Less Wrong is a community blog devoted to refining the art of human rationality. Please visit our About page for more information.

Comment author: Riothamus 05 April 2017 10:01:30PM 0 points [-]

I disagree, for two reasons.

  1. AI in conflict is still only an optimization process; it remains constrained by the physical realities of the problem.

  2. Defense is a fundamentally harder problem than offense.

The simple illustration is geometry; defending a territory requires 360 degrees * 90 degrees of coverage, whereas the attacker gets to choose their vector.

This drives a scenario where the security trap prohibits non-deployment of military AI, and the fundamental problem of defense means the AIs will privilege offensive solutions to security problems. The customary response is to develop resilient offensive ability, like second-strike...which leaves us with a huge surplus of distributed offensive power.

My confidence is low that catastrophic conflict can be averted in such a case.

Comment author: roystgnr 06 April 2017 09:28:21PM 1 point [-]

The simple illustration is geometry; defending a territory requires 360 degrees * 90 degrees of coverage, whereas the attacker gets to choose their vector.

But attacking a territory requires long supply lines, whereas defenders are on their home turf.

But defending a territory requires constant readiness, whereas attackers can make a single focused effort on a surprise attack.

But attacking a territory requires mobility for every single weapons system, whereas defenders can plug their weapons straight into huge power plants or incorporate mountains into their armor.

But defending against violence requires you to keep targets in good repair, whereas attackers have entropy on their side.

But attackers have to break a Schelling point, thereby risking retribution from otherwise neutral third parties, whereas defenders are less likely to face a coalition.

But defenders have to make enough of their military capacity public for the public knowledge to serve as a deterrent, whereas attackers can keep much of their capabilities a secret until the attack begins.

But attackers have to leave their targets in an economically useful state and/or in an immediately-militarily-crippled state for a first strike to be profitable, whereas defenders can credibly precommit to purely destructive retaliation.

I could probably go on for a long time in this vein.

Overall I'd still say you're more likely to be right than wrong, but I have no confidence in the accuracy of that.

Comment author: ingive 08 February 2017 03:42:42PM *  0 points [-]

And -2 * $-1 = $-2, yes.

I think you meant (-2 * -1 = $2) I meant, multiply by a negative count of debt and not itself. So a debt multiplied by a negative count of debts leads to no debt at all, a positive. I'm not sure how you can have a negative count of debts.

$2 debt squared does make sense, though, it is $4 and no debt. So by our mathematics, I could call the bank and ask them to multiply my debt with yours, I would return a positive.

The point I am making is that we've made it this way because have chosen to. It says to me that mathematics is more of a mental creation, albeit a very useful one and that nature might be infinitely greater than our own self-imposed boundaries.

Take a look at this picture does the water goes up or down? Is bivalent thinking necessarily nature or simply a mental creation? When it comes to truths (true or false) or computers (by primordial decisions) (1 or 0)

Comment author: roystgnr 11 February 2017 02:06:55AM *  2 points [-]

$2 debt squared does make sense, though, it is $4 and no debt.

No, it is $$4.

If that's what you meant to write, and it's also obvious to you that you could have written 40000¢¢ instead and still been completely accurate, then I'd love to know if you have any ideas of how this computation could map to anything in the real world. I would have thought that "kilogram meters squared per second cubed" was utter nonsense if anyone had just tried to show me the arithmetic without explaining what it really meant.

If that's not what you meant to write, or if it takes a second to figure out why $$4 isn't 400¢¢ instead of 40000¢¢, then you've just got the illusion of sense going on. And yes, I just noticed that pun and it wasn't intentional.

Comment author: Lumifer 12 December 2016 04:05:50PM *  5 points [-]

Facebook is supposed to have a semblance of privacy

Whaaaaa..?

That looks to me like a fundamental misunderstanding of teh intertubes and, of course, FB itself.

Netiquette has rules about what/how/when you could repost, but that doesn't mean that FB has any kind of privacy.

Comment author: roystgnr 14 December 2016 02:50:36PM 1 point [-]

Facebook has privacy settings, such that anyone who wants to limit their posts' direct visibility can.

