Researchers at the Massachusetts Institute of Technology have quantified a phenomenon educators and managers are only beginning to name. When young people delegate writing tasks to generative models such as ChatGPT, their brains show up to 55 per cent less connectivity in the prefrontal cortex than during unassisted reflection, according to work led by Nataliya Kosmyna of the MIT Media Lab. The finding gives a neural correlate to what cognitive scientists call “cognitive debt”—the gap between what a student produces with AI and what they can understand, memorise or create independently. Over time, Kosmyna warns, that debt can harden into cognitive atrophy, eroding analytical capacity and memory.

This pattern of offloading is not confined to schools. Canadian survey data from Employment Hero shows 43 per cent of workers feel guilty using AI, 39 per cent consider it cheating, and 34 per cent hide their use from employers. Educators like David Williams describe cognitive offloading as their “number one concern,” noting that circumventing one’s own thinking undermines learning. The secrecy, researchers argue, blocks the very conversations needed to teach safe and responsible use. In Iran, epidemiologist Hamid Souri has observed patients arriving at clinics with AI-generated diagnoses, a trend he links to high treatment costs that push people toward self-medication. He cautions that AI cannot replace clinical examination and that reliance on it risks disrupting the doctor-patient relationship and worsening outcomes.

In the labour market, the effects are structural. An OECD report notes that AI is eliminating specific tasks—report-writing, data compilation—rather than whole jobs, but the burden falls unevenly. PwC’s global AI employment survey finds that entry-level roles in highly exposed fields are now seven times more likely to demand advanced skills such as judgement and leadership, effectively raising the first rung of the career ladder. Boston College’s Center for Retirement Research reports that workers over 55 in high-exposure roles are leaving the workforce at elevated rates, choosing early retirement over the strain of adapting to AI-driven workflows. Microsoft’s recent cut of 4,800 positions, while not a direct substitution, reflects how task automation compresses training pathways and middle-management functions.

Hiring processes are buckling under the strain. Harvard Business Review research indicates that AI-written résumés and real-time interview prompts are making early recruitment filters unreliable. Nearly half of job applicants now use AI somewhere in their search, and 13 per cent admit to using chatbots live during interviews, according to Gartner. The gap between polished applications and genuine competence is widening, pushing firms like Google, L’Oréal and Anthropic to mandate in-person interviews or declare AI-free zones. The financial stakes are measurable: the Society for Human Resource Management puts average hiring costs at $5,475 for non-executive roles, with replacement costs reaching 1.5 to two times annual salary.

Viewed from Washington, the US Bureau of Labor Statistics projects a 6 per cent decline in computer programmer positions over the next decade even as demand for AI-related roles grows, a bifurcation that mirrors the broader reallocation of cognitive work. The next milestone to watch is not a single regulatory step but the speed at which institutions—schools, hospitals, corporations—build AI literacy frameworks that address dependency before it becomes structural. Without them, the data suggests, the hidden costs of cognitive offloading will compound.