E-Paper
Artificial intelligence and the friction paradox
The difficulty is that the savings from removing friction are visible at once, while the lost capability surfaces years later. Nobody knows how to measure this
Prefer HTon Google
Those of us who learned to drive before smartphones will remember getting lost. A wrong turn meant stopping to ask a shopkeeper for directions and eventually finding the right road back, and knowing the city a little better for it. Today, Google Maps tells us where to turn, and we arrive knowing almost nothing about the roads we took. Researchers at McGill University found that heavy GPS users had poorer spatial memory when they had to find their way unaided. This trade, an easier time today for a weaker skill tomorrow, is not new. What is new is its scale. And, GPS merely does our navigating, but artificial intelligence (AI) offers to do our thinking and knowledge gathering. This is the friction paradox: The effort we are so eager to remove is often what makes us capable.
The clearest evidence of this comes from a classroom. Researchers at the University of Pennsylvania studied nearly 1,000 high-school students in Turkey learning mathematics. Those who practised with ChatGPT solved far more problems correctly. But in the exam, without the chatbot, they scored 17% worse than classmates who never used AI. Parents understand this instinctively; it is why we don’t do our children’s homework for them.
Evidence shows that beginners gain the most from AI, and also lose the most. In a large study of customer-service agents, AI raised the productivity of the least experienced by about a third while barely helping veterans. Yet, an experienced agent can sense when a customer’s patience is running out and recognise when the standard fix will not work, and escalate early, whereas a novice may not.
The efficient shortcut that is harmless for the expert can hollow out the novice. Socrates warned that writing would ruin memory, and teachers worried that calculators would erode mental arithmetic. We survived both. But a calculator does not need oversight. AI can confidently give wrong answers, and only human judgment catches them. That is the very skill we stop building when the machine does the work. In 2009, the autopilot on Air France Flight 447 switched off over the Atlantic after ice blocked its speed sensors. The pilots, who rarely flew manually at that altitude, made a series of errors, leading to the plane crashing and killing all 228 people on board. Automation removed daily practice, then handed control back to people at the moments that demand the most skill.
Medicine is already seeing this. Polish specialists who had grown used to an AI system for spotting early signs of bowel cancer found fewer such growths when working without it, their detection rate falling from about 28% to 22% within months.
Ask accomplished leaders what shaped them, and they invariably talk about overcoming big challenges such as the project that failed or the year everything went wrong. Warren Bennis called these experiences “crucibles”; overcoming hard challenges shapes character and builds capabilities and resilience. But hardship can break some people even as others grow stronger. The difference is in whether a person has already faced smaller hardships. The worry is that AI will remove the hundreds of small trials that prepare young people for bigger ones. Those small trials usually happen in a first job. India’s IT services industry has been one of the world’s largest apprenticeship systems, where young engineers from small towns wrote routine code, tested other people’s software and slowly learned how real work gets done. That routine work is exactly what AI does best. Stanford economists found that since generative AI arrived, employment of 22-25-year-olds in the most exposed jobs has fallen about 13% relative to less exposed work, while older workers have been largely untouched. Each company that trims junior hiring is acting sensibly. If all of them do, where will the senior people of 2040 come from?
However, not all friction is worth keeping. The weeks that small entrepreneurs lose chasing licences and waiting on inspectors build nothing but frustration, and all of this should go. But the same entrepreneur pitching to a sceptical customer or investor, failing, and then figuring out why is becoming more capable. An AI tool that writes the entrepreneur the perfect pitch would make her first year easier and the later years worse. The test of good friction is whether an effort builds something in the person making it. If not, eliminate it. If yes, think hard before automating it, especially for beginners.
The Turkish study gives hope. Students given a chatbot designed as a tutor that offered hints rather than answers did about as well in the exam as those who hadn’t used AI. The problem is not AI, but how it is set up. Schools should make students attempt a problem before the machine does, and judge learning by what they can do unaided. Boards should ask management, “If we hire fewer juniors, how are you going to grow our future leaders?” And AI companies should design AI as a tutor, not the machine churning out answers, for students and new workers.
The difficulty is that the savings from removing friction are visible at once, while the lost capability surfaces years later and nobody yet knows how to measure this. Unless we choose deliberately, the paradox can hollow out humanity. Our lives will get easier, and we will become less capable.
Ravi Venkatesan is the former chairman of Microsoft India and founder of GAME Global Alliance for Mass Entrepreneurship). The views expressed are personal