Whether you should take someone else's settings as explicit consent should probably vary from person to person, but I think the "if he didn't want it to be widely seen he wouldn't have set it up to be widely seen" heuristic is probably accurate when applied to EY, even if it's not applicable to every Joe Grandpa.

Even in the Joe Grandpa case, it doesn't seem like merely avoiding citing and moving on is a good solution. If you truly fear that someone is sharing more to the world than they intend to, the kind thing to do is inform them and help them fix it, not ignore them and pray that everyone else who stumbles upon it shares your sense of decorum.

Comment author: Luke_A_Somers 15 November 2016 03:55:54PM *  1 point [-]

It would have been even nicer if they'd changed to a system with fewer pathologies than IRV, like Schulze or Ranked Pairs or Score or Approval...

but I'll take what I can get!

Comment author: roystgnr 18 November 2016 06:03:37PM 0 points [-]

The form of the pathologies makes a difference, no?

IIRC the worst pathology with IRV in a 3 way race is basically: you can now safely vote for a third party who can't possibly win, but if they show real support then it's time to put one of the two Schelling party candidates in front of them. So it's not worse than plurality but it's not much of an improvement. Plus, with more candidates IRV becomes more and more potentially insane.

With Schulze or Ranked Pairs the pathology is DH3: If a third party can win, you can often help them win by voting a "dark horse" candidate ahead of their two real competitors... but if enough people try that then the "dark horse" actually wins. That could actually end up much worse than plurality.

With Score the unique pathology is: If you know how to vote tactically then your vote is almost certain to look like an approval vote. So the only reason to use it rather than Approval is to encourage non-game-theory-geeks to sabotage themselves and water down their own votes.

With Approval (and Score, to a slightly lesser extent) the pathology is the "Burr dilemma". Okay, if Approval voters are good at tactical voting and there are many iterations of accurate public polls, then they pick the Condorcet winner... but if either of those things aren't true then people attempting to vote tactically can end up hurting front-runner candidates and electing a "dark horse".

I'd say this all starts to make plurality look reasonable by comparison, because the two party system tends to only fail by electing overly-conservative (in the Burkean sense) candidates, which is a less dangerous failure than its opposite would be, but I admit there is a recent very-salient counterexample.

Comment author: James_Miller 19 June 2016 08:14:09PM 4 points [-]

Huge, thin mirrors in orbit over the equator that reflect light that would have hit the equator to the area of earth experiencing winter.

Comment author: roystgnr 20 June 2016 07:05:51PM 2 points [-]

There's a high-stakes variational calculus problem. For what seasonal temperature profile do we get the best long-term minimum for the sum of "deaths due to extreme cold" and "deaths due to tropical diseases whose vector insects are stymied by extreme cold".

Comment author: gwern 29 April 2016 12:12:31AM *  2 points [-]

And they also vary CO2 levels systematically by geography as well; if that was enough for a detectable effect on IQ, then the lower CO2 levels around Denver should make the rest of us at lower altitudes, such as sea level, look obviously handicapped. If you believe the altitude point refutes effects of oxygen, then it must refute effects of carbon dioxide and nitrogen as well...

Which is part of my original point about implausible effect sizes: the causal effect is underidentified, but whether it's oxygen or CO2 or nitrogen, it is so large that we should be able to see its repercussions all over in things like the weather (or altitude, yes).

Comment author: roystgnr 29 April 2016 03:12:50PM 1 point [-]

The magnitude of the variation isn't nearly the same in the O2 vs CO2 cases. "16% O2 reduction is lost in the noise" is devastating evidence against the theory "0.2% O2 reduction has significant cognitive effects", but "16% CO2 reduction is lost in the noise" is weaker evidence against the theory "66% and 300% CO2 increases have significant cognitive effects".

I'm not arguing with you about implausible effect sizes, though. We should especially see significant seasonal effects in every climate where people typically seal up buildings against the cold or the heat for months at a time.

Comment author: gwern 27 April 2016 12:13:53AM 1 point [-]

Small proportional changes seem unlikely to drive big effects, unless there is some feedback mechanism that is keeping the level precisely balanced.

