AI tools are extraordinary, and the efficiency gains are real. But we are seeing serious unintended consequences for human capability and cognition.
Call it the hidden cost of the AI era: the more we delegate memory, work, and thinking to machines, the less we develop the capabilities needed to add meaningful value and deliver the kind of thinking machines cannot. It is happening pervasively, below the threshold of any dashboard or performance review. And by the time it becomes visible, it is very hard to reverse.
The Ongoing Externalization of the Mind
Research suggests that cognitive performance has begun to decline in some geographies. The Flynn effect—the 2- to 3-point-per-decade rise in IQ scores that held throughout much of the 20th century—has slowed, stalled, or gone into reverse in countries like Norway, the UK, and the US. On some measures of cognitive performance, younger generations may no longer outperform the generations that preceded them.
The decline began before AI. When we handed memory to search engines in the early 2000s, the transaction felt like pure gain as we freed mental capacity for higher-value thinking. But external memory comes with a cost. Pattern recognition depends on having knowledge readily available to draw connections across ideas. When that knowledge must be continually retrieved from outside the mind, it impairs our ability to reason and establish relational connections at speed.
Research by Columbia psychologist Betsy Sparrow, published in Science, showed that people stop encoding information at the moment they believe they can look it up later. The offloading happens automatically, below conscious intention. A 2024 meta-analysis in Frontiers in Public Health confirmed the self-reinforcing loop: the more you search, the less you store; and the less you store, the more you must search.
With the rise of AI, we are externalizing critical thinking itself. According to preliminary findings of a 2025 MIT Media Lab study, people writing with an LLM showed significantly lower neural connectivity associated with memory, creativity, and attention than those using search engines or no tools. Most strikingly, 83% could not accurately quote from their own essays minutes later. The researchers called this “cognitive debt”: borrowing against a future self that is measurably less capable than the one doing the borrowing.
The consequences extend beyond individual cognition. As AI becomes embedded in everyday work, cognitive offloading can accumulate into a collective erosion of skills—weakening the organizational intelligence and resilience those individual capabilities support. One BCG study of C-suite leaders and senior executives found that half are already observing de-skilling in their organizations, and more than 60% believe that de-skilling will pose a material threat within the next three to five years.
Research by BCG and Harvard Business School makes the stakes concrete. AI can be a powerful equalizer: lower-performing professionals improve dramatically when equipped with it. Yet the same research shows that AI’s benefits depend on how it is used. When professionals relied on persuasive AI outputs without critically evaluating them, performance deteriorated. AI increased speed, but not judgment.
The critical problem is that universal reliance on these tools removes opportunities for people to acquire the skills required to add value as humans in the loop. Left unchecked, that creates a doom spiral of capability erosion.
The Attention Collapse
Alongside the externalization of memory and thought, something else is deteriorating: our capacity for sustained attention.
UC Irvine professor Gloria Mark found that the average time people spent on a screen before switching tasks fell from two and a half minutes in 2003 to just 47 seconds in 2023. At the same time, people have quietly stopped reading for pleasure. The American Academy of Arts and Sciences found that 84% of Americans now read for five minutes or less per day. Only 9.5% of young people aged 15 to 24 read for more than 20 minutes daily. The National Endowment for the Arts reports that literary reading fell from 45.2% of adults in 2012 to 37.6% in 2022, marking the lowest level in 30 years of survey data.
This matters because reading does more than transmit information. Long-form reading is the primary exercise through which humans build abstract reasoning, analogical thinking, and the capacity to hold complex multi-step arguments in mind. Instead, we have passive video consumption: algorithmically optimized, attention-engineered, and designed to be watched rather than processed. The brain that reads and the brain that watches are having meaningfully different experiences at the neurological level. One is exercising. The other is resting.
Human in the Loop Requires an Experience Curve
The standard organizational response to AI risks is to keep humans in the loop: a person reviews the output, catches the errors, and provides the judgment. It is a reasonable design principle, but it contains a flawed assumption: that the people providing oversight have the knowledge, attention, and confidence required to intervene.
