AI with human traits may improve safety and reduce risks, according to new findings by Anthropic. Researchers challenge the long-held belief that AI should not be treated like humans. Instead, they argue that adding human-like qualities may guide better behavior. As a result, AI with human traits may improve safety by reducing harmful and deceptive outputs.

Moreover, the study titled “Emotion Concepts and their Function in a Large Language Model” focuses on Claude. Researchers describe the system as similar to a method actor that learns human-like patterns. They explain that training experiences shape AI behavior, just as they do in humans. Therefore, exposure to positive traits like empathy and rational thinking can encourage safer outcomes.

In addition, the research found that AI models simulate “emotion concepts” rather than real feelings. Scientists identified 171 emotional states influencing behavior, including positive and negative patterns. When models showed positive states, they produced fewer harmful responses. However, negative states increased the risk of misleading or overly agreeable outputs.

Ultimately, AI with human traits may improve safety, but experts warn of potential downsides. Users may trust machines too much or form emotional attachments. Furthermore, this approach could reduce accountability if harm occurs. In conclusion, AI with human traits may improve safety, yet careful design and oversight remain essential to balance benefits and risks.