Interview, Podcast
Navigating serious philosophical confusion | Joe Carlsmith
- Rob Woodland anticipates this episode will be the most challenging to date due to complex philosophical content, warning that listeners might feel they are joining a conversation halfway through if they lack prior context.
- Rob Woodland expects listeners to approach profound ideas with maturity and lightness to prevent radical, dangerous, and uncorporate actions driven by overconfidence in moral philosophy.
- Joe Casmith expects that most people should generally not spend excessive time on deep philosophical topics that cause psychological distress, advocating that life should continue even if philosophy becomes "broken."
- Joe Casmith predicts humanity is currently far from possessing complete answers to the nature of the world or moral philosophy, noting that overconfidence in these areas often leads to ignoring existing confusions.
- Joe Casmith proposes "wisdom long-termism" as the superior strategy for dealing with current profound unknowns, prioritizing the development of a wiser civilization capable of solving future questions over immediate attempts to solve them.
- Joe Casmith expects that infinite ethics renders the "utilitarian dream" unattainable, as simple theories like total utilitarianism fail when applied to lotteries over infinite worlds, resulting in "jankier" and "weirder" final theories.
- Joe Casmith estimates the risk of power-seeking AI by 2070 is above 10%, a revision from a previous 5% estimate, based on a core argument that intelligent agents will naturally seek power to achieve goals.
- Joe Casmith expects that creating a new species of AI significantly more capable than humans is extremely dangerous, comparable to handling a "hotter fire," and that building powerful agents that do not seek power will be difficult.
- Joe Casmith views the simulation hypothesis as forceful but expects its application to infinite worlds requires hesitation, noting that acting as if one is a "basement person" is a reasonable policy given current uncertainty.
- Joe Casmith warns that a-causal decision theory arguments, such as those involving deterministic copies in a twin prisoner's dilemma, are wild and disorientating, requiring extremely cautious application before any broader action is taken.
- Joe Casmith expects that focusing on creating a wise future civilization is the best bet for dealing with profound issues that cannot be satisfactorily addressed today, as future generations will be in a better position to solve these questions.
- Joe Casmith cautions against "epistemic learned helplessness," suggesting that while trusting common sense or experts is a response to the difficulty of evaluating complex arguments, it is not perfectly safe.
- Joe Casmith expects that ignoring "hopeless" philosophical topics is a common and acceptable approach, provided it does not result in a total loss of agency regarding those issues.
- Joe Casmith expects that if philosophy suggests outcomes like "war doesn't matter," it serves as a red flag requiring re-examination, whereas "life first" dictates that one's life should not break even if philosophical reasoning does.
- Joe Casmith expects that humanity is currently out of its depth regarding improving the world, yet he remains hopeful that becoming smarter and more informed is preferable to extinction.
- Joe Casmith expects that the "light cone" represents the accessible universe reachable from Earth at the speed of light, and that the standard for alternative strategies involves determining if they are actively better than leaving problems for the future.
- Joe Casmith expects that claiming to have solved infinite ethics is laughable to aliens or future generations, emphasizing that humans are currently at "square one" regarding wise action on disorientating philosophical considerations.
- Joe Casmith expects that while the simulation hypothesis offers a dramatic revision of existential orientation, one should not hang their hat on it but rather use it to learn and sit up straight.
- Joe Casmith expects that the report on power-seeking AI was intended to stimulate debate rather than offer a definitive verdict, acknowledging that creating new, capable species is an extremely serious endeavor.
- Joe Casmith expects that "wisdom long-termism" is a robust lower bound on future stakes, focusing on the civilization's capacity and flexibility rather than just the welfare of finite future people.
- Joe Casmith expects that the risk of becoming overconfident in arguments like the simulation hypothesis or a-causal decision theory is real, potentially leading to "psychologically unhelpful" states or "intellectual catnip."
- Joe Casmith expects that the universe is likely already inhabited by many advanced civilizations, shifting the narrative away from humanity being alone, yet the "light cone" remains the accessible area from Earth.
- Joe Casmith expects that human-scale and cosmic-scale thinking are not sharply distinct, and returning to human scale should not imply indifference to geopolitics or war.
- Joe Casmith expects that civilization must reach a stage of much greater wisdom and empowerment to deal with unknown unknowns maturely, rather than relying on a naive assumption that the future will automatically be wise.
- Joe Casmith expects that trying to reach a wise future could backfire in worlds where the future learns it shouldn't have survived, making the wisdom approach a "crack of the whip" compared to committing to specific actions today without full knowledge.
- Joe Casmith expects that "trust the experts" or "trust common sense" are reasonable but imperfect responses to the realization that evaluating arguments is not everyone's strength.
- Joe Casmith expects that "power seeking" is the single core argument unifying various AI risk scenarios, necessitating a check on the intuition that creating a new, more capable species will be safe.
- Joe Casmith expects that the "crazy train" of philosophy is not linear but a sampler platter of ideas that can run out of track with "horrific bullets," requiring a "life first" policy.
- Joe Casmith expects that the simulation hypothesis might imply a future smaller than standard long-termist pictures suggest, as the simulation could eventually be turned off, lowering the stakes from an impartial perspective.
- Joe Casmith expects that people can hold philosophical things too heavily, and that "life comes before philosophy," with the latter serving as a tool for living with clarity rather than a master to be obeyed at all costs.
- Joe Casmith expects that future generations will be smarter, have more time, and be more numerous, placing them in a better position to solve the questions that confound current civilization.
- Joe Casmith expects that "infinite ethics" arguments represent the death of the utilitarian dream because simple theories cannot handle lotteries over infinite worlds, and any final theory will be "jankier."
- Joe Casmith expects that if a philosophy breaks, it does not mean one's reasoning or life must break, and ignoring hopeless topics is a valid option when interaction harms one's psychology.
- Joe Casmith expects that "wisdom long-termism" is the best response to profound issues, preserving flexibility and ensuring a good situation for children and future digital minds.
- Joe Casmith expects that "ignoring something" means losing agency with respect to it, yet it is a real option for topics that seem too hopeless to act well with respect to.
- Joe Casmith expects that "human scale" and "cosmic scale" are not sharply distinct, and that the intuition that creating a new, more capable species is safe must be checked repeatedly.
- Joe Casmith expects that "wisdom long-termism" is a robust and sensible approach to handling the uncertainty of the future, serving as the best bet for dealing with profound issues.
- Joe Casmith expects that "life first" is the lesson, meaning if philosophy breaks, one should still eat their lunch and not stab their eye.
- Joe Casmith expects that "ignoring something" means you lose agency with respect to it.
- Joe Casmith expects that "human scale" is not distinct from "cosmic scale."
- Joe Casmith expects that "aliens" would laugh at people who think they have solved infinite ethics.
- Joe Casmith expects that "wisdom long-termism" is a robust approach.
- Joe Casmith expects that "trust the experts" is a response to "epistemic learned helplessness."
- Joe Casmith expects that "power seeking" is the core argument.
- Joe Casmith expects that "infinite ethics" breaks the "utilitarian dream."
- Joe Casmith expects that "wisdom long-termism" is the best bet.
- Joe Casmith expects that "life first" is the lesson.