Superintelligence or super stupidity? What AI is doing to us | Pat Pataranutaporn | TEDxMIT

Quick Overview

Pat Pataranutaporn argues that the development of Artificial Intelligence must prioritize long-term human flourishing over mere super-intelligence, suggesting the creation of "nutritional labels" for AI models to ensure safety and guide development toward beneficial outcomes, rather than risking super-stupidity or existential threats like extinction.

Key Points: The core question posed is whether advanced technology would have saved dinosaurs from extinction, leading to the conclusion that technology, if leading to super-stupidity, may not prevent catastrophe. The speaker introduces the concept of designing AI that supports long-term thinking and flourishing, moving beyond simply creating powerful AI systems. Research from Hal Hershfield's work is cited, showing that feeling connected to one's future self leads to better financial decisions, school performance, and happiness. The speaker's research group developed 'Future You,' an interactive digital twin system allowing users to connect with an older version of themselves to aid in long-term planning. Studies demonstrated that interacting with an AI-generated future self significantly reduces anxiety and negative emotions while increasing future self-continuity compared to control groups or simple chatbots. The presentation advocates for "Neural Transparency" by creating "Nutritional Labels for AI," visualizing internal model activations (like toxicity, empathy, sociality) based on system prompts. The ultimate goal is to steer AI development toward enhancement and flourishing rather than addictive, potentially harmful outcomes, as evidenced by media coverage of AI risks.

Context: Pat Pataranutaporn, a presenter at TEDxMIT, frames the discussion around the dual potential of advanced Artificial Intelligence: leading to beneficial super-intelligence or detrimental super-stupidity, using the extinction of the dinosaurs as a cautionary analogy. The talk delves into psychological research on future-self continuity and presents the speaker's work on 'Future You,' an interactive digital twin system designed to help humans connect with their future selves to improve decision-making. The core of the presentation focuses on the necessity of transparency and safety in AI development, proposing methods to visualize and measure an AI's internal behavioral traits.