EconomicModel

[Question] What are good mod­els of col­lu­sion in AI?

• EconomicModel 22 Jun 2021 21:55 UTC

  1 point

  in reply to: Neel Nanda’s comment on: The Point of Trade

  “Diminishing marginal returns” is underspecified. Is this diminishing marginal returns of productivity or utility? Is it in money? Goods? Some other thing? I think it’s diminishing marginal utility in goods, but I’m not sure. Some types of diminishing marginal returns do imply gains from trade, but others don’t.

• EconomicModel 6 Feb 2019 17:18 UTC

  1 point

  in reply to: Stuart_Armstrong’s comment on: Why is this util­i­tar­ian calcu­lus wrong? Or is it?

  capturing all your money is better for the producer than just capturing the producer surplus (and the marginal utility of the consumer surplus to you is sufficiently low that adding it on doesn’t bring the surpluses to a higher number).

  By assumption, the consumer surplus to me is $30. Which is high enough to bring the surpluses to a higher number. I’m not denying that there are slightly different constructions of the problem where donation is the trivially more moral action. That’s not the point, though. The point is that, in this particular scenario (where, by preference fulfillment/​K-H efficiency, buying>donating>keeping my money) my moral intuition says to donate or keep the money. You’re making further assumptions (U(consumer surplus) is approximately 0) which make the problem easier, but less interesting.

• EconomicModel 6 Feb 2019 17:03 UTC

  3 points

  in reply to: Dagon’s comment on: Why is this util­i­tar­ian calcu­lus wrong? Or is it?

  Past a certain point, this is certainly true. But you need a certain degree of reflection before you can tell whether further reflection is likely to produce valuable insights. Apparently you hit your limit, but I haven’t yet. If you have some reason why you think this is a particularly unenlightening thing to think about, I’d love to hear it, but this seems like a matter of different tastes.

  See Vaniver’s comments below his answer for reasons I think this is worth thinking about. I basically agree with them.

• EconomicModel 4 Feb 2019 5:40 UTC

  1 point

  in reply to: Said Achmiz’s comment on: Why is this util­i­tar­ian calcu­lus wrong? Or is it?

  > You seem to be assuming that utility ends up being linear in money, so let’s stick with that assumption.

  > For a function to convert dollars to utils u($), u’($)>0, u″($)<0

  ^^^non-linear function^^^

  ^^^U″($)!=0^^^

  [/​snark]

  That is important, though. The whole point of the thought experiment was that the strictly selfish result (I buy $100 of ice cream) is different from the Kaldor-Hicks/​utility efficient outcome (I overpay for a widget) in a situation where my (normally very utilitarian) moral intuition backs the selfish action. Your scenario is only equivalent if you take the U″($)=0 condition which I explicitly rejected.

• EconomicModel 4 Feb 2019 5:29 UTC

  1 point

  in reply to: Stuart_Armstrong’s comment on: Why is this util­i­tar­ian calcu­lus wrong? Or is it?

  >It’s certainly not more moral (because the extra benefits flow to you, and that is generally not seen as a moral plus).

  Not in the calculus as I run it. I consider a util to me, a util to Warren Buffett, and a util to an impoverished african farmer to be equivalent (insofar as interpersonal utility comparisons are possible etc. etc.). The only reason I consider a dollar donation to GiveDirectly>a dollar spent on ice cream>a dollar donation to Warren Buffett’s personal checking account is because “for a function to convert dollars to utils u($), u’($)>0, u″($)<0”.

  >I believe the empirical evidence is that that argument is wrong, but it’s certainly not wrong in theory.

  What empirical evidence? It’s a contrived thought experiment, not something I’m actually debating.

  > the total value added is not relevant to the receivers, just to yourself, unless you use that extra value to trade or donate more.

  Again, this is just plain wrong. Utilitarianism!=self-flagellation.

• EconomicModel 4 Feb 2019 5:21 UTC

  1 point

  in reply to: Dagon’s comment on: Why is this util­i­tar­ian calcu­lus wrong? Or is it?

  Some context. I do not, in fact, believe that Kaldor-Hicks efficent actions are inherently moral. But I do think that Kaldor-Hicks efficiency is a pretty good first-pass heuristic. This thought experiment was meant to set up a dilemma between Kaldor-Hicks efficiency (which says to buy) and my moral intuitions (which says not to buy). The problem is that I can’t figure out exactly what my intuition is trying to tell me about what seems to be a fairly straightforward utility-maximizing transfer. For the purposes of this contrived thought experiment, suppose that the only decision is whether or not to buy from the commune. There isn’t an option to donate some or all of the money to GiveDirectly if I choose not to buy. Just buy a widget or buy $100 worth of ice cream.

[Question] Why is this util­i­tar­ian calcu­lus wrong? Or is it?