The junk food economy

There is a third force working against human development in the AI era, and it may be the most intractable of all. It is not about cognition or social dynamics. It is about the basic architecture of human motivation. Humans are systematically terrible at doing what is good for them in the long term when a more comfortable option exists in the present.

Behavioural science calls this present bias, a form of temporal discounting where the perceived value of future rewards drops sharply relative to immediate ones. The marshmallow experiment is the famous illustration, but the phenomenon extends far beyond childhood impulse control. It governs how adults manage their health, finances, relationships, and, critically, their own development. The cost of learning is immediate: effort, discomfort, the ego bruise of being wrong. The reward is distant and uncertain: capability that may pay off months or years from now.

AI collapses this calculation entirely. When a knowledge worker can get a polished draft in seconds, the immediate reward of using AI is enormous: time saved, effort avoided, output produced. The alternative, struggling through the problem independently, building genuine understanding, developing the judgment to know when the AI output is wrong, offers no immediate reward whatsoever. It only pays off later. And the human brain, governed by the limbic system's preference for immediate gratification over the prefrontal cortex's capacity for long-term planning, will choose the easy path almost every time.

This is the junk food economy of knowledge work. Junk food succeeds not because people do not know it is bad for them. It succeeds because the reward is instant and the cost is deferred. AI-generated answers operate on exactly the same dynamic. The person using AI to skip the hard thinking is not making an uninformed choice. They are making a perfectly rational short-term choice that happens to be catastrophic in aggregate. Every instance of cognitive offloading is individually trivial. Accumulated over months and years, it represents a systematic hollowing out of capability.

What makes this especially pernicious is that the damage is invisible in the short term. A person who outsources their thinking to AI today will perform just as well tomorrow. Their output may even improve. The decline in underlying capability only surfaces when the AI is unavailable, when the problem is genuinely novel, when judgment is required under conditions of true uncertainty. By the time the deficit is apparent, the atrophy is deep.

No amount of awareness solves this. Knowing about present bias does not neutralise it. Just as knowing that junk food is unhealthy does not make the salad more appealing at the moment of choice. The intervention has to be structural, not informational. It has to change the environment in which the choice is made, not lecture the person making the choice.