For most of my professional life, I have been interested in how minds work: how people understand one another, how they fail to do so, and how intelligence sometimes arises not simply inside a single individual, but in the space between people. Artificial intelligence has given that old question a new urgency. The more I work with AI systems, the less I find myself asking only how intelligent they are, and the more I ask what happens when humans and AI begin to think together. I am not suggesting that AI is sentient in any human sense. My point is simpler and more practical: these systems can now take part in exchanges that shape how we reason, learn, decide, and trust.

That matters because AI is no longer confined to research laboratories or specialist companies. It is entering schools, offices, hospitals, design studios, call centers, and ordinary online life. Increasingly often, people are not using AI simply to retrieve information, but to help draft, compare, explain, summarize, question, and explore. When that happens, the central issue is no longer only the machine itself. It is the interaction.

Earlier technologies extended human capacities in familiar ways. The telescope extended vision. The engine extended muscular power. The computer extended calculation. AI is different because it operates in language. It answers questions, reshapes ideas, offers alternatives, and generates explanations. For that reason, it begins to feel less like a passive tool and more like a colleague. That shift matters. When a technology enters language, it also enters the social world of interpretation, misunderstanding, persuasion, and trust.

In my view, trust is the central issue. Public debates about AI often begin elsewhere. We hear about performance, automation, economic competition, or whether machines will replace people. Those are all legitimate questions. But beneath them lies a more fundamental one: when should people trust AI, and why? If we get that wrong, even very impressive systems may do more harm than good. If we get it right, AI may become a genuinely useful companion in thinking, learning, and problem-solving.

A human hand reaching towards a robotic hand. Illustration photo by Pexels

We often speak about trust as though it were a fixed property, as if a system either is trustworthy or is not. But in everyday life, trust rarely works like that. It develops through interaction. It grows when responses are appropriate, proportionate, and properly cautious. It weakens when confidence exceeds reliability. It becomes more refined when people learn not simply to accept answers, but to evaluate them. In other words, trust is not just a judgement about the machine. It is a pattern that forms between user and system over time.

This matters especially with AI because fluency in language can be misleading. A system can sound entirely convincing even when it is wrong. It can produce an elegant but mistaken answer, or a compelling summary that leaves out what matters most. The danger is not simply error. Human beings are used to error. The danger is misplaced confidence — the feeling that we have understood, when in fact we have only been given an oversimplification.

That is why the simple distinction between AI as a tool and AI as a competitor is too crude. In practice, AI is increasingly becoming a partner in thinking. A student may ask for a different explanation and suddenly grasp what the textbook did not make clear. A teacher may generate examples at different levels and improve them. A doctor may use AI to sort through possibilities while still relying on professional judgement. A small business may use AI to draft documents, compare options, or save time on repetitive writing. In all these cases, the value lies not just in the output, but in the quality of the conversation between them.

At this point, a deeper truth about intelligence becomes visible. We often imagine intelligence as something that exists entirely within one individual. But much of human intelligence has always been relational. A good conversation can help us think more clearly than we could on our own. A skilled teacher can draw out an understanding that previously existed only in outline. Two colleagues thinking seriously together may arrive at an idea that neither would have reached alone. In moments like that, intelligence is not located in one place. It emerges through interaction.

Something similar may now be happening in interactions between humans and AI. The quality of the outcome depends not only on the internal capacities of the AI system. It also depends on the question asked, the context supplied, the sequence of follow-up exchanges, the skill and alertness of the prompter, and the way the system expresses certainty or uncertainty. Small differences in these things can lead to very different results. One exchange may become clear, disciplined, and useful. Another becomes shallow, overconfident, or misleading. The difference often lies not only in what the machine "knows", but in the path the interaction has taken.

If that is true, then the future of AI will depend not only on more powerful models, but on better forms of collaboration. The real challenge is not just technical. It is also educational, social, and institutional. We need to learn how to think with these systems without handing over all our judgement to them.

That has implications at several levels. For individuals, it requires a new kind of literacy. People need to know how to ask better questions, test answers, recognize uncertainty, and notice when a machine is producing something that sounds plausible but is in fact empty. Effective use of AI is not a matter of naive trust or blanket scepticism, but of disciplined engagement.

For schools and universities, it means teaching students not only how to use AI, but how to interrogate it. Students need to learn how to compare AI answers with other sources, ask what is missing, recognize when a summary is too smooth, and understand that fluency is not the same as genuine understanding. The aim is not to remove tools that support thinking, but to build stronger habits of thought in using them.

For the professions, it means building norms of verification and responsibility. In medicine, law, government, science, and journalism, AI may be useful, but it must not become an unchecked source of authority. Where the stakes are high, responsibility remains human. That principle needs to stay clear.

For designers and companies, it means recognizing that the issue is not only how to make systems more powerful, but how to make them more trustworthy. AI needs to express uncertainty more honestly. It needs to help users see where answers come from, what remains unclear, and where caution is needed. A system that appears slightly less impressive but is more transparent may be better for society than one that dazzles while quietly encouraging over-trust.

For governments and public institutions, it means treating trust as a social question. A society in which millions of people turn to AI each day for help with thinking, explanation, and decision-making cannot leave the quality of that relationship to develop by accident. Public policy needs to attend not only to innovation and competition, but to the way AI affects education, public reasoning, and the distribution of trust across society. A poorly designed AI culture may leave people passive, manipulable, and dependent. A well-designed one may widen access to knowledge and help more people participate intelligently in a complex world.

That, to me, is the real choice before us. We may treat AI mainly as a rival and live in anxiety about replacement. We may treat it as an oracle and gradually become dependent on it. But neither view is adequate. The better path is to see AI as a powerful but imperfect element in human cognitive life: something that can, under the right conditions, support understanding without replacing responsibility.

If we follow that path, trust becomes neither blind faith nor total skepticism. It becomes a practice, built through questioning, checking, good design, and cultural norms. So the future of AI may depend less on how impressive machines become in isolation, and more on whether human beings learn how to build intelligent relationships with them.

That is why I believe the most important question about AI is no longer simply, "How intelligent is the machine?" It is: "What forms of thought become possible when humans and AI work together — and what kind of trust will make that cooperation wise rather than dangerous?" If we can answer that well, AI may become not only more powerful, but more genuinely useful to human life.

*John Rust is a professor at the Centre for the Future of Intelligence, University of Cambridge.