Nvidia's Jensen Huang accepts lost math skills as AI's price

The evidence behind the exchange

Klein raised a Chinese study of 26,000 students who began using AI at different times. That timing allowed researchers to compare students who adopted AI earlier with those who delayed using it.

The students who used AI completed assignments faster. But the study also found convincing signs that cognitive skills needed for schoolwork were gradually disappearing. Over the long term, those students received lower exam scores.

Huang agreed that students’ cognitive abilities were declining. He pointed to skills that many people would consider basic:

Then he questioned whether losing those abilities really mattered. Klein said that was precisely what he wanted to understand.

Huang’s answer was that the loss probably did not matter. New abilities would emerge in the future, he said, even if they were not the abilities people use today.

An executive’s personal example

Huang then offered himself as evidence that people can live without some basic knowledge. He said he does not know his own address. Several years ago, he stopped to refuel his car and panicked when asked for his postal code because he could not remember it.

For the same reason, he does not know his phone number and sometimes forgets similar details. Huang said he can live with that.

His broader position is that people may lose some subtle intellectual abilities while becoming better at systems thinking. An AI assistant, in this future, would remember ordinary details such as a person’s home address.

The bargain is more concrete than superintelligence

The usual AI safety conversation focuses on a hypothetical system that escapes human control. Huang’s exchange with Klein points to a less dramatic form of dependence: people gradually surrendering the small acts of memory and reasoning that make independent thinking possible.

The trade is not between effort and convenience; it is between kinds of ability.

I think that is what makes Huang’s answer more revealing than reassuring. He did not dispute the study’s warning or deny that students are losing skills. He simply treated those losses as manageable because AI may provide different capabilities in return.

That argument also changes the standard for harm. If an assistant remembers an address, a phone number or the result of a square root, the immediate convenience is obvious. The harder problem is deciding whether the skill has disappeared because it became unnecessary, or because people stopped exercising it before they had developed a replacement.

The future Huang describes does not require anyone to remember even their own phone number, as long as an AI assistant remembers it for them. That may be a workable arrangement, but it makes human competence dependent on the systems that are replacing it.