The Cost of Letting AI Do Your Thinking

An AI researcher who works on measuring Gemini’s capabilities reflects on a habit she sees spreading in herself and around her: handing over more and more cognitive work to AI. She draws a line from search engines—which still forced you to decompose a question, weigh sources, and synthesize an answer—to today’s assistants, which deliver finished responses to complex questions in minutes, quietly performing the intermediate reasoning steps that used to be yours to do. The through-line, illustrated with Ken Liu’s 2012 story ‘The Perfect Match’ and a ‘Microphone Man’ at an SF startup event who claims Claude Fable thinks better than he does, is a question of autonomy: who is actually making the decisions that matter in your life?

The author resists a purely alarmist read. She catalogs genuine wins—automating tedious, low-paid drudgery, translating reports, tutoring an MCAT student from scratch, freeing researchers to focus on analysis—and concedes that offloading mundane thinking to reclaim time for harder thinking can raise both productivity and satisfaction. Her worry is the slide from assistant to substitute, and the lazy thinking that convenience invites; her physics-teaching mother already suspects nearly all her online students submit AI-generated assignments that come back near-identical.

The piece lands on a practical distinction rather than a verdict. Some questions merit fast answers; others merit slow ones. Her model example: traveling in Portugal, she and her sister first generated their own hypotheses about why the country still celebrates its colonial-era ‘explorers,’ argued and speculated and drew on half-remembered history, and only then queried AI to test and extend their thinking. The AI confirmed some theories, added others, and missed a few—but the value came from thinking first and consulting second, not outsourcing the reasoning wholesale.