by Dr Bradley Hastings
I remember my shock when a student submitted an essay that that included their entire conversation with AI. Somewhere between the chatbot and the submission portal, they had copied across not just the answer but the prompts they used to generate it. They’re far from alone. Many of my students have stopped trying to disguise AI’s work at all, leaving in the telltale em-dashes and overconfident, under supported, conclusions.
Four years ago, I wrote in The Conversation about why I don’t fear AI taking our jobs. I stand by that. What I didn’t see coming was something subtler, and arguably worse: AI stealing our thinking.
Skills atrophy is a thing. Ask anyone who has surrendered their sense of direction to a mapping app. The evidence this is happening with AI is mounting. A recent Swiss study found a negative correlation between frequent AI use and critical thinking ability, driven by what researchers call “cognitive offloading” – handing mental work over to the machine. This effect was most pronounced in younger people.
The professional world offers its own cautionary tales. Deloitte agreed to partially refund the Australian government after a researcher found its report on welfare compliance was riddled with AI hallucinations – including citations to academic papers that don’t exist and a fabricated quote from a Federal Court judge. Nobody, it seems, had thought to check.
Haven’t we been here before?
Before panicking, it’s worth remembering that human evolution has encountered similar conundrums before. Plato worried that the invention of writing would destroy memory (see Phaedrus). The arrival of the pocket calculator triggered fear that students would never again master long division. In my case, that fear proved entirely accurate – yet my critical thinking survived.
Offloading thinking is how human cognition has always scaled. Be it writing, calculators, or search engines, each outsourced a mental function and freed up capacity for something new. Philosophers Andy Clark and David Chalmers have taken this idea further, arguing in their famous ‘extended mind’ thesis that human-plus-technology thinking can be smarter than the human alone.
So which is it? Is AI a mind-extender, or a thinking thief?
It depends on the stance you take with AI
The emerging answer is both. Neuroscientist Vivienne Ming, in her book Robot-Proof, describes two divergent approaches to how people engage with AI. One group defer to its answers, taking them as gospel, and see their thinking diminish. While another group engage critically, interrogating its outputs, testing them, pushing back – thus amplifying their own cognitive effort. Sadly, these critical engagers were the minority.
The decisive variable, in other words, isn’t the technology – it’s how we choose to engage with it. Whether we become subject to AI’s answers and are swept along by whatever it tells us, or whether we treat this output as an object we hold up to the light, examine, and question.
Why cognitive offloading feels so good
The challenge with engaging more deeply with AI is that doing so uncovers a paradox in how people learn. In a Harvard study, students taught through active engagement with problems learnt significantly more than students given polished lectures. Yet the active learners reported feeling they had learnt less. This is because our brains mistake the smooth, fluent sensation of being confidently told something for the harder, messier business of actually understanding it.
And fluent, confident telling is precisely what AI is trained to produce. Every interaction is engineered to feel like learning – whether or not any learning is taking place.
An AI that refuses to answer
In the leadership programs we run at the John Grill Institute for Project Leadership, we use an AI sandbox (built by the Vertical Development Institute) that does the opposite of what AI normally does. Rather than serving up fluent answers, it responds only with questions: “What’s your assumption here?” “What information would make this decision simpler?” “What are you not seeing?”
The results have been striking. Stripped of the option to defer, students think more deeply about their problems – and when they return to ordinary AI tools, they engage with them more critically. The muscle, once worked, stays active.
Steal your thinking back
You don’t need a custom-built sandbox to do this. The shift starts with the prompts you write. Instead of “write me a proposal for…”, “summarise this data” or “which option should I pick?”, try: “What assumptions am I making?” “What am I missing?” or “Ask me five questions before you give me an answer.”
AI will happily do your thinking for you – that is, after all, what the product says in the advertising. But it can only steal the thinking we hand over. For me, I’m still happy to hand over the long division, but a good thinking partner beats a doer, any day.