AI language models, such as ChatGPT or Claude, sometimes get stuck in digital "groupthink." An AI language model is a computer program that understands human language and generates it itself. This "groupthink" means: The AIs repeat similar patterns instead of being truly creative. This also happens with tasks that are supposed to seem random.
This "groupthink" shows the limits of current AI research. If even top models cannot generate true randomness, their creative problem-solving ability is questionable. For anyone who needs diverse and unbiased AI results, this is a clear warning sign.
A report by MIT Technology Review shows: Leading AI language models often give the same answers. Claude, ChatGPT, and Gemini were asked for a random number between 1 and 10. They almost always answered with 7. When asked again, 3 or 4 often followed. After that came 8 or 9. This pattern was observed across different models. It is reminiscent of a team of robots throwing work at each other instead of thinking independently.
As a user or creator, you should not blindly rely on the supposed "randomness" of AI models. If you use AI for brainstorming or creative ideas, the results may be less diverse. Your AI-generated content could appear predictable and less original. This can affect your creative workflow.
Companies using AI must take this bias seriously. This is especially true for marketing texts, design suggestions, or product development. True diversity is important there. If your AI always delivers only similar solutions, you could fall behind in the competition. This costs money and important innovative power, i.e., the ability to develop new things.
These findings offer an opportunity for new research. Developers can find methods to improve true randomness in AI models. This could lead to new training approaches. These approaches would make models more unbiased and creative. Specialized tools could also emerge that recognize and correct these "groupthink" patterns.
The biggest risk is that this "groupthink" remains undetected. It could lead to distorted or poor results in important applications. For example, if AIs recommend products, they might always repeat the same patterns. This would suppress true diversity. This could also reduce trust in AI technology.
