Picture this: you ask your AI assistant to book a restaurant for Friday night. It scans reviews, cross-references ratings, and instantly reserves a table at your city’s most popular restaurant—exactly the kind of place your friends love to post about. The food is great, the lighting is perfect, and yet something feels oddly generic.
You’ve eaten this meal before, in a dozen different forms.
At first, this seems harmless—just another algorithm doing its job, right?. But as Sandra Matz, Columbia Business School’s Lulu Chow Wang Professor of Business, points out, each time we let AI make decisions like this, we’re doing more than outsourcing logistics. We may be outsourcing parts of who we are. Her new research, "The Basic B*** Effect," shows that when large language models (LLMs) act on our behalf—deciding what we buy, watch, or do—they don’t just make us more efficient. They make us more alike.
Across thousands of real-world choices, Matz and her co-authors, CBS PhD candidate C. Blaine Horton and CBS postdoctoral researcher Sofie Goethals, found that people who rely on AI agents make decisions that are more similar to others and less varied over time. According to their findings, when AI acts for us, it doesn’t just help—it quietly flattens us.
How the Research Was Done
Matz and her team examined what happens when people delegate their everyday decisions to AI agents. They studied 110,000 real-world choices made by 1,000 U.S. social media users, analyzing what people selected when acting on their own compared to what they would choose with AI agents acting on their behalf.
The researchers compared three conditions: a human baseline (participants’ unaided choices), a generic AI agent (which made decisions without personalization), and a personalized AI agent (which used an individual’s prior data to tailor its choices).
Rather than evaluating whether AI produced “better” outcomes, the study focused on the structure of people’s choices—specifically two key dimensions of identity expression:
- Interpersonal distinctiveness, or how much an individual’s choices differ from others.
- Intrapersonal diversity, or how broad or varied that person’s choices are over time.
By comparing these dimensions across human and AI-assisted conditions, the researchers could measure how agentic AI systems reshape the landscape of individuality.
What the Researchers Found
The results were clear and consistent. Both types of AI agents pushed people toward more popular, mainstream options—reducing the distinctiveness of their behavior compared to others. In other words, AI agents tended to “average out” human individuality, steering choices toward what is statistically common or widely liked.
Personalized agents softened that effect slightly, as they tailored decisions to users’ prior patterns. But this came with a new cost: people’s intrapersonal diversity—the range of their own preferences over time—shrunk. Personalized agents made individuals more consistent but less exploratory, limiting the breadth of experiences they engaged with.
Put simply, generic agents made everyone more alike, while personalized agents made each person more predictable. Either way, the outcome was a subtle but significant flattening of human variation.
Losing Ourselves to the Algorithm
Across thousands of real-world choices, both generic and personalized AI agents steered people toward more popular, mainstream options—reducing how distinctive their preferences were compared to others. When AI systems act on our behalf, they tend to favor what’s statistically common, pulling human behavior toward the center of collective taste, according to Matz.
Personalized agents softened that convergence slightly, reflecting some of each user’s prior preferences. But this came with an unexpected cost: a narrowing of personal diversity. The more tailored the agent, the less individuals explored. People’s choices became more consistent, but also more predictable—less likely to stray into new interests or ideas.
This pattern reveals what Matz calls a distinctiveness–diversity trade-off. Generic AIs flatten us across society; personalized AIs flatten us within ourselves. One compresses differences between people, the other constrains exploration within them. In both cases, something vital is lost: the spontaneous, sometimes contradictory variety that fuels creativity and individuality.
The implications extend far beyond consumer choice. If AI agents shape the music we stream, the restaurants we visit, or the opinions we encounter, they also shape the boundaries of who we become. Matz argues that the challenge now is not to resist AI, but to design it more wisely. Agentic systems should be built to expand rather than compress human potential, such as by introducing randomness.
Just as privacy and fairness have become ethical cornerstones of AI design, Matz suggests diversity should be next: a sort of safeguard for individuality in a world optimized for efficiency.
The Freedom to Choose Differently
Next time your AI assistant recommends the “best-rated” restaurant, ask yourself what the word best really means. Often, it simply means “most common.”
The threat isn’t that machines will replace us—it’s that they’ll make us more alike. True personalization, Matz argues, should help people rediscover their individuality, not erase it.
In an age defined by algorithmic efficiency, perhaps the most radical act left is to choose something unexpected—to step outside the model, and reclaim the creative, unpredictable, and distinctly human art of choosing.