Such as in the body, dealing with tightly regulated and critical aspects of metabolism like oxygen consumption.

But 1% changes in oxygen should be happening all over the place.

Perhaps they are. You don't know the effect because the existing experiments do not vary or hold constant oxygen levels. All you see is the net average effect, without any sort of partitioning among causes.

Comment author: roystgnr 28 April 2016 03:20:22PM 1 point [-]

You don't know the effect because the existing experiments do not vary or hold constant oxygen levels. All you see is the net average effect, without any sort of partitioning among causes.

Existing experiments do vary oxygen levels systematically, albeit usually unintentionally, by geography. Going up 100 meters from sea level gives you a 1% drop in oxygen pressure and density. If that was enough for a detectable effect on IQ, then even the 16% lower oxygen levels around Denver should leave Coloradans obviously handicapped. IIRC altitude sickness does show a strong effect on mental performance, but only at significantly lower air pressures still.

Comment author: gwern 21 April 2016 05:59:02PM *  6 points [-]

It doesn't matter if some of the group differences are genetic in origin, given that others are not, we can still resolve those.

Those can be resolved but they will not make nearly as large a difference as currently expected, where current ideologies hold that all of that 3-190x per capita difference is due to environmental conditions, history, and racism. HBD implies that, just as with individual differences and the systematic failure of welfare and education randomized experiments to 'close the gap', we can expect this futility to occur on a country-level basis at some level of development. Countries like China (maybe) and North Korea (definitely) will be predicted to escape their current poverty levels with appropriate interventions... and countries like Subsaharan Africa to possibly not escape. (Which countries can be made more concrete in a HBD context by taking Piffer's country/group-level polygenic scores and looking at the residuals of a GDP/score regression for the countries which most over and underperform; the former can be predicted to not grow substantially, and the latter can be predicted to grow substantially.)

Remember how heritability works. If environments improve, genetics will explain more and more of variance. It's Liebig's barrel. Shared-environment in the USA is very small.

Do the implications here change?

Yes, because those environmental factors are causally downstream and cannot be improved without the locals. As development aid has discovered again and again, you cannot force improvements on a country. Pakistan, for example, is so dysfunctional and clannish that iodization and polio programs have had serious trouble making any headway.

The steps we should take don't depend on HBD being true or false.

Yes, they do! These causal models are fundamentally different. If genetics is a major limiting factor, iodine and all other environmental factors are not going to help past a certain level of development. (You can feed some Americans or New Zealanders iodine supplements, but it won't give them +10 IQ points even though they are probably somewhat deficient). If genetics is the major limiting factor, then at a certain point, you are basically polishing a turd and this can either be accepted or more radical interventions must be considered.

In response to comment by gwern on Suppose HBD is True
Comment author: roystgnr 21 April 2016 07:02:42PM -2 points [-]

Pakistan, for example, is so dysfunctional and clannish that iodization and polio programs have had serious trouble making any headway.

To be fair, that's not entirely Pakistanis' fault. Is paranoia about Communist fluoridation plots more or less dysfunctional than paranoia about CIA vaccination plots? Does it make a difference that only the latter has a grain of truth to it?

In response to Positivity Thread :)
Comment author: username2 09 April 2016 11:13:57AM 2 points [-]

What are your favorite puns?

Comment author: roystgnr 12 April 2016 09:02:21PM 6 points [-]
Comment author: roystgnr 12 April 2016 08:38:30PM *  0 points [-]

This looks like a special case of a failure of intentionality. If a child knows where the marble is, they've managed first-order intentionality, but if they don't realize that Sally doesn't know where the marble is, they've failed at second order.

The orders go higher, though, and it's not obvious how much higher humans can naturally go. If

Bob thinks about "What does Alice think about Bob?" and on rare occasions "What does Alice think Bob thinks about Alice?" but will not organically reason "What does Alice think Bob thinks about Alice's model of Bob?"

then Bob can handle second and third but can't easily handle fourth order intentionality.

It may be a useful writing skill to be comfortable with intentionality at one level higher than your audience.

View more: Next