Job Loss by AI? Credit: linkedin.com
Introduction
AI layoffs are looking more and more like corporate fiction, hiding a darker reality, the research institute Oxford Economics suggests. Despite breathless headlines warning of a robot takeover in the workforce, a new research briefing from Oxford casts doubt on the narrative that artificial intelligence is currently causing mass unemployment. According to the firm’s analysis, corporations do not appear to be replacing workers with AI on a significant scale, suggesting instead that they may be using the technology as a cover for routine headcount reductions. While isolated examples of AI-related displacement exist, Oxford Economics found no macroeconomic evidence of a structural employment shift caused by automation. Instead, the researchers pointed to more traditional factors including weaker consumer demand and excessive hiring in the past .“We suspect some firms are trying to dress up layoffs as a good news story rather than bad news,” the report states.
The Problem with Data
In a recent interview, Wharton management professor Peter Capelli told Fortune Magazine that layoffs are widely publicised because stock markets typically celebrate news of job cuts. Firms announce so-called ‘phantom layoffs’ that never actually occur by simultaneously buying and selling the same or similar assets in different markets to profit from tiny price differences, realizing a risk-free profit by exploiting market inefficiencies. When asked about the supposed link between AI and layoffs, Cappelli urges people to look closely at this kind of announcement. The headlines suggest that AI is the cause of layoffs but if one reads what they actually say, it is obvious that they are just hoping that AI is the cause for job loss as this message expresses what investors want to hear. Oxford Economics support their argument about layoffs with a simple analogy: If machines were truly replacing humans at scale, output per remaining worker should skyrocket and as AI-systems were replacing labour at scale, productivity growth should be accelerating. Generally, this clearly is not the case. The report observes that recent productivity growth has actually decelerated, a trend that aligns with cyclical economic behaviours rather than an AI-driven boom. While Oxford Economics acknowledges that productivity gains from new technologies often take years to materialize, the current data suggests that the application of AI remains experimental in nature and is not yet replacing workers on a major scale. This corresponds with what Bank of America head of Research, Savita Subramanian, told Fortune Magazine in August last year. Companies have learned already in the 2020s to generally replace people with processes as productivity measures have not really improved much since 2001, referring to the famous productivity paradox, described by Nobel prize-winning economist Robert Solow: “You can see the computer age everywhere but not in the productivity statistics.”
The briefing from Oxford Economics also addresses fears that AI is eroding entry-level white-collar jobs. While U.S. graduate unemployment rose to a peak of 5.5% in March 2025, Oxford Economics argued this is likely cyclical rather than structural, pointing to a ‘supply glut’ of degree-holders as a more probable cause. The share of 22-to-27-year-olds with university education in the U.S. rose significantly since 2019, with even sharper increases observed in the Eurozone. Ultimately, Oxford Economics concludes that shifts in the labor market are likely to be evolutionary rather than revolutionary.
The Hidden Cost of AI
A new analysis from MIT, titled ‘The Iceberg Index’, suggests that what we are observing represents just the smallest, most visible fraction of a much larger structural shift. The report describes in great detail how much of the U.S. workforce workload could already be performed by current AI systems in a not-distant future scenario as in fact, this shift is happening right now. MIT’s simulation estimates that today’s AI technologies have the technical capacity to perform the equivalent of approximately 11.7% of all U.S. jobs, representing nearly USD 1.2 trillion in annual wages. Only 2.2% of that exposure is in the occupations we typically talk about. The rest sits beneath the surface ,quiet and largely unexamined.
Clinical neuropsychologists study how humans think, work and how they make decisions under pressure. Hence, the MIT analysis is less a story about automation and more a reflection on the cognitive architecture of the modern workplace. It tells us not just what AI can replace, but what humans uniquely contribute and what we risk losing if we are not careful. To fully understand cognitive workload, it is important to define what it means. Cognitive workload can be simply described as the amount of mental effort or resources required to complete a task successfully. It encompasses various cognitive processes such as attention, memory, problem-solving, decision-making and perception. The more demanding a task is, the higher the workload required to perform it effectively. When it comes to cognitive workload, it is not just about the quantity of mental effort, but also the type of tasks engaged as part of the workload. Different tasks can demand varying levels of cognitive load depending on their complexity, novelty and familiarity. For example, solving a complex mathematical equation may require a high cognitive workload due to the need for intense concentration and problem-solving skills. On the other hand, performing a routine task that has become familiar may require less cognitive workload as it requires less effort to complete the task.
What the Iceberg Index Actually Measures
The Iceberg Index goes beyond typical labour forecasts. Instead of projecting future technological change, it maps the current capabilities of AI systems against the current skill composition of about 900 U.S. occupations. It builds a ‘digital twin’ of the workforce and identifies the specific tasks within each job that is technically reproducible by existing AI tools. This distinction matters as it shifts the question from ‘What might AI do someday?’ to ‘What could AI do today, if adopted right now?’. For corporate activities already strained by staffing shortages, burnout and administrative overload, including education and behavioural health, the index highlights where the system is under the greatest tension. One of the most striking findings is where the exposure to change actually takes place: Not in the glamorous tech sectors, not in robotics factories and not among software engineers. It is focused on administrative roles, logistics support, HR, finance, legal assistance, scheduling, documentation, data processing and other forms of cognitive ‘glue work’. These are tasks that keep institutions functioning but rarely show up in headlines. They represent activities characterized by structured decision-making, rule-based judgments and predictable patterns of information flow. These tasks do not depend on deep expertise but on sustained attention, accuracy and working memory, precisely the domains where generative AI systems excel. From a neuropsychological point-of-view, these are tasks that rely heavily on mapping the functionality of the human brain such as maintaining multiple streams of information, updating decisions as new inputs arrive, inhibiting irrelevant details and shifting flexibly between subtasks. AI models are built to do these functions cheaply, tirelessly, and at scale, humans are not. That does not mean the work performed according to these models is unimportant. In fact, the opposite is true. It just means the cognitive cost of performing these tasks is disproportionately high for humans compared to machines as in many work environments these costs are showing up in the form of chronic overload and burnout.
AI and the Future of Work
AI’s engagement with cognitive middle-skill work is an indication of the nature of human expertise. The tasks that are most easily automated are those that rely on executive functions such as organization, monitoring, rule-following, patterning and documentation. The tasks that remain stubbornly human are those grounded in high-context reasoning, emotional intelligence, ethical judgment, rapport and trust. These draw on neural networks mapping the human brain which are shaped not just by algorithms, but by lived experience, historic developments, culture and emotion. Because they are computationally complex and biologically grounded humans are not replaceable. If organizations respond to the Iceberg Index with the idea of cost-cutting, they will miss the point. If they use it to rearchitect work, the result could be a healthier, more sustainable cognitive ecology applied to solve today’s problems. The danger is not that AI will replace us. The danger is that we may continue designing workplaces that burn through and negatively impact human attention as if it were an infinite resource in view of the fact that intelligent machines are much better in executing routine tasks.