Sign in. It’s quick, free and it’s up to you.
An account is an optional way to support the work we do. Find out more.
Alamy Stock Photo
OpinionArtificial Intelligence
AI can spare us the frustration of thinking. That might just be the problem
Artificial intelligence might be making us more productive, but if we rely on it too much, we risk losing our ability to truly understand anything.
ON A RECENT episode of Inside Learning, the podcast from our Trinity College-based technology centre Learnovate, host Aidan McCullen shared a story he’d heard from a chief operating officer.
A colleague had sent the COO a lengthy report, and he duly read it carefully and gave detailed feedback. Then he asked the colleague for his thoughts on the report. The reply? He had generated it using Gen AI and hadn’t formed any personal thoughts. No doubt the report was well written, neatly structured and perfectly plausible. Yet the colleague had never even read it, let alone struggled to understand what it said. So how could he stand over it?
This is the quiet risk in the age of AI – we are producing more while understanding less. Understanding can’t be generated on our behalf. It’s built slowly by expending effort and enduring frustration.
Learning requires the uncomfortable internal work of fitting new ideas into what our brains and bodies already know or believe. This is how we come to learn and through which our understanding becomes our own. AI can help us work through that pain, or it can allow us to bypass it entirely. Which we choose, and our children choose, has consequences for our capacity as human beings to think, judge and understand.
Everything becoming same same
You have no doubt noticed that while AI-generated emails, posts and reports are polished, well-structured and perfectly reasonable, they are also somehow interchangeable. When researchers, led by the University of Washington, put tens of thousands of open-ended questions to AI models, the kind with no single right answer, they found the models gave strikingly similar responses as described in the paper Artificial Hivemind: The Open-Ended Homogeneity of Language Models (and Beyond). Even when systems were built by different companies, they arrived at the same ideas.
The researchers named this effect the Artificial Hivemind. There’s a simple reason for this. AI is designed to produce the most likely answer, the one that best fits the patterns in everything it has learned from. That makes it amazingly useful, but it also means it pulls inevitably and persistently toward a homogenised centre.
In the same way that cosmetic surgery is producing a convergence of faces, the so-called Instagram face, that is flawless but slightly unreal, AI content shaped with too little human negotiation risks producing a similar effect: polished, but generic and somehow uncanny.
In authority, we trust
Part of what makes AI so persuasive is the way it converses. It answers instantly, fluently and with confidence, whatever our question. Kate O’Neill, known as the Tech Humanist, will be in Dublin as keynote speaker at Learnovation, our learning conference in Dublin this week, which looks at how human skills are essential in an AI world. O’Neill argues that this AI fluency triggers something deeply human in us. It feels like talking to an intelligence like our own so we begin to trust it and hand over more and more of our thinking.
Yet AI can be confidently wrong and seductively agreeable, reflecting our own assumptions back to us with optimised grammar and spelling. When we continually accept generated content on trust, we are not just swallowing AI answers undigested. We gradually surrender our critical faculties, our judgment, our discernment.
O’Neill has made the point that we assume learning happens when we ask a question and get an answer. In reality, we usually get partial answers, often conflicting ones, from different people who see the problem from different sides. Insight comes from within us as we grapple with why those answers differ, and what the tension between them reveals. That weighing up is the practice that builds judgment.
AI quietly removes that effort. It gathers the partial answers for us and smooths them into a single, tidy response. The gaps disappear, and so does the tension.
Learning hurts
Why are we so ready to hand that work over? Partly because learning hurts. It takes effort, and most challengingly in the modern world, it takes time. And time is something in short supply. Psychologists know that cognitive effort encompasses feelings of frustration. Thinking hard about knowledge that doesn’t align with what we already know feels unpleasant.
Some learning is easy. It’s easy to assimilate learning when it already fits what we know and believe. What’s hard is when we bump up against a new idea that upends what we believe or refuses to fit alongside what we thought we knew. We experience confusion and frustration until eventually a new picture emerges that accommodates our new learning and all fits together again.
It’s no surprise, then, that we reach for a tool that promises to spare us all this effort. That isn’t laziness. It’s a natural human response to discomfort. We’ve grown accustomed to reaching for Google and other search engines over the last few years, so why not an AI chatbot?
But that discomfort is doing essential work. World-renowned learning scientist and professor at ETH Zurich Manu Kapur has spent years studying what he calls “productive failure”. As he recently told us, his research shows that people learn more deeply when they wrestle with a problem before being shown how to solve it, even though they fail at first.
The struggle activates what they already know, exposes the gaps and prepares the ground for the new idea to take root. When the explanation finally comes, it lands in a mind that is ready for it. Skip the struggle and the explanation has nowhere to land. We may be able to mimic an explanation, even perhaps apply it in a specific context, but it never quite becomes ours.
No one can understand for you
The knowledge we each assemble is shaped by what we already know, the experiences we’ve had, the way we think and the body we live in. Two people can meet the same new idea and fit it into their understanding in completely different ways, because the gaps waiting for it, and the pieces that resist it, are different. Each of us fits new knowledge into a different framework that is ours alone.
Our differences and divergence are part of how we learn. Some minds make leaps between distant ideas; others go deep and notice the detail everyone else misses. Kate O’Neill argues that meaning grows out of being a person in a body, taking in the world through our senses and connecting those embodied experiences to language.
When AI hands us a ready-made piece, it’s been cut to fit a typical jigsaw, not ours. We can press it into place, but it hasn’t been shaped by our own understanding, and so it doesn’t truly connect to anything we know.
Crutch or scaffold?
AI can be a crutch or a scaffold. A crutch takes the weight for you and the more you lean on it, the less your own legs are used. A scaffold holds you steady while you do the building yourself. The technology is the same either way. What differs is how we use it.
Used as a scaffold, AI can be remarkably helpful during the most frustrating part of learning. When a new idea won’t fit, you can talk it through. You can ask about where it conflicts with what you already believe and how others have reconciled friction between concepts, or what parts you might be misunderstanding. You can ask it not to give you the answer at all, but to interrogate your fledgling understanding until you work it out.
Edtech leader and author of Super Skills, Rahim Hirji recently suggested a simple rule to us at Learnovate: a “human at the start” approach to this tech. Frame the problem in your own mind before turning to the machine. He also distinguishes between drift, passively going wherever the algorithm leads, and design, deliberately choosing when and how to use it. Both ideas point the same way. The struggle needs to begin with you and in you.
Tolerating discomfort
Think back to the colleague with the unread report. He had the document, but not the understanding, because understanding is the one thing that can’t be produced on our behalf. It has to be developed with some effort by the person who will hold it.
That building is often uncomfortable. It involves confusion, frustration and the unsettling sense that something we thought we knew no longer fits. In a world where answers arrive instantly and smoothly, it would be easy to treat that discomfort as a problem to be engineered away. But the discomfort is the sign that learning is happening, that a new idea is reshaping our picture of the world.
Let AI scaffold our work by helping us interrogate conflicting ideas, but don’t let it spare us the effort of building understanding. The jigsaw is yours, and only you can insert new pieces or wipe it away and start building again.
Nessa McEniff is Managing Director of The Learnovate Centre in Trinity College Dublin. Learnovate is a global research and innovation centre on the future of work and learning. Its annual Learnovation Summit is Ireland’s leading learning event and takes place at the Aviva Stadium this Thursday (8 October) with the theme ‘Human skills that power the AI world’.
have your say