AI Is Already Making Us Less Human | Enterprise workflow automation software
- Gammatek ISPL
- 7 hours ago
- 7 min read
By Gaammatek ISPL
Last updated: September 2026 | 14 min read
Why This Matters
You've probably noticed it in small moments: reaching for ChatGPT before trying to remember a fact yourself, letting an AI draft the email you used to write in two minutes, asking an assistant to summarize an article instead of reading it. None of these feel significant individually. But a growing body of research from MIT, Carnegie Mellon, Oxford, UCL, and other institutions is now measuring what happens when these small moments compound — and the early data suggests something worth paying attention to: measurable declines in memory, independent problem-solving, and confidence in one's own thinking, tied directly to how heavily people lean on AI. This isn't a philosophical thought experiment anymore. It's showing up in controlled studies, and it affects anyone who uses these tools regularly — which by 2026, is most of us.
The Research: What's Actually Been Measured
The concept researchers keep returning to is cognitive offloading — the well-documented tendency to delegate mental effort to an external tool rather than exercise it yourself. It's not new (calculators and GPS do this too), but AI offloading is different in scale and depth: it doesn't just replace one narrow skill like mental math, it can replace the entire process of forming an argument, evaluating information, or solving a novel problem.
A few specific findings from recent research:
Researchers at Carnegie Mellon, Oxford, MIT, and UCLA ran an experiment giving one group of participants an AI assistant to help solve fraction-based math problems, while a second group solved them independently. When the AI assistant was removed, the group that had relied on it struggled significantly more to solve similar problems on their own — a measurable performance gap appearing after as little as ten minutes of AI-assisted practice.
A 2025 study published in the journal Societies, led by researcher Michael Gerlich, surveyed 666 participants across age groups and education levels and combined that with in-depth interviews. The research found that heavier reliance on AI tools for everyday problem-solving correlated with reduced critical thinking performance — though notably, higher educational attainment appeared to partially offset this effect, suggesting that people with stronger foundational reasoning skills navigate AI use somewhat more safely than those without.
Separately, Microsoft and Carnegie Mellon researchers surveyed 319 knowledge workers and found a pattern worth sitting with: workers who expressed higher confidence in AI's capabilities also reported feeling that critical thinking now required less personal effort — and, notably, less confidence in their own cognitive abilities. The tools didn't just replace some of the thinking; they appeared to be quietly eroding people's trust in their own minds.
A study out of Middlesex University, published in Technology, Mind, and Behavior, examined what researchers termed "executive function attenuation" in high-frequency generative AI users — looking specifically at the mental functions responsible for planning, focus, and self-regulation, and finding behavioral evidence that heavy AI reliance measurably dampens them.
Not every researcher agrees the picture is this stark. Dr. Sam Gilbert of UCL's Cognitive Neuroscience department has publicly pushed back on the more alarming interpretations, expressing skepticism that AI use is meaningfully damaging human cognitive flexibility in the ways some coverage suggests. That disagreement is worth taking seriously — the research here is young, sample sizes are often modest, and "correlation" is doing a lot of work in several of these studies. But the direction several independent research groups are converging on, even accounting for that skepticism, is consistent enough to warrant real attention rather than dismissal.
It's Not Just Thinking — It's Writing, Too
One of the more nuanced findings comes from research published in the Journal of Educational Psychology in early 2026, which tracked student writing quality and found a growing split, or bimodal distribution, in outcomes based on howstudents used AI, not just whether they used it. Students who wrote a first draft themselves and then used AI to refine it showed genuine improvement in their work. Students who asked AI to generate a draft first and then edited it showed measurable declines in argument structure and original insight over a two-year period.
This distinction matters enormously, and it's the piece most alarmist coverage skips: the sequence of human effort relative to AI assistance appears to determine the outcome. AI used as a second pass after genuine effort seems to sharpen thinking. AI used as the first pass, with humans editing afterward, seems to dull it.
Beyond Cognition: What "Less Human" Actually Means Here
The phrase "AI is making us less human" can sound overwrought, so it's worth being precise about what the research actually supports and what it doesn't.
