Artificial intelligence now writes essays, generates code, summarizes research and drafts plans in seconds. As it takes on work that once required years of training, a basic question has moved from the margins of research to the center of public debate: is AI strengthening human abilities or slowly weaning people off using their own minds?
Researchers call the underlying behavior cognitive offloading, the handing over of part of one’s memory, analysis or decision making to an external system. The habit is old. People have leaned on books, calendars, calculators and maps for centuries. The difference, as recent coverage in Scientific American and elsewhere has noted, is that today’s AI systems do not just store information. They generate answers, offer advice, steer decisions and often sound as confident as a human expert.
Three voices on the same problem
The question was the subject of a panel on the BBC program AI Decoded, hosted by the broadcaster’s AI correspondent Marc Cieslak. The guests were Owase Jeelani, a pediatric neurosurgeon; Tapani Rinta-Kahila, a researcher who studies how digital technology affects human work and expertise; and Stephanie Hare, an author on technology and ethics.
Jeelani argued that AI can be a powerful aid to human capability if used at the right level and for the right purpose. He noted that discussion tends to focus on the risks of AI while overlooking the weaknesses of human thinking: bias, limited information, poor estimates and opaque judgments. Used well, he said, AI can reduce some of those weaknesses in narrowly defined tasks.
He suggested the technology might be better described as augmentative intelligence than artificial intelligence, because its best role is to extend a person’s ability rather than replace it. In his view, the best doctors of the future will be those who learn to use AI as a supporting tool while keeping their core clinical skill, creative judgment and responsibility for decisions.
Three routes to deskilling
Rinta-Kahila described three distinct ways AI can erode skill. The first affects experienced professionals who come to rely so heavily on automation that their own abilities atrophy over time. The second affects newcomers who train from the start with tools that do the hard parts for them, and who therefore never reach real depth in the craft.
The third operates at the level of organizations. Companies simplify work so that less experienced staff can do it, which lowers labor costs in the short term but can reduce the value placed on high level expertise over time. He framed this not as a purely technical change but as a shift in how work, employment and professional development are structured.
In medicine the stakes are higher than in most fields, because a wrong decision affects health and life rather than money alone. A young doctor who leans on AI recommendations from the first day of training may develop differently from earlier generations. Such a doctor may reach information faster and complete complex tasks more easily, but without practice in diagnosing and deciding under pressure, the ability to recognize when the AI is wrong may stay underdeveloped.
AI as a bounded tool, not a decision maker
Jeelani stressed that AI is built on human generated knowledge and human interpretations, so it cannot be treated as infallible. The biases and gaps in human knowledge carry over into the systems trained on it. A doctor, engineer or lawyer cannot accept a recommendation without questioning it.
He gave an example from his own practice. In a complex operation to separate conjoined twins, his team used an AI model for one specific technical problem: estimating how much expansion could be achieved between two bones using particular devices over several weeks. For that bounded question the model was very useful. Planning the overall operation, integrating different sources of information and bearing final responsibility stayed with the human surgeon.
The example points to a working principle. Rather than handing an entire complex task to a machine, AI can be applied to the parts where it offers better speed, calculation or precision than a person, while the overall plan, interpretation of circumstances, handling of surprises and ethical responsibility remain human.
Trust sits at the heart of this. Patients and parents do not choose a surgeon only for technical skill; they entrust that person with a life. If something goes wrong, the surgeon cannot say the AI gave bad advice. The panel also drew a distinction between autonomous cars, which operate in relatively repetitive conditions with measurable rules, and surgery, where every patient and anatomy differs. Social, ethical and legal acceptance of fully automated decisions on the human body will be far harder to earn.
Expertise grows at the edge of ability
Real skill, Jeelani said, develops when people work at the limit of their current capability. He became an expert because he repeatedly faced problems that had not been fully solved before. If AI removes those difficult experiences from future specialists, they may hold a great deal of information without becoming true experts.
The risk is not confined to medicine. Students and employees can now produce polished presentations, detailed reports and professional documents with AI. The work can look excellent while the person behind it struggles to answer a basic question about its content. AI can make someone appear more skilled than they are, while their understanding remains shallow.
Researchers connect this to automation complacency: when an automated system keeps performing well, people trust it more and monitor it less. Aviation offers a long studied parallel. Pilots depend on sophisticated automation, yet are required to keep manual skills sharp because those skills decide the outcome when the technology fails in an unexpected situation.
Education, daily life and business models
In education, the panel raised the question of whether children should learn with AI from an early age or first build reading, writing, math and problem solving skills by traditional means. The view expressed was that a student who outsources every hard question from the start misses the trial, error and struggle that produce understanding. In math, the method matters as much as the answer; in writing, forming ideas and building an argument matter as much as the finished essay.
Some panelists favored strengthening basic literacy, numeracy, creativity and critical thinking first, then introducing AI as a supporting tool. A student who has learned to do a task unaided is better placed to spot and evaluate an AI’s errors.
Jeelani also observed that the world talks a great deal about artificial intelligence and relatively little about understanding organic intelligence. He described the human brain as fundamentally a predictive system that constantly anticipates what comes next based on experience and context, a capacity sharpened by practice, mistakes and real world exposure. Delegating even small decisions to AI reduces the training that system receives.
The discussion also touched on commercial incentives. Hare and others pointed to the razor and blade strategy, in which a core product is cheap or free and revenue comes later from continued use. AI services offered free or with frictionless access can become embedded in work and life, making a subscription easier to accept once people find it hard to function without them. If dependence extends from work into writing, research, emotional advice and relationships, the issue becomes one of mental autonomy rather than a monthly fee.
Practical habits
The panelists did not reject AI. They noted it can extend a person’s reach into areas outside their specialty, such as a researcher using it to decode tax law before verifying against official sources. They offered several habits for preserving independent thinking:
- Think first. Form an initial answer or position before asking the AI, then use it to review, explain or offer alternatives.
- Ask for reasoning, not just results. Request the assumptions behind an answer, the uncertain points and the plausible alternatives.
- Keep practicing core skills. Doctors should keep diagnosing unaided, pilots should fly manually, programmers should read code structure and students should write independently.
- Keep creative ownership. Machines can generate ideas, designs and drafts, but purpose, meaning and priorities should remain human choices.
The conclusion of the discussion was that AI is neither an outright threat to human intelligence nor an automatic benefit. Its effect depends on where it is placed. If it handles repetitive tasks, large volumes of information and specific computational problems while people retain creative thought, ethical judgment, responsibility and foundational skill, it can expand what humans can do. The challenge is to build a relationship with the technology in which convenience does not replace competence and automated answers do not become a permanent substitute for human judgment.
