We Are Not Machines — But We Are Becoming One

AI is not only changing what workers do. It is changing how they think, read, and engage with their own labour — and no workplace AI policy currently addresses that.

Sign up for my Substack daily AI newsletter here.

See my AI Training course portfolio for corporate Business Leaders here.

Article link: https://open.substack.com/pub/johanosteyn/p/we-are-not-machines-but-we-are-becoming?r=73gqa&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true

Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/

Recently, a robot magician called D4RYL was rejected for membership of the Magic Circle. Its tricks were technically exemplary. The organisation decided nonetheless that it did not engage the audience's emotions as a flesh-and-blood performer would. Magic Circle president Marvin Berglas put it directly: "Technology can enhance magic beautifully, but historically there is a human storyteller at the centre." The story is charming, and it is also precise: there is something about human presence, human error, and human feeling that the technically proficient robot could not replicate, and the people making the decision knew it when they saw it.

The same week, Guardian columnist Heather Stewart reviewed Sarah O'Connor's new book We Are Not Machines: The Fight for the Future of Work — a title that captures both a statement of fact and an increasingly urgent aspiration. O'Connor, an award-winning Financial Times journalist who has spent a decade reporting on the labour market, develops her central question through sustained reporting on Amazon warehouse workers, content moderators in India and Costa Rica, and translators whose creative craft has been reduced to correcting mediocre AI output: "We think we're robotising our work," she writes, "but what if we're robotising ourselves?"

That question does not appear in any AI impact assessment, any workplace AI policy, or any board-level AI strategy document I have encountered. It should.

CONTEXT AND BACKGROUND

The dominant frame for discussing AI and work in 2026 is the displacement frame. How many jobs will AI eliminate? Which occupations are most exposed? What retraining will be required? These are important questions and they deserve serious answers. They are also, in an important sense, the wrong questions — because they focus on whether workers will have jobs and say almost nothing about what those jobs will do to the people performing them.

O’Connor’s reporting illuminates a different and deeper concern. The translators she speaks to have not lost their jobs. They have kept them — but the jobs have been transformed. Where once they exercised linguistic creativity, cultural judgment, and the hard-won craft of rendering meaning across languages, they now spend their days correcting mediocre AI-generated text for a fraction of their former pay, in a role known as machine translation post-editing. One translator named Petr tells her: “I want to have something creative, but I’m not sure that I can have a creative job that’s not endangered. Everywhere you step, there’s AI.” He still has work. What he has lost is harder to name — and it is not captured by any employment statistic.

The Amazon warehouse workers she visits have their tasks constantly surveilled. The content moderators in India and Costa Rica watch hours of mind-numbing footage to train the AI systems that monitor the warehouses. They are not being replaced. They are being used as components in a system whose purpose is to eventually need fewer of them — and in the meantime, to require them to perform work that is specifically designed around what the machine needs rather than what the human is capable of. “We think we’re robotising our work,” O’Connor writes, “but what if we’re robotising ourselves?”

INSIGHT AND ANALYSIS

The evidence that AI is doing something to human cognition beyond changing employment is accumulating in academic research at a pace that the workplace AI policy conversation has not yet absorbed. A 2025 MIT study found that participants who exclusively used AI to help write essays showed weaker brain connectivity, lower memory retention, and a fading sense of ownership over their work. More concerning still, the cognitive declines measured in the AI-only group continued long after the study was completed — even after participants stopped using AI, they still showed reduced neural engagement. The researchers described a phenomenon in which outsourcing thinking to AI does not merely substitute for a cognitive function but degrades the capacity itself.

A January 2025 peer-reviewed study published in Societies, authored by Michael Gerlich of the SBS Swiss Business School, found a significant negative correlation between frequent AI tool usage and critical thinking abilities, mediated by increased cognitive offloading. Younger participants exhibited higher dependence on AI tools and lower critical thinking scores than older participants. The study concluded that AI tools can enhance efficiency for time-consuming tasks but introduce risks to originality and critical thinking when users become overly dependent on AI-generated content.

