely more dependent. If the answer is the second one, the productivity gains are already being paid for in advance.
