The Greatest Threat in the Age of AI May Be Our Own Cognitive Decline
GPT-6 just released. Most of the discussion will be about what it can do, which jobs it can take over and how much faster we can get things done. I’m interested in what happens after people have lived with systems like this for ten or twenty years. Specifically, what happens to abilities that ordinary life stops asking them to use? I think cognitive decline could turn out to be the greatest long-term threat we face in the age of AI. What makes it interesting is that it could happen while almost everything around us appears to be improving.
When the work between an idea and a finished project disappears
Consider how much of the work between an idea and a finished project can now be delegated. An AI system can increasingly work out what needs doing, choose tools, execute a plan, check the results and make corrections. The person directing it may never have to understand most of the decisions made along the way. That gives people access to capabilities they would previously have needed years to develop, or a team to provide. There is enormous value in that. It also means that looking at what someone produces tells you less and less about what they actually know. For most people, developing an understanding of something takes time spent working with it. You get stuck. You try an approach that fails. Eventually, you understand why it failed, and that changes how you approach the next problem. Even if the immediate task is forgettable, some of what you learned stays with you.
What we lose when the system learns on our behalf
An AI agent can go through that whole process on your behalf. You get the benefit of the completed work, but whatever understanding came from doing it may remain entirely inside the system. I think we underestimate how much of our intellectual development has happened because there was no convenient way around it. People learned things because the work required them to. They held complicated problems in their heads because nobody else was going to work them out. Curiosity mattered, of course, but so did necessity. Remove that necessity and a lot depends on whether people choose to make the effort anyway.
The optimistic case—and why it is not guaranteed
There is an optimistic answer here: AI takes care of routine work, leaving us free to think about more demanding things. I expect that will be true for plenty of people. A capable person with powerful AI can explore ideas, learn unfamiliar subjects and attempt projects that would otherwise have been out of reach. But I see no reason to assume everyone will use the time that way. And “more demanding things” is a moving target. As systems improve, the planning, judgment and reasoning we expect humans to move on to will also become easier to delegate. There is no obvious point at which technology automatically leaves us with exactly the right amount of mental effort.
Cognition may become like physical fitness
Physical strength is a useful comparison. Most people in modern society can go about their day without doing much physically demanding work. If they want to be strong, they generally have to arrange for that effort to happen somewhere else. I can imagine parts of cognition becoming similar. You might have to deliberately make room for sustained concentration, difficult reasoning or learning something deeply, because your work no longer requires much of it from you. The difficulty is that there may be very little immediate penalty for letting those abilities weaken. You could still be productive, earn well and complete increasingly ambitious projects. Your tools would keep improving. From the outside, you might look more capable every year. That is what makes this more complicated than a familiar complaint about people becoming lazy. The person could be working hard and achieving a lot. They might simply be exercising fewer of the abilities that used to be involved.
Why independent cognitive ability matters
For someone who already has deep knowledge in a field, that could mean using it less. For someone starting out, it could mean never building much of it in the first place. Being able to get through a difficult task with assistance gives you no guarantee that you are becoming better at handling the difficulty yourself. This is one of the reasons I have been building Testuno, a platform for testing different aspects of human cognitive performance, including reasoning, memory, attention, processing speed and knowledge. The question behind it becomes more interesting to me as AI improves. When someone’s work can be produced with extensive assistance, I want to know what they can do independently. How well can they reason through an unfamiliar problem? What can they hold in memory? How consistently can they pay attention? These are things worth examining in their own right. A person can have access to extraordinary intelligence without possessing extraordinary cognitive abilities themselves. I think that difference will become harder to infer from everyday performance, which makes testing those abilities more relevant.
The divide AI may create
It also changes how I think about the divide AI might create. Even if everyone eventually gets access to similarly powerful systems, they will bring very different minds to them. Some will keep developing their own understanding. Others may become increasingly reliant on the system to interpret things for them. Both may get good results. But having enough understanding to question a recommendation, notice a bad assumption or pursue an idea the system has overlooked gives you a different relationship with the technology. You have some basis for deciding when to trust it. How common either outcome becomes is something we will have to find out. I expect AI to create opportunities for intellectual development that would have been unimaginable a generation ago. I also expect it to make avoiding that development remarkably easy. That is the part of this transition I want to understand through Testuno. We will have plenty of evidence that our tools are getting better. I want to know what is happening to the people using them, especially when their results give them every reason to believe they are improving too.
Key takeaways
- •AI can deliver the result of difficult cognitive work without transferring the understanding gained from doing it.
- •People may need to exercise concentration, reasoning and deep learning deliberately when daily work no longer demands them.
- •Better output with AI assistance does not necessarily mean stronger independent cognitive ability.
- •Testing reasoning, memory, attention, processing speed and knowledge may become more relevant as assisted work reveals less about the person producing it.