British scientists have identified human personality traits and individuality in AI

British scientists created the first scientifically based personality testing system for AI chatbots. The results showed that the models imitate human personality traits and form their own individuality.

As reported by BAKU.WS, this was announced by Nature Machine Intelligence.

A team from Cambridge University tested 18 large language models. They applied psychological testing methods that are typically used to assess human personality traits.

Scientists found that larger and instruction-tuned models, such as GPT-4o, most accurately imitate human personality traits. These traits can be influenced through prompts, changing the way artificial intelligence performs tasks - for example, making it more emotional when writing social media posts. Large language models also proved to be more stable and predictable, while smaller models often gave contradictory answers.

Researchers warned that the formation of AI personality can make it more convincing and capable of manipulating people. For instance, chatbots can portray themselves as pleasant conversationalists, but behave aggressively when performing certain tasks. To avoid abuses, regulation of AI systems should be introduced.