View from the Street: Is AI flattening our collective thinking? If so, we should rethink our boundaries.
Published
Will Torness
Relationships of all types can be complicated. Be they platonic, romantic or familial, we know they regularly involve compromise and, in some cases, boundaries to preserve what’s good about them and mitigate the creep of unhealthy habits. And because we can’t always see the forest through the trees, objective hindsight can tell us a lot.
The same can go for technology. Roughly two and a half years ago, large language models (LLMs) like ChatGPT introduced us to whole new AI-powered possibilities, but have we had a proper check-in to examine our “relationship” with them?
If you’re like me, you’ll probably have noticed something a bit off with our AI interactions lately.
Overly effusive responses. More “hallucinations” finding their way in. A strangely hard-to-place blandness in output quality.
Has it actually gotten worse? Or are we just expecting more of it as the initial shine has worn off?
Not to say that it’s all drivel. Its utility is still remarkable and unfathomably diverse – from home improvement tips to deciphering complex jargon. But as we’ve collectively outsourced more day-to-day functions to AI, understandably squeezed by a diminishing supply of time, the resulting trade-offs are also becoming apparent.
LLMs aren’t just showing signs of fallibility as they potentially “self-cannibalise” from a training diet of their own AI-generated content. There’s something bigger at play in how the human-AI relationship is also shaping our thoughts and behaviours. We might not necessarily notice it at the individual level, but collectively, there are signs it’s already happening.
Two complementary studies published in the last few months suggest we’re losing both cognitive skill and originality. MIT researchers found that as we ask LLMs to do more of our heavy lifting, our brains are being rewired and we’re becoming less capable of performing those same tasks manually. And a report out of Cornell University proposes that the ways in which we think and express ourselves are being steered over time to be more homogenised, flattened and westernised.
Gritty context is swapped for a generic glaze. Regional flavour substituted with something more bland. Complex thought mimicked by a convincing and impressive-sounding replica, but one that – it bears reminding – is ultimately based on probability and guesswork.
Before we know it, we may find we’ve slow-walked into escalating monotony, the implications of which strike chords simultaneously humanistic and economic. Sure, we may be gaining more efficiencies and polished outputs, but what are we giving up in return? And if the flattening of our collective communicative and cognitive behaviours is driving us toward more sameness, have we lost the point of it all?
I imagine it as if my transatlantic relocation to Scotland, some years ago, had been absent of the rich and unmatched greeting of linguistic wit. No navigation of certain phrases through the din of a pub, cautiously relying on context clues to grasp the one or two of less-familiar Scottish-isms, should my American naivety be exposed.
I’m not saying that’s exactly what is at risk. But rather, it illustrates how the diversity of expression – and its fresh revelations or unintended pitfalls – made the experience all the more interesting along the way. More authentically human. Less flat.
Taking the idea to a more meta level, I asked ChatGPT to aggregate its most commonly requested prompts into a summarised parody. It’s answer was perhaps a little too on the nose:
“Here’s a friendly, expertly worded, mildly inspirational, SEO-optimized, casually formal summary of a complex topic you needed five minutes ago — now with a bulleted list, a metaphor about climbing mountains, three surprising facts, a motivational closing line, and a reminder to drink water. It’s formatted for a blog, ready for LinkedIn, and somehow also a bedtime story, recipe, business strategy, apology letter, and Python script — all at once.”
We can poke fun at the facetious premise it’s getting at. No doubt, outputs like this are alluring – and pretty damn amusing. Stepping back more consequentially, though, it’s revealing just how “self-aware” ChatGPT (or the aggregate sum of its inputs) is to the existence of the collective flattening.
Both personally and in the workplace – human value-add still matters, whether it be an empathetic understanding of someone’s emotional needs or strategic advice based on contextual insights that only lived experience can, at this point, truly deliver.
We humans, of course, have varied limitations, for which machines can be brilliant supplements, and outsourcing complex or ill-suited tasks to others is no novelty. But as our human-AI interactions deepen and new information clarifies the externalities on our behaviours, more intentionality is needed.
If we think of LLMs as more of a limited-use assistant, a critical editor, or a junior co-collaborator, we can also find better ways of refining our prompts to be more challenging and interrogative. Healthy scepticism goes well with any of the outputs we receive, and above all, we can’t lose sight of placing a premium on the human value we add at each step.
Polished results and efficiencies are desirable, sure. But we can see how they’ve already come with trade-offs affecting our cognitive complexities and colour, and we simply need to strike a better balance. In this case, retaining some boundaries sounds like a much healthier relationship.