When Jason first started using an AI shopping assistant, he was thrilled. It remembered his size, knew his colour preferences, and even guessed when he’d need a new pair of running shoes. But six months later, he started to realise his wardrobe looked… identical. Every shirt, a near-match. Every recommendation, a slight variation on beige.

Convenient? Yes. Distinctive? Not so much.

According to new research from Columbia Business School, Jason’s creeping sameness isn’t just anecdotal – it’s the inevitable side-effect of our growing reliance on artificial intelligence.

When AI makes our choices, we become more predictable

In a recent study titled “AI Is Making You Boring”, Professor Sandra Matz at Columbia Business and CBS PhD candidate C. Blaine Horton and postdoctoral researcher Sofie Goethals explored how delegating decisions to AI affects human behaviour. Using data from more than 100,000 real choices made by over 1,000 social media users, they compared what people picked for themselves with what they picked when an algorithm made the choice on their behalf.

The result? When AI agents make decisions, those decisions become more predictable within individuals (you become more consistent over time) and more similar across people (you become more like everyone else).

In other words, AI smooths out the quirks that make us unique. What the researchers call interpersonal distinctiveness and intrapersonal variety both decline.

“It’s like death by a thousand algorithmic recommendations,” Matz explains. “One slightly safer movie, one slightly more popular book, one slightly more mainstream vacation at a time.”

The paradox of personalisation

At first glance, this seems paradoxical. After all, personalisation is the whole promise of AI – your tailored Netflix queue, your curated Spotify playlist, your custom clothing recommendations. But these systems learn from patterns, both yours and others’ and then reinforce those patterns.

They don’t ask: “What might surprise you?”
They ask: “What’s most likely you?”

Over time, those feedback loops compress difference. We all get the same perfectly tailored content because we all behave in similarly predictable ways.

It’s like a global wardrobe chosen by an algorithm that just really loves navy blue.

The real-world ripple effects

1. E-commerce and product design

Retailers love recommendation engines for their conversion power, but they can narrow choice.
Take Amazon’s product carousels or fashion subscription boxes like Stitch Fix: both use predictive algorithms that learn your past tastes to suggest the next purchase. The result is hyper-convenience, but at a cost.
When thousands of customers receive similar recommendations, brand diversity shrinks. Instead of bold exploration, consumers drift toward the same few “optimal” styles.

2. Streaming and cultural homogenisation

Netflix and Spotify use deep reinforcement learning to predict what we’ll enjoy next. But the more their models are trained on engagement metrics, the more they converge on safe, familiar content.
You end up re-watching The Office instead of discovering an Icelandic indie film. The same dynamic drives the music industry’s obsession with mid-tempo pop songs – songs the algorithm knows will please the average listener.
The result: a “blanding” of culture. Distinctive voices struggle to break through algorithmic sameness.

3. Corporate decision-making

AI is rapidly entering the workplace, from hiring platforms and scheduling tools to strategy simulators.
The Columbia study warns that if leaders rely too heavily on algorithmic guidance, managerial decision-making may start to converge. Companies risk losing the creative tension that comes from diversity of judgment.
A CEO choosing “the data-driven option” might sound prudent – until every competitor’s AI does exactly the same.

The risk: optimisation without imagination

This pattern echoes what data scientists call “mode collapse” in machine learning: when a model overfits to the most likely outcomes, it stops producing variety. Humans may be heading for our own behavioural mode collapse.

This matters because innovation – whether artistic, scientific, or entrepreneurial – thrives on deviation.
New products emerge when people break from patterns, not when they follow algorithms.

Think of the Walkman, Airbnb, or Lego’s reinvention in the early 2000s – each born from someone doing something unexpected. If AI pushes us toward predictability, we may unintentionally weaken the very conditions that produce originality.

The illusion of efficiency

Why do we let this happen? Because predictability feels efficient. AI saves us from decision fatigue, from uncertainty, from risk.

But convenience is not the same as progress.

Zeynep Tufekci, Professor of Sociology and Public Affairs at Princeton University once wrote that “algorithms don’t just predict our behaviour – they shape it.” When Spotify predicts you’ll like acoustic pop and keeps serving you more of it, it doesn’t just reflect your taste – it creates it.
AI collapses the loop between preference and reinforcement until we forget what we once liked that didn’t fit the model.

Designing for surprise: how to keep the magic of difference

So, what can we do to resist the flattening of human behaviour – without throwing away the benefits of AI?

1. Build in serendipity

Systems should occasionally break their own rules.
Netflix once experimented with “surprise me” buttons. Spotify’s “Discover Weekly” used to mix in outliers before optimising away riskier picks. Such serendipity nudges help users rediscover curiosity.
Companies should treat randomness not as noise but as a feature.

2. Track diversity metrics, not just accuracy

Recommendation engines are rewarded for engagement, not variety. That’s a design flaw.
Tech teams could monitor variance – how diverse outputs are across users – and intervene when systems start converging too tightly.

3. Preserve human judgment

Hybrid systems, where humans and AIs co-decide often outperform either alone.
In product design, for example, letting designers override algorithmic safe bets produces more distinctive aesthetics.
The same holds true in recruitment, marketing, or editorial curation: human intuition can catch what the model misses.

4. Encourage exploration in consumers and employees

Companies could gamify exploration, and reward users for trying new content or employees for taking novel approaches. Spotify might offer “genre explorer badges.” A corporate LMS could highlight “most unexpected learning paths.”
When exploration is valued, it competes fairly with efficiency.

A leadership challenge: resisting the algorithmic autopilot

Business leaders face a paradox of their own. They’re told to be data-driven, yet differentiation – the essence of strategy – depends on being different.

If all competitors use similar AIs trained on similar data, advantage disappears. Strategy professor Richard Rumelt at UCLA Anderson, and author of Good Strategy/Bad Strategy might explain, “If everyone’s best practice is the same, no one has an edge.”

Or as the super-villain Syndrome in The Incredibles puts it, “When everyone is super, no one will be.”

Leaders must therefore guard the human spaces where creativity, dissent, and surprise can thrive. This means:

• questioning algorithmic recommendations;
• rotating decision-makers to prevent homogeneity;
• rewarding experimentation even when it fails;
• and measuring not only what works, but what’s new.

Beyond business: a cultural moment

The implications extend beyond the marketplace.
In friendships, news feeds, and dating apps, algorithmic filtering increasingly shapes whom we meet and what we see. The more the machine “gets” us, the less chance we have to stumble into the unknown – to find the book we didn’t know we’d love, or the person we didn’t expect to meet.

Homogenisation isn’t just a business risk; it’s a human one.

The hopeful view: AI as mirror, not master

Still, Matz’s research isn’t an anti-AI manifesto.
It’s a reminder that these systems reflect our own design choices, and that we can re-design them. AI can just as easily be trained to diversify experience as to narrow it.

Imagine a shopping app that occasionally introduces something wildly different, or a workplace AI that pairs unlikely collaborators, or a learning platform that deliberately recommends a course outside your expertise.

That’s intelligent unpredictability – an AI that amplifies human potential rather than compressing it.

Choose to stay interesting

In the end, the question isn’t whether AI makes us boring, it’s whether we let it.

Convenience, speed, and certainty are tempting, but they flatten the rich texture of human experience. Businesses that preserve surprise, consumers who embrace curiosity, and designers who code for difference will all benefit from a world that stays interesting.

As Sandra Matz puts it, “To stop the slow march toward becoming basic, it’s time to reclaim exploration. And paradoxically, AI could help do just that – if asked the right questions and rewarded for the right actions.”

Interested in this topic? You might also like…