A May 2026 paper published in ScienceDirect introduced the concept of AI-chatbots-induced cognitive atrophy, describing the potential deterioration of essential cognitive abilities resulting from excessive dependence on AI tools — including critical thinking, decision-making, and memory — and noted that empirical evidence for these effects is growing across disciplinary silos of education, organisational studies, and clinical medicine.

These findings are not arguments against AI. They are arguments for understanding what AI does to people when it is introduced into the workplace without deliberate attention to the cognitive conditions of those using it. A landmark study presented at the CHI Conference on Human Factors in Computing Systems by researchers from Microsoft Research and academic partners found self-reported reductions in cognitive effort and confidence among knowledge workers using generative AI — workers reported engaging less in deep, reflective thinking and preferring quick AI-generated solutions over independent analysis. That is a finding about what AI does to the thinking of people who still have jobs, and it has implications that reach well beyond the displacement question.

IMPLICATIONS

O’Connor’s governance argument is direct and practically important. Just because a robot may technically be able to perform a task does not mean we should accept that it should. She offers three examples from her reporting that make this concrete. A Dutch nurse caring for an elderly patient at home does so with a humour and empathy that a robot carer could not provide. A Hollywood writers’ strike negotiated genuine creative control over whether and how AI is deployed in the production process. Swedish miners at the Renström mine negotiated the introduction of autonomous underground trucks alongside management, producing a deployment that workers understood, accepted, and participated in designing.

The governance lesson that connects all three is the one that O’Connor draws most directly: “Technology is designed by people, made by people, and adopted by people. And it is perfectly reasonable for policymakers, business leaders, workers and consumers to say ‘yes’ to some uses of new technology, or workplace changes induced by technology, and ‘no’ to others.” That is not a Luddite argument. It is a design argument — and it is one that most organisations have not yet made explicit in their AI deployment decisions.

For boards and executives, the implication is specific. The AI impact assessments your organisation is conducting — if it is conducting them at all — are almost certainly evaluating AI deployment in terms of efficiency gains, cost savings, headcount implications, and regulatory compliance. They are almost certainly not evaluating what the deployment does to the cognitive capacity, creative engagement, and professional ownership of the people performing the work. The MIT and Microsoft Research findings suggest that those dimensions are not trivial. They are the dimensions that determine whether the organisation retains, over time, the human capability that made it worth deploying AI to augment in the first place.

The distinction between augmentation and subordination is the one that workplace AI governance most urgently needs to develop. Augmentation means AI handles the tasks that are routine, repetitive, or below the threshold of what makes the work meaningful, freeing the human to apply judgment, creativity, and relational intelligence to the remainder. Subordination means AI defines the task structure, the pace, the evaluation criteria, and the output format, and the human performs whatever the system requires. The first makes workers more capable over time. The second makes them less so. Both can be described as AI deployment. Only one deserves the governance frameworks that boards are currently designing.

CLOSING TAKEAWAY

The Magic Circle’s rejection of D4YRL was, on one level, a minor institutional curiosity. On another level, it was an organisation making exactly the kind of deliberate choice that O’Connor argues we need more of — asking not whether the robot can perform the task, but whether the performance of the task by the robot produces the same thing that the task was for in the first place. The answer, for the Magic Circle, was no. Not because the tricks were inadequate, but because the performance of magic is not separable from the human presence that gives it meaning.

Most organisations deploying AI in the workplace have not asked the equivalent question about their own work. They have asked what AI can do and whether it is efficient and whether it is compliant. They have not asked what the work is for, whether AI’s performance of it produces the same thing, and what happens to the people whose cognitive engagement with that work is the source of the capability AI is being deployed to support. Those questions will not answer themselves. The organisations that ask them deliberately will be better positioned than those that discover the answers in the data, when the data is measuring something they no longer have.

Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net