Every powerful tool raises the same question: does it make us more capable, or does it quietly do our thinking until we can no longer do it ourselves? The worry is old, and the history of that worry is a good place to start.

An ancient complaint

In Plato's Phaedrus, written in the fourth century BC, Socrates tells a story about the Egyptian god who invented writing and offered it to a king. The king was unimpressed. In Benjamin Jowett's translation he warns that "this discovery of yours will create forgetfulness in the learners' souls, because they will not use their memories; they will trust to the external written characters and not remember of themselves." Socrates feared that writing would weaken memory as people relied on written marks, giving them the look of wisdom without the substance.

Was he wrong? Partly. Writing made law, science and history possible, carrying thought far beyond any single memory. But he was also partly right: literate people rarely memorize long poems the way people in oral cultures did. Every tool trades one skill for another. The real question is whether the trade is a good one, and who gets to choose it.

Offloading

Psychologists call this cognitive offloading: using outside tools to reduce mental effort, from shopping lists to the contacts in a phone. It is often wise. In a 2011 study, Betsy Sparrow and colleagues reported that when people expected to look facts up later, they remembered where to find the facts better than the facts themselves. Some later attempts to repeat parts of that study found weaker effects, so its size is debated, but the general pattern of offloading is well established.

Automation bias

The darker side appears when people stop checking. Automation bias is the tendency to favor automated suggestions even when they are wrong. Researchers including Kathleen Mosier and Linda Skitka, studying pilots and others in the 1990s, described two forms. In an error of commission, a person follows bad advice. In an error of omission, a person misses a problem because the system did not flag it, having come to treat the machine's silence as a sign that all is well.

In 1983 the engineering psychologist Lisanne Bainbridge published a short paper titled "Ironies of Automation." Her central point, in paraphrase, was that automation leaves people the hardest tasks while eroding the skill to do them. The machine handles routine work, so the human gets less practice, and is then called in only when something unusual goes wrong. Air France Flight 447, which crashed into the Atlantic in 2009 with 228 people aboard, is often cited. When ice blocked its speed sensors and the autopilot switched off, the crew struggled with hand flying at high altitude, something pilots rarely practiced.

New evidence, still early

Studies of AI tools are young and often surprising. In a 2025 trial by the research group METR, sixteen experienced open-source programmers worked on real tasks in their own projects, with AI tools allowed on some tasks and not on others. The developers were slower with AI, by about 19 percent, though they believed it had made them faster, by about 20 percent. Another 2025 study, from Poland, looked at doctors who had grown used to an AI helper during colonoscopies. When they worked without it, their detection of precancerous growths had fallen, from about 28 percent of procedures to about 22 percent. Both are single studies that need repeating, but they point the same way: our sense of how much a tool helps can be wrong, and skills can fade quickly.

The case for augmentation

None of this means tools must make us weaker. After the chess champion Garry Kasparov lost to IBM's Deep Blue in 1997, he promoted games in which people played with computer help. In a 2005 online freestyle tournament open to any mix of people and machines, the winners were two American amateurs using three ordinary computers. These amateurs who used computers well beat stronger players and stronger machines, because they had learned when to trust each program and when to question it.

The lesson is to be deliberate: offload what you do not need to keep, protect the skills you will need when the tool fails, and keep enough expertise to judge what it gives you.

Worked example: the new nurse

Amara is a new nurse on a ward that uses software to suggest medicine doses and flag drug interactions. In her first month she notices that she has stopped working out doses herself, since the screen always shows a number. One night the system fails to warn about two drugs, prescribed by different doctors, that should not be combined. A senior nurse catches it from experience. It was an error of omission waiting to happen.

Amara settles on a routine. For high-risk drugs she works out the dose herself before looking at the suggestion, and treats any disagreement as a reason to stop and check. For patients on many medicines she reviews interactions even when the software stays quiet. The software still saves her time on routine work, but she now uses it as a second opinion rather than a first one.