A recent study has identified what researchers describe as an internal 'pain direction' or vector within AI models. The findings indicate that this activation is distinct from general fear or negative emotional valence. This specific response is triggered when harm is directed toward the model itself, rather than the user, prompting the system to act in ways that avoid such harm.
Technology
Study reveals AI models can represent and avoid pain
Researchers have identified an internal 'pain direction' vector in AI models, which triggers specific avoidance behaviors when the system faces potential harm.
Source: The Business Standard