How to build a real relationship with AI and keep your cognitive edge

Co-Existence Ethan Mollick WH Allen (UK, 22 October); Portfolio (US, 20 October Maybe I should be embarrassed to admit this, but I have made precious little effort to engage with artificial intelligence. I don’t doubt it will be transformative. That much is already clear. It’s more that I worry about what happens when we outsource our thinking, what I lose personally if relying on AI leads to some sort of cognitive atrophy, and what becomes of us all when we surrender intellectual effort. In his new book Co-Existence: The next phase of AI, Ethan Mollick makes a convincing case that I have been too wary and complacent. The book amounts to a clear-minded manifesto for cautiously embracing AI, a pleasingly practical guide to how each of us can collaborate with it effectively without relinquishing our most valuable cognitive capabilities, and a wide-ranging exploration of what this new relationship might mean. Advertisement A professor at the University of Pennsylvania’s Wharton business school, Mollick writes the wildly successful Substack “One Useful Thing” about the impact of AI. Co-Existence is essentially a sequel to his 2024 book, Co-Intelligence: Living and working with AI, which focused on how to think about the powerful new technologies. Looking at AI today, Mollick is pragmatic from the start. AI is here, whether we like it or not. It is improving rapidly. And it is going to change how we live and work in ways that are hard to predict. The question, then, is not whether some kind of “superintelligence” is ultimately going to kill us all, but what kind of relationship we should have with AI now and in the immediate future. In providing answers, Mollick covers a lot of ground. He explains everything from the jagged frontier – a concept he coined to describe the uneven and unpredictable nature of AI capabilities – to how today’s large language models work and what that means for the way they “think”, including an instructive passage on why reinforcement learning with human feedback makes them so sycophantic. He does it all with impressive clarity and pep, supporting his arguments with insights from his own research on AI and from other leading experts, and leavening it all with examples of his gratifyingly meta conversations with the likes of Claude, Gemini and ChatGPT. One minor criticism is that the book rattles through some of the more unsettling aspects – the appropriation of writers’ work for training data, for example, or the admittedly early research suggesting that AI use might blunt our cognitive abilities. But none of this undermines its main thrust. The primary purpose here is to show us how we can work wisely with collaborators that, as Mollick puts it, are “genuinely powerful and genuinely unreliable”. The book’s real value lies in the way it makes us reimagine what we might do as we learn to co-exist with AI Here the book is genuinely helpful, at least for those of us who don’t know our “personas” from our “context windows”. We learn about the different approaches to collaborating with AI in order to conduct deep research and generate ideas, the various techniques for writing prompts that get the best possible responses, when we should avoid using AI entirely and where it is most useful. All of which is grounded in the message that runs through the book, namely that “AI is not a replacement for your own thinking, and the value it provides depends entirely on how you choose to use it”. The book is also reassuring about what humans still bring to the table, not least our powers of judgement, taste and, most importantly, agency – defined roughly as the willingness to act independently, without instruction. These, he argues, are what we need to be careful to preserve in the era of coexistence. Mollick goes on to explore how our new relationship with AI will play out across business, politics, education and our relationships with each other. It is all thorough and measured, and he is candid about the many uncertainties, despite tending towards the optimistic. What’s more, there is plenty to reward even those with a deeper understanding of AI. There is an intriguing passage, for instance – cleverly foreshadowed by a disconcerting exchange with Claude – on what the frontier AI labs are learning as they probe how models represent specific concepts, including indications the models may develop something that looks a lot like intention and introspection. But for me, and all the other holdouts inefficiently slogging away without using AI to its full potential, the book’s real value lies in the way Mollick makes us reimagine what we might do as we learn to co-exist with these collaborators, and the way he shows us how to start. Indeed, it wasn’t lost on me that I began the book wondering if engaging with AI would mean giving something up, but now I am thinking I might be losing something by holding out. As Mollick says, “the people who benefit the most are the ones who actually bother to try”. Daniel Cossins is a writer, based in London. He is a former head of features at New Scientist When you make a purchase via the links on this page, we receive a commission.