It does not show that AI is changing human biology, emotion, or fundamental capacity for connection. What it does show, with growing consistency, is a narrowing of certain everyday practices that we've historically associated with being a reflective, self-directed thinker: forming an argument from scratch, sitting with a hard problem before reaching for help, tolerating the discomfort of not immediately knowing an answer, trusting your own judgment enough to act on it without external validation.
Those aren't abstract philosophical qualities — they're skills, in the same sense that physical strength is a skill. Skills atrophy without use. The research increasingly suggests that AI, used carelessly, provides a very comfortable way to stop using them.
An Implementation Consideration: Where This Shows Up at Work
This isn't only a personal or educational concern — it has direct operational implications for any organization deploying AI tools broadly, which is where this connects to something concrete rather than abstract.
Consider a common pattern inside companies rolling out AI copilots for knowledge work: initial productivity metrics often look great — tasks complete faster, output volume rises. But the Microsoft/CMU confidence-erosion finding above suggests a second-order effect worth watching for: employees who lean on AI heavily for complex judgment calls may become less capable of handling the moments when the AI is wrong, unavailable, or simply not equipped to catch a nuance specific to the situation. An organization that doesn't track this can end up with a workforce that looks more productive on paper while becoming quietly more fragile in exactly the moments that matter most — a system outage, an edge case, a decision that needs contextual judgment a model doesn't have.
The practical fix isn't avoiding AI tools — it's structuring how they're used. Based on the writing-quality research above, the sequencing principle likely generalizes beyond writing: having people attempt independent judgment first, then using AI to check, refine, or extend that judgment, preserves more of the underlying skill than defaulting to AI output first and editing afterward.
This is also where workplace automation and HR software decisions matter more than most companies realize. As enterprise workflow automation software and enterprise human resources software increasingly mediate hiring decisions, performance reviews, and day-to-day task routing, the same sequencing question applies at an organizational level: is the automation replacing the moment where a manager exercises judgment, or is it removing the busywork around that judgment so the manager can spend more attention on it? Companies evaluating enterprise HR software and workflow automation platforms would do well to ask vendors directly which model their tools are built around — because the difference isn't cosmetic, it's the difference between automation that preserves institutional judgment and automation that quietly erodes it.
A Reasonable Middle Ground
It's worth naming directly: some of the loudest "AI is making us dumber" commentary overstates certainty the research doesn't yet support. Sample sizes in several of the cited studies are in the hundreds, not tens of thousands. Most track short-term behavioral changes, not confirmed long-term cognitive decline. And UCL's Dr. Gilbert isn't alone in urging caution against treating early, correlational findings as settled fact.
At the same time, dismissing the pattern entirely ignores that multiple independent research teams — with different methods, different institutions, and different sample populations — are converging on a similar directional finding: heavy, unstructured reliance on AI for thinking tasks correlates with measurable declines in independent performance and self-confidence, at least in the short term. That convergence, even without long-term certainty, is a reasonable basis for changing how you personally and organizationally use these tools — not a reason for alarm, but a reason for intention.
What Actually Helps, Based on the Evidence
Pulling directly from what the research above supports, rather than generic advice:
Attempt the task yourself before reaching for AI, even briefly — the writing-quality research suggests this single sequencing choice is more predictive of outcome than whether AI is used at all.
Notice when you're using AI to avoid discomfort, not to save time — the discomfort of not immediately knowing something is often exactly the moment where the mental exercise the research is measuring actually happens.
Treat AI-generated confidence as separate from actual capability — the Microsoft/CMU findings specifically flagged a gap between how confident people felt using AI and how capable they remained without it; checking your own reasoning independently, occasionally, is a way to keep that gap visible to yourself.
In organizational settings, audit automation for what it removes, not just what it adds — a workflow or HR tool that removes busywork around a decision is different from one that removes the decision itself; the first supports human judgment, the second replaces it.
The Honest Takeaway
The evidence doesn't support a dramatic claim that AI is stripping away our humanity in some sweeping, irreversible way. It supports a narrower, more useful one: specific cognitive habits — independent problem-solving, tolerance for productive struggle, confidence in one's own judgment — are measurably weaker in people who lean on AI without intention, and measurably preserved or even strengthened in people who use it as a second step rather than a first one. The technology isn't the deciding factor. The sequence is.
Comments