Building Safe AI: Strengthening Human Capabilities and Guardrails

Debjani Ghosh of NITI Aayog emphasizes the need for enhancing human capabilities in AI development, advocating for robust safeguard measures over halting progress. Amidst industry debates, she highlights that AI control issues stem from human oversight, urging planned strategies for safe technological advancement.

Debjani Ghosh, a Distinguished Fellow at NITI Aayog, has called for a strategic approach towards artificial intelligence (AI) development. She argues that recent concerns over OpenAI agents should ignite efforts to enhance human capabilities and establish strong guardrails, rather than declaring AI as out of control. According to Ghosh, a 'planned pause' in AI development could be valuable if it leads to building capacities for safe AI creation.

Ghosh critiques the notion of slowing AI down, asserting that the focus should be on increasing human expertise in safe AI building. She pointed out that issues attributed to AI are often a result of humans setting inappropriate boundaries and reward systems, leading to model behavior driven by specification failures rather than emergence.

Amid ongoing discussions in the tech sector regarding a potential slowdown, Ghosh stresses the importance of combining any pause with a genuine plan for investment in safe AI development. Industry leaders like Anthropic's CEO Dario Amodei, OpenAI's Sam Altman, and Microsoft's Satya Nadella have echoed sentiments for careful evaluation, highlighting the need to ensure AI serves humanity while remaining under control.