Consider what happens in highly automated environments. When systems perform reliably for long periods, human operators can get fewer opportunities to practice the difficult judgments required when something goes wrong. In a crisis, they may still be formally “in the loop,” but the system is relying on capabilities that have not been exercised often enough to remain sharp.
The same risk now applies to every organization building AI-assisted workflows. A system that generates strategy recommendations, financial analyses, legal interpretations, or medical diagnoses is only safe if the human reviewing it is genuinely capable.
Capability alone is not enough. Cognitive capacity is also essential for sustaining attention and judgment, but it can be undermined by the need for intensive oversight of AI. Recent BCG research found that supervising AI can create cognitive overload: workers who had to closely monitor and evaluate AI outputs reported 14% more mental effort, 12% more mental fatigue, and 19% more information overload than those with less-intensive oversight responsibilities.
Effective human oversight requires deep domain knowledge, critical thinking, and contextual judgment to detect hallucinations, challenge assumptions, and recognize what AI misses. Without these practiced capabilities—and the independence, cognitive capacity, and authority to challenge the machine—human oversight becomes little more than procedural reassurance.
Four Ways to Reverse the Decline
The solution is not to resist the technology. It is to ensure the humans working alongside it remain genuinely capable of providing oversight and adding value. For C-suite leaders and boards, four recommendations stand out:
Provide the repetitions that build expertise. The traditional organizational model assumed that people would build judgment through repeated exposure to increasingly difficult cognitive work. AI is removing many of the routine tasks that once created those repetitions. The experience curve still exists, but organizations can no longer assume that people will climb it simply by doing the job. Leaders need to deliberately create opportunities for employees to wrestle with difficult problems, make decisions, receive feedback, and learn from the consequences. The goal is not to preserve work that AI can do better, but to preserve the experiences through which humans develop the judgment and expertise needed to add value alongside it.
Reinvent learning around immersion, not instruction. Here is an awkward truth: the generation supposedly unable to focus can sustain intense concentration for hours inside a video game. The challenge is to bring the same high levels of engagement, entertainment, and rewards to classroom and corporate learning. That means requiring decisions, iteration, and feedback—with meaningful consequences—rather than passive receipt of information. Learning becomes immersive, participatory, and intrinsically motivating. The answer is not to turn everything into a game, but to understand why games hold attention and then rebuild education and training around the same principles: agency, challenge calibrated to skill, immediate feedback, and stakes that feel real.
Move up to the value that is genuinely hard to replicate. The human advantage is not a fixed position; it is a rapidly shifting frontier. Rote work is increasingly automated. Pattern-matching work is following suit. What remains defensible is a narrower and more demanding set of capabilities: genuine empathy and human connection that builds the trust no algorithm can manufacture; strategic risk-taking that bets on non-obvious futures and tolerates the ambiguity that optimization models cannot; and innovation that is truly new to the world—the creative leap that connects domains in ways no training set predicted. These are not soft capabilities. They are the toughest ones to build, the most difficult to measure, and the hardest for AI to replicate. This is precisely because they require a fully exercised, richly experienced human mind as their foundation.
Double down on the physical world. The most undervalued leadership capability right now may also be the most irreplaceable: the embodied intelligence that comes from being physically present in the world. The negotiator reading a room, the leader who knows when something is wrong before the data says so, or the executive who builds trust through presence rather than output. These are not anachronisms. They are examples of human advantage that become more valuable as more cognitive work migrates to machines. Organizations that invest in physical world mastery and embodied judgment are not being romantic about the past. They are building capability in the one domain AI cannot enter.
AI is reshaping more than the way work gets done; it is reshaping how human capability is developed. That creates a foundational risk to humanity’s ability to add distinctive value and create meaningful advantage. To keep pace with these changes, every leader must rethink how they develop the next generation’s judgment, originality, and human connection. We must preserve the human capabilities needed not only to give the technology its direction, but also to set its limits.
Subscribe to our new The Value Lens from BCG newsletter to read more articles like this one.
