Is AI turning us all into the same person? | Sandra Matz

Sandra Matz, a computational social scientist with 15 years working across psychology, computer science, and business, argues that AI is quietly making people boring by narrowing taste, flattening personality, and scrubbing away the edges that make individuals interesting. She describes the effect as a death by a thousand algorithmic recommendations, subtle and cumulative rather than immediately noticeable.

The core problem is that AI systems including Spotify, Netflix, and ChatGPT are trained to optimize for exploitation, meaning short-term engagement and satisfaction, rather than discovery. Spotify has over 100 million songs and Netflix over 5,500 movies, yet both platforms funnel users toward a narrow band of safe, popular choices because recommending familiar content reduces customer churn and protects against disappointment.

Matz ran two experiments with ChatGPT to illustrate the point. When she asked the AI to recommend an ice cream flavor from Baskin-Robbins' 31 options, each time presenting itself as a new customer, it recommended one of the chain's two most popular flavors 96 out of 100 times. In a second experiment, she told ChatGPT she had chosen nutty coconut 70 percent of the time across 100 visits and split the remaining 30 percent among chocolate fudge, cookie dough, and mango, then asked it to make the next 100 choices on her behalf. ChatGPT picked nutty coconut every single time, eliminating all variety entirely.

Broader research reinforces what the experiments suggest. When people use AI for guidance, their preferences become more normative and less diverse, their creative output becomes less unique, and their choices of important scientists, athletes, and historical figures converge with everyone else's. Even when AI learns individual quirks, it still optimizes for what a person is most likely to like rather than introducing variety. Each decision outsourced to AI makes a person more narrow, and the AI then learns from that narrower version and narrows its recommendations further, creating a compounding loop.

Matz raises a structural consequence beyond individual taste. If AI consistently recommends only the most popular options, less popular choices may disappear entirely because producers have no incentive to offer what no one selects. She also warns that AI is moving from suggesting to actively choosing and acting on behalf of users, which raises the existential stakes for human complexity and diversity at a societal scale.

Matz does not see the problem as unsolvable but is precise about what will not fix it. Simply asking AI to be more creative or adventurous fails because its reward structure defaults to tried and tested output. Instead, AI needs to be rewarded for taking smart, informed risks rather than penalized every time it takes a swing and misses. Because AI has already processed the entire universe of human preferences, it knows what lies just beyond the boundaries of any individual's typical choices, which makes intelligent exploration technically feasible rather than speculative. She proposes that platforms like Netflix or Google could offer users a dial allowing them to choose how far from their typical preferences to stray, preserving a spot-on setting for those who want it while giving agency to break out when desired. Achieving this at scale, she says, requires companies to be incentivized to harness AI for exploration rather than pure engagement optimization.