Eric C. So, W. W. Norton
US: out now; UK: out 20 November

It was the type of question that Eric C. So had posed to his class for years. As a professor of behavioural economics at the Massachusetts Institute of Technology, So liked to ask his students what impact they believed technology was having on our way of thinking.

But this time, before he could even finish asking the question, a student raised her hand. “I just ChatGPT’d it,” she said, before reading out the answer the AI chatbot had given. When So asked her what she thought, the student seemed genuinely puzzled. ChatGPT’s answer was now what she thought.

“She couldn’t distinguish between her own reasoning and AI-generated reasoning. Or perhaps she could. But she didn’t see why the distinction mattered,” So writes in The Collision: What AI does to us, which explores its impact on our brains.

It is the sort of anecdote people are hearing – and worrying about – around the world. Are we embracing a technology that will see us all become like this student, outsourcing our thinking to AI without even realising?

So certainly seems to think so. uses research from the fields of behavioural science and neuroscience to argue that AI, especially generative AI tools like chatbots, can cause many of us to suffer from “cognitive atrophy”. This is the claim that intellect is like a muscle, with our brains rewiring neural pathways around our habits and pruning skills we no longer have use for. If we don’t use our independent thinking skills, they will deteriorate, while the “cognitive reflex” to use AI for anything mentally challenging will strengthen, So argues.

Eric C. So proposes the interesting idea that we have evolved over millennia to find the lure of AI irresistible

He points to a study that revealed how people learning to juggle over three months showed visible tissue growth in the regions of the brain designed to track movements, but after three further months of not practising, this brain tissue then regressed. In other words, use it or lose it.

You can see the problems that could cause, especially in children. “When we hand students AI tools before they’ve built their own cognitive foundations, we may be preventing those essential connections from ever forming properly,” So warns. The same risks apply to the skills we usually build up over a career. Within a generation, for example, we could have doctors “whose brains never developed the architecture for independent diagnosis”.

So was inspired to examine what AI was doing to our minds after he found himself asking ChatGPT to finish a simple email he was writing to a colleague and realised “some fundamental relationship between me and my own thinking had shifted”.

The danger of writing a book on such a premise is that it becomes tempting to selectively pick the arguments that support it. So falls into that trap, relying on a few small studies to do the heavy lifting and he doesn’t feature any interviews or case studies. I would have loved to have seen some insight from neuroscientists on the latest thinking around what AI is doing to our brains.

One of the small studies So describes measured the neurological activity of three groups of students writing an essay, with one set allowed to use only their mind, another, search engines and the final group ChatGPT. While completing the task, the unassisted writers’ brains were “alight with widespread, robust connections”, but the brains of the ChatGPT group had “gone quiet”.

Straight after finishing, participants were asked to quote any sentence from their essay. For those who wrote unassisted or used a search engine, recall was nearly flawless. But 83 per cent of the ChatGPT users struggled to recall any direct quote at all.

“The information was never encoded,” So writes. This, he argues, shows how we are in danger of creating a world of “hollow competence”, with a generation that can “produce without comprehending”. But the book doesn’t mention that the study was of just 54 participants, 18 per group, and they were given only 20 minutes to write the essay.

never provides quite enough solid scientific evidence to back up its bold claims. Still, it is a well-argued and thought-provoking read. The book points to studies that show how AI can warp our perception of how good we are at something, essentially mistaking its knowledge for our own. Yet because AI makes us feel more capable, we become even more inclined to rely on it, further eroding our thinking skills.

So also proposes the interesting idea that we have evolved over millennia to find the lure of AI almost irresistible, because our brains use up a lot of the body’s energy on thinking and so a preference for “cognitive ease” has become encoded in our very nature. This makes it “extraordinarily difficult for us to resist off-loading our thinking to AI”, So states. “Our biology practically demands it.”

Again though, the author pushes this argument slightly too far by claiming the world is experiencing something he calls “AI gravity” – a force acting on us constantly to use AI as much as possible, unless we actively work against it. That would be an apt description of the smartphone, but it still feels some way off for AI. At times, it felt like this book was written for the small subset of people who are already completely addicted to chatbots. At one point, the author asks: “If AI vanished tomorrow, what would you be capable of doing? Most people don’t know until they try.” Really? I think most people know they would be completely fine.

So posits that the key factor to working with AI successfully is implementing friction into the process. He points to studies that suggest when people have to work through prompts before they get an answer, then AI as a “cognitive partner” can extend our intellectual capabilities.

Otherwise, our intellects could be in trouble, the book warns. “You won’t notice the moment you stop thinking for yourself. No bells will ring, no alarms will sound. The slow surrender will be… almost imperceptible.”

Here’s hoping, for all our brains’ sake, we don’t let that happen.

When you make a purchase via the links on this page, we receive a commission.