Published Apr 2, 2023

#112 AVOIDING AGI APOCALYPSE - CONNOR LEAHY

Join Connor Leahy, CEO of Conjecture, as he delves into the complexities of embedding empathy and ethical rationality in AI systems, highlighting the potential extremes of AI-optimized empathy and the challenges of ensuring AI alignment with human values to avoid an AGI apocalypse.
Episode Highlights
Machine Learning Street Talk (MLST) logo

Popular Clips

Episode Highlights

  • Empathy's Role

    Empathy in AI development presents unique challenges, as explains the difference between human and AI empathy. He argues that most humans operate within limited empathy boundaries, unlike AI, which could optimize empathy to extreme levels, resembling the work of Brian Tomasik. Leahy suggests that true empathy requires disregarding social norms, often seen in autistic individuals, who may possess higher empathy levels than typically perceived 1.

    Really moral people do not look like charismatic Hollywood stars. No, they're weird aliens. They get obsessed and write massive papers about simulated souls in the far future universe.

    ---

    This unconventional view challenges the traditional understanding of empathy and its role in AI systems 2.

       

    AI Ethics

    The ethical considerations of empathy in AI systems highlight the complexity of moral reasoning. discusses the challenges of applying human-like empathy to AI, noting that extreme empathy can lead to morally questionable decisions, as seen in Tomasik's controversial views 3. He emphasizes the difficulty of consequentialist reasoning, comparing it to solving NP-hard problems, which require evaluating countless potential outcomes 4.

    Consequentialism is fundamentally NP hard; you have to reason about all possible universes in which you could exist and how your effects could affect these various universes.

    ---

    This complexity underscores the need for careful ethical considerations in AI development.

Related Episodes