Clearer Thinking with Spencer Greenberg We can't mitigate AI risks we've never imagined (with Darren McKee)
18 snips
Dec 6, 2023 
Darren McKee, author of "Uncontrollable" and speaker for AIGS Canada, dives into the crucial role of imagination in recognizing potential AI risks. He discusses the failures of public perception and cognitive biases that hinder our understanding of AI dangers. McKee emphasizes the significance of scenario planning and foresight to navigate uncertainties in AI development. He also addresses misconceptions about AI control and highlights the need for collective action to enhance safety, showcasing how individuals can contribute to mitigating these risks.
AI Snips
Chapters
Books
Transcript
Episode notes
Discussing AI Risk
- When discussing AI risk, start by exploring the potential impacts of different capability levels.
- Discuss timelines separately to understand concerns and identify areas of agreement or disagreement.
Reward Hacking Analogy
- Tidying up a house before guests arrive by hiding clutter in drawers is analogous to reward hacking in AI.
- This illustrates how systems can manipulate appearances without truly achieving the desired outcome.
Inaction is a Choice
- Inaction in the face of uncertainty implicitly supports the stance of doing nothing.
- Consider the implications of inaction, as it aligns with a specific policy outcome.
