Artificial intelligence could one day supercharge human cognition, leading to significant advances in science, technology and other fields. It could also make us dull, new academic research suggests.
That’s because the so-called large language models that power AI apps often yield information that is predictable and normative for the population as a whole, reducing life’s complexity to a bland mulch of watered-down ideas.
“LLMs predict the most likely next word in a sentence or event in a sequence, and by definition, that’s average,” Columbia Business School professor Sandra Matz, author of the study, told CBS News. “It tells you what the most likely thing to appear is if you ask it for a movie recommendation or what color to paint your wall. It homogenizes decisions, and we all get the same output.”
To arrive at their findings, Matz and her study co-authors analyzed more than 110,000 real-world decisions made by 1,000 people and compared them to choices made by both generic and personalized AI agents. They also used data from a Facebook application called the myPersonality project, which conducted personality tests on users who shared their Facebook profiles for research purposes.
“AI hates risk”
For individuals, relying on AI to shape a decision — say, on where to go on vacation or what walking shoes to buy — steers people toward the most common choices and away from more distinctive, or even quirky, behaviors and preferences, according to Matz, a computational social scientist with a background in psychology and computer science.
In effect, AI narrows what users “explore across topics and psychological affinities,” she wrote, adding that “LLMs play it safe within a user’s preferences.”
In other words, even if an AI agent knows its user occasionally makes an out-of-the-box or uncharacteristic decision about any given subject, like what to eat for dinner, “LLM agents nudge behavior toward more normative options and narrow the range of what individuals explore,” Matz added.
“AI hates risk because we train it that way,” Matz said. “It wants to keep you on the platform, so it shows you what you already like and not stuff on the outskirts of what you do.”
AI apps don’t have to operate this way, but it’s how they’re programmed to work, she added.
To prevent AI agents from diminishing the richness and diversity of human experience, Matz encourages tech developers to build in an “exploration mode” option for users who want more unexpected, less conventional recommendations.
That would help ensure “we prevent ourselves as individuals from becoming boring, and making sure culture doesn’t collapse into a single set of preferences,” Matz said.
