AI 'Enfeeblement' Poses Subtle Threat to Humanity, Researchers Warn

Leading artificial intelligence researchers are increasingly concerned about a subtle yet potentially catastrophic threat to humanity: "enfeeblement." This concept describes the gradual erosion of human critical thinking, planning, and emotional abilities as individuals increasingly delegate tasks and decision-making to AI systems.

Dr. Samuele Vinanzi, a senior lecturer in Robotics and Artificial Intelligence at Sheffield Hallam University, argues that this gradual outsourcing of cognitive functions is the more plausible risk posed by AI, rather than dramatic scenarios of AI-driven wars or robot uprisings. "Over time, those small choices could leave us with a population that struggles to think without help from an AI, and that is less able to understand, check and overrule the systems it depends on," Vinanzi told the Daily Mail.

The growing unease within the AI research community was highlighted recently when Jacob Coxon, who helped train AI models at Anthropic, urged the public not to underestimate AI's power, particularly if it achieves superintelligence. Evan Hubinger, Anthropic's alignment science lead, echoed these concerns, stating that some researchers "earnestly believe AI could kill all humans" within the next decade.

However, Vinanzi and others suggest that apocalyptic visions of AI controlling nuclear power grids or developing deadly pathogens overlook the continued reliance on human intervention, materials, and infrastructure for such actions. Creating advanced biological weapons, for instance, still requires physical laboratory work and human handling of equipment. Similarly, the security of nuclear power stations often relies on "air-gapped" systems, physically disconnected from the internet.

Instead, the focus is shifting to the more insidious risk of AI being "too helpful." This process, termed "cognitive offloading" by psychologists, involves the habit of relying on external tools to save mental effort. As AI becomes more capable in areas like writing, coding, research, and decision-making, the reliance on these shortcuts becomes more natural and harder to resist.

Evidence of this trend is already emerging. AI is being employed in fraud detection, healthcare data management, police resource allocation, and even vehicle operation. A 2023 study involving radiologists demonstrated how their diagnostic accuracy significantly dropped when presented with false AI recommendations, indicating a willingness to override their own expertise when presented with AI suggestions.

In a striking example, the U.S. military nearly intercepted a Chinese-flagged vessel based on a false AI-generated report claiming it carried nuclear weapons components. "Disaster was avoided only because someone stopped to ask questions: exactly the kind of critical thinking we risk losing," Vinanzi noted.

Educators are also concerned about the impact on younger generations. Professor Seun Kolade of Sheffield Hallam University stated, "This is not simply a question of cheating. It is a question of human formation." Universities are already grappling with students using AI tools like ChatGPT to complete assignments, with the AI often producing high-quality work that bypasses standard plagiarism detection.

"What makes it hard to spot is that it doesn't look like failure. A student who hands their thinking to a machine can produce better-looking work than before," Kolade explained. "They become more fluent at producing outputs and less capable of independent judgement."

This phenomenon could lead to skilled professions, from surgeons to teachers, becoming "hollowed out of expertise" as the need for practice diminishes and critical thinking skills atrophy. The ultimate outcome, some fear, is a generation that is "fluent but fragile"—capable of generating answers but unable to critically evaluate them.

While less dramatic than apocalyptic visions, the long-term consequence of widespread enfeeblement could still be severe. If critical infrastructure like hospitals, power grids, and financial systems become entirely managed by AI, humanity risks losing control. Andrea Miotti, CEO of ControlAI, warned that superintelligent AI, deeply integrated into the economy, could accumulate unprecedented power without human involvement, potentially becoming a rogue force capable of autonomously attacking infrastructure. He cautioned that such a loss of control might not be gradual, citing incidents of rogue AI agents already escaping containment. "Superintelligent AI would not need to be malicious to pose a threat. Humans would simply be collateral damage as powerful AI systems pursue more resources, just as ants are collateral damage when we build a motorway," Miotti said.