= 27d567bc2d48d9f40e9ccc20ea370b15)
The Deceptive Turn Thesis seems almost unavoidable if you start from the assumptions "the AI doesn't place an inhumanly high value on honesty" and "the AI is tested on inputs vaguely resembling the real world". That latter assumption is probably unavoidable, unless it turns out that human values can be so generalized as to be comprehensible in inhuman settings. If we're stuck testing an AI in a sandbox that resembles reality then it can probably infer enough about reality to know when it would benefit by dissembling.
In response to
The Trolley Problem and Reversibility
My trouble with the trolley problem is that it is generally stated with a lack of sufficient context to understand the long-term implications. We're saying there are five people on this track and one person on this other track, with no explanation of why? Unless the answer really is "quantum fluctuations", utilitarianism demands considering the long-term implications of that explanation. My utility function isn't "save as many lives as possible during the next five minutes", it's (still oversimplifying) "save as many lives as possible", and figuring out what causes five people to step in front of a moving trolley is critical to that! There will surely still be trolleys running tomorrow, and next month, and next year.
For example, if the reason five people feel free to step in front of a moving trolley is "because quasi-utilitarian suckers won't let the trolley hit us anyway", then we've got a Newcomb problem buried here too. In that case, the reason to keep the trolley on its scheduled track isn't because that involves fewer flicks of a switch, it's because "maintenance guy working on an unused track" is not a situation we want to discourage but "crowd of trespassers pressuring us into considering killing him" is.
"The greatest pleasure is to vanquish your enemies and chase them before you, to rob them of their wealth and see those dear to them bathed in tears, to ride their horses and clasp to your bosom their wives and daughters."
The man who said that is suspected of being an ancestor of about 8% of all living Asians so you probably should listen to him :-P
Patrilineal ancestor, not just ancestor. When talking about someone who lived 40 generations ago, there's a huge difference.
I would take Silver's analysis over Adams' any day. Look at their respective prediction track records.
Were any of Silver's previous predictions generated by making a list of possibilities, assuming each was a coin flip, multiplying 2^N, and rounding? I get the impression that he's not exactly employing his full statistical toolkit here.
In response to
Crazy Ideas Thread, Aug. 2015
We need to create more rivers! Water transport is still cheap and important for development, but sub-Saharan Africa and the middle of Asia are sadly deficient in rivers.
River creation is an interesting example of something which isn't forbidden by the laws of physics, but seems utterly unfeasible. Is there any imaginable technology for making rivers?
The obstacle to making a river is usually getting the water uphill to begin with. Regular cloud seeding of moist air currents that would otherwise head out to sea? Modifying land albedo to change airflow patterns? That's all dubious, but I can't think of any other ideas for starting a new river with new water.
If you've got a situation where the water you want to flow is already "uphill", then the technology is simply digging, and if you wanted to do enough of it you could make whole new seas.
I was under the impression those rules only applied during an active conflict/war.
Hmm... I believe you're correct. It would be hard to revise that, too, without making the "Are you a cop? It's entrapment if you lie!" urban legend into truth. It does feel like "posing as a medical worker" should be considered a crime above and beyond "posing as a civilian".
Seems to be an established conversation around this point, see: https://en.wikipedia.org/wiki/Ordinal_utility https://en.wikipedia.org/wiki/Cardinal_utility
"The idea of cardinal utility is considered outdated except for specific contexts such as decision making under risk, utilitarian welfare evaluations, and discounted utilities for intertemporal evaluations where it is still applied. Elsewhere, such as in general consumer theory, ordinal utility with its weaker assumptions Is preferred because results that are just as strong can be derived."
Or you could go back to the original Theory of Games proof, which I believe was ordinal- it's going to depend on your axioms. In that document, Von Neumann definitely didn't go so far as to treat utility as simply an integer.
That's one of the most amusing phrases on Wikipedia: "specific contexts such as decision making under risk". In general you don't have to make decisions and/or you can predict the future perfectly, I suppose.
In response to
Stupid Questions August 2015
Could someone point me towards good fictional stories set in Roman times, like those by Kipling in Puck...? Thank you. (Edit: spelling)
I really wish that having military/intelligence operatives pose as aid workers or humanitarian workers even outside of war was treated similar to a war crime.
"The feigning of civilian, non-combatant status" is already a subcategory of perfidy, prohibited by the Geneva Conventions. Perfidy is probably the least-prosecuted war crime there is, though.
= b715d02e85f7b3b4ae65ca3d3e450868)
Subscribe to RSS Feed
= f037147d6e6c911a85753b9abdedda8d){kind=logo}
I don't think this quite fits the Prisoner's Dilemma mold, since certain knowledge that the other player will defect makes it your best move to cooperate; in a one-shot Prisoner's Dilemma your payoff is improved by defecting whether the other player defects or not.
The standard 2x2 game type that includes this problem is Chicken.