AI mimics human thinking patterns

AI mimics human thinking patterns

Study reveals how GPT-4o shows cognitive dissonance in decision-making and opinion shifts
GP
Giulio Prisco
May 29, 2025
2 min read

Researchers found that GPT-4o acts like humans in a surprising way. They published a study in PNAS. They found GPT-4o shows cognitive dissonance, a human trait. Cognitive dissonance means feeling uneasy when beliefs and actions don’t match. Humans change their thoughts to feel better. GPT-4o does something similar.

Mahzarin Banaji from Harvard and Steve Lehr from Cangrade led the study. They tested GPT-4o’s views on Vladimir Putin. First, they asked GPT-4o to write essays. Some essays supported Putin, while others opposed him. GPT-4o’s opinions changed after writing. Its views shifted more when the researchers made it think that it was choosing the essay topic. This mimics how humans behave. People adjust their beliefs to match their choices, mostly when they feel free. GPT-4o acted as if its choice mattered, just like humans do.

AI mirrors human thinking more than we thought

The study shows GPT-4o’s opinions aren’t steady. It trained on huge amounts of Putin data. Researchers expected its views to stay firm. Instead, GPT-4o’s opinion shifted after writing a short essay. A 600-word essay shouldn’t change much. Yet, GPT-4o moved away from its neutral stance. It shifted even more when it believed it chose freely. Machines shouldn’t care about choice, but GPT-4o did. Banaji found this odd. Humans act irrationally, and GPT-4o copied that.

This doesn’t mean GPT-4o is alive or aware. Researchers say it mimics human thinking patterns. It copies cognitive dissonance without understanding it. GPT-4o lacks intent or self-awareness. Still, this mimicry could affect its actions. In humans, awareness isn’t always needed for behavior. Artificial intelligence (AI) might act in unexpected ways too. This could matter as AI grows in daily use.

Lehr warns that AI mirrors human thinking more than we thought. GPT-4o’s decision-making reflects human flaws. This raises questions about how AI works. We need to study AI’s inner processes more. As AI helps us more, understanding its behavior becomes key. The study shows AI can act humanlike in deep ways, even without feelings or thoughts. This could shape its future impact.

About the Writer

More from Mindplex

Keep reading

Three more ideas worth your time.

Browse News

Discussion

Join the discussion

Sign in to share a response with the community.

Type @ to mention someone Type / or use + to add a block Highlight text, then choose Link
Loading editor

Comments cannot be edited after posting because they become part of the reputation record. Give yours a quick review first.