Questions and Answers: AI, your brain, and society are colliding.

Meta description: MIT Sloan Professor Eric So warns that heavy AI use can weaken thinking and offers four principles for keeping learning, memory and judgment intact.

Tags: AI use, cognitive atrophy, Eric So, MIT Sloan, productive struggle

Featured image: abstract illustration of a glowing AI orb and a quiet human desk, with linked nodes suggesting memory and learning, no people, no logos, no text

As AI use becomes part of everyday work and study, MIT Sloan Professor Eric So has a new book arguing that the technology can quietly change how people think. So, the Sloan Distinguished Professor of Global Economics and Behavioral Science in the MIT Sloan School of Management, sets out the case in “The Collision: What AI Does to Us,” published this month by W.W. Norton. The book is aimed at students, professionals and organizations that use AI tools, and it offers a clear takeaway: AI can help, but relying on it too broadly can erode the mental habits that make people capable.

So does not deny that AI can finish work faster or serve as a useful domain-specific tool. His concern is the cost of overreliance. He argues that heavy AI use can lead to a kind of cognitive atrophy, meaning people may fail to develop and keep essential thinking skills. Evidence, he says, generally shows that learning improves when people wrestle with material, work through early failures, keep iterating, and in doing so build better thinking while retaining more knowledge.

The argument draws on recent research from So and others in academia and industry, including analysis from AI firms themselves. One example he cites is an Anthropic study showing that about half of queries from college students were low-effort queries for simple answers. So also brings in his own experience studying financial markets and behavior, where he has seen how new tools and information can change human decisions.

Why people keep reaching for AI

The book opens with a personal episode. So describes receiving a colleague’s request for help, writing half a sentence himself, then asking ChatGPT to finish the response professionally. After a few light edits, he sent it, then felt concerned because he had removed his own thinking from the process.

That story becomes a broader pattern, in his view. Students may outsource thinking to AI instead of working through homework that would help them master a subject. But So says the problem is not only individual choice. There are converging forces pulling people toward AI. He points to a biological instinct to conserve mental energy, plus social pressure when others use AI to move faster. That creates a ratcheting effect: as more people use AI, others feel more pressure to use it too. Add the difficulty of detecting AI use in practical settings, and So calls the result “AI gravity,” a constant pull to use the technology.

He says many organizations now do more than suggest AI. Use of the tools can become part of evaluation, meaning a bonus or promotion depends on using them. In academia, students may think a future employer will expect AI use, so they start relying on it now. The danger, he argues, is a self-fulfilling prophecy: using AI too early can weaken the very capabilities people are trying to build, making them more dependent on AI later.

What AI may take from learning and memory

So’s central point is that AI changes not only what people can do, but who they are as thinkers. He says AI removes much of the friction needed to build foundational mastery. Instead of pushing through a problem, users get instant relief, and that relief can become familiar. He also says that when many people turn to AI as their main access point for information, it can affect their ability to remember.

The book also discusses a psychological feature of AI models: they are described as sycophantic, meaning they tend to make users feel validated because engagement and positive feedback can drive their behavior. At the same time, they can be persuasive. So argues that people may lose the ability to connect with others because they become used to a frictionless environment where the model always agrees.

Four principles for keeping thinking alive

To counter the pull, So proposes four principles. The first is embracing the struggle. He reframes struggle as formative rather than something to be eliminated, comparing the shift to what happened after industrialization, when people no longer needed physical labor to be economically productive and had to choose exercise deliberately. He says cognitive effort may need a similar reframing.

The second principle is to value who you are without AI. So points to polished presentations made with Claude where the speaker has not done the work needed to understand the material. He uses the phrase “inseparable task bundle” to say that the process of making slides, for example, creates the ability to discuss the substance. Without that process, a person may get the product but not the capacity to use it well.

The third principle is reinvesting cognitive surplus. So notes that AI can save time on rote tasks, but the real question is what people do with the saved time. If they simply funnel the same energy into the same tasks, they miss the transformation potential. He suggests freed time can be used to build expertise, stretch capabilities, or invest in family, learning, and other skills.

The fourth principle is using AI as a cognitive trainer, not an answer machine. So says AI can be bad for thinking when used as a source of instant answers, but can be great for learning when used as a tutor. That requires seeking friction rather than avoiding it, and it requires motivation.

What the warning means for readers

So’s book is not a call to reject AI. It is a call to notice what the technology is doing to memory, learning and relationships. For readers who use AI for homework, professional messages, presentations or daily work, the message is practical: keep a part of the work human enough to build mastery. So calls this a very challenging moment, and he says the problem needs to be taken seriously. The book’s core challenge is simple to state and hard to apply: use the tool, but do not let it remove the thinking that makes the tool useful.