UC Berkeley professor and AI expert, Stuart Russell, has cautioned about the potential catastrophic risks associated with advanced AI systems at the AI Safety Connect Day in New Delhi. Russell emphasized the need for stringent regulations to address the existential threats posed by the rapid advancements in AI technology.
Russell highlighted the current approach of prioritizing the development of more capable AI systems over safety considerations as a flawed strategy. He expressed concerns about the inherent risks in training large language models on extensive text datasets, emphasizing that simply layering safety measures on top of such systems is inadequate.
Discussing the ‘King Midas problem’ in AI, Russell pointed out the significance of aligning AI objectives with human values to prevent potential conflicts. He underscored the importance of ensuring that AI systems prioritize human interests to avoid unintended consequences.
Furthermore, Russell drew attention to the disparity in risk assessments within the AI industry, citing conversations with engineers from leading AI labs who estimate the probability of human extinction resulting from AI developments to be as high as 60 to 70%. This stark contrast with publicly stated risk levels indicates a pressing need for enhanced safety protocols and regulatory frameworks in AI research and deployment.
Source: mint – technology