The "overlearning the game" problem • Hacker News Board

Re: The "overlearning the game" problem

#21

Post by apsurd » Mon, Aug 29, 2011, 6:46 AM UTC

wisty wrote: post #10

It's related to Goodhart's Law: Any observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes. This is often the result of attempting to overoptimize a system. You can optimize a race car to a huge degree, because you know exactly what you want it to do. You can't optimize a schooling system, because you don't know exactly what you want it to do. A little noise is a go…

Good comment. I smiled at the sheer simplicity of injecting randomness to mitigate the >95% in order to deter exploitation. Of course it would be an interesting experiment to conduct and compare results against.

Also regarding your racecar example; they've gotten so good at making them go fast they actually are required to have restrictor plates (http://en.wikipedia.org/wiki/Restrictor_plate) so this fits the example of exploiting a well-defined goal to the point it's no longer competitive/safe/fun.

I find that quite amusing and cool. (I don't follow the sport I just like watching science-type shows on tv)

Re: The "overlearning the game" problem

#22

Post by DasIch » Mon, Aug 29, 2011, 6:54 AM UTC

sendos wrote: post #11

kragen wrote: post #8

Earlier quoted context omitted.

It's pretty bizarre that you're using the derogatory definition of "hack" given that you're posting on Hacker News . And it doesn't incline me to want to talk to you further.

Sorry, I was just using the colloquial meaning of hack. Did not at all mean to appear derogatory for the type of hacking exemplified by HN, of which I take part in. In any case, I do see your point about similarities between hacking and overlearning, but I think there are some differences. If I can articulate those difference better, I'll reply a little later.

I think overlearning can be considered to be a subset of hacking. In both cases laws (as in physics) are exploited however while hacking may or may not break rules as imposed by society overlearning does not.

Re: The "overlearning the game" problem

#23

Post by jacques_chester » Mon, Aug 29, 2011, 6:55 AM UTC

drewcrawford wrote: post #12

If you pay a man by the hour, he'll work a lot of hours. If you pay him by the brick, he'll lay a lot of bricks. These "games" are basically the equivalent of counting lines of code or checkins. We're measuring poor proxies instead of the things we're actually interested in. The solution isn't an arms race to build bigger and better proxies, the solution is to measure real things instead of artificial ones. Here's ju…

The idea of directly polling everybody is not new -- that 'everyone' could vote on everything was at the core of Athenian government.

The US founders specifically rejected that system because it leads to majoritarianism. And it would do so again, if you let it.

Re: The "overlearning the game" problem

#24

Post by drewcrawford » Mon, Aug 29, 2011, 7:15 AM UTC

jacques_chester wrote: post #23

drewcrawford wrote: post #12

If you pay a man by the hour, he'll work a lot of hours. If you pay him by the brick, he'll lay a lot of bricks. These "games" are basically the equivalent of counting lines of code or checkins. We're measuring poor proxies instead of the things we're actually interested in. The solution isn't an arms race to build bigger and better proxies, the solution is to measure real things instead of artificial ones. Here's ju…

The idea of directly polling everybody is not new -- that 'everyone' could vote on everything was at the core of Athenian government. The US founders specifically rejected that system because it leads to majoritarianism. And it would do so again, if you let it.

> The US founders specifically rejected that system because it leads to majoritarianism.

It would be wonderful if you could cite a few primary sources discussing a direct vs indirect legislative branch as leading to majoritarianism, whether a longitudinal study or quotation of the appropriate historical person.

You can still have separation of powers, judicial oversight, bills of rights, local governments, federalism, and all the checks we have today against majoritarianism within a direct legislative system. Direct legislature doesn't mean only one branch of government.

To put it another way, I don't believe the solution to the "tyranny of the majority" is to set up little tyrants to rule over the majority instead. Even if those were the only two options (and they're not), I would prefer majority rule to what we have.

Re: The "overlearning the game" problem

#25

Post by jowiar » Mon, Aug 29, 2011, 7:16 AM UTC

jacques_chester wrote: post #23

drewcrawford wrote: post #12

If you pay a man by the hour, he'll work a lot of hours. If you pay him by the brick, he'll lay a lot of bricks. These "games" are basically the equivalent of counting lines of code or checkins. We're measuring poor proxies instead of the things we're actually interested in. The solution isn't an arms race to build bigger and better proxies, the solution is to measure real things instead of artificial ones. Here's ju…

The idea of directly polling everybody is not new -- that 'everyone' could vote on everything was at the core of Athenian government. The US founders specifically rejected that system because it leads to majoritarianism. And it would do so again, if you let it.

Further along these lines, modern communication has caused the US political system to more closely approach direct democracy, causing problems along the lines of those foreseen.

Re: The "overlearning the game" problem

#26

Post by tomjen3 » Mon, Aug 29, 2011, 7:24 AM UTC

wisty wrote: post #10

It's related to Goodhart's Law: Any observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes. This is often the result of attempting to overoptimize a system. You can optimize a race car to a huge degree, because you know exactly what you want it to do. You can't optimize a schooling system, because you don't know exactly what you want it to do. A little noise is a go…

I don't think it is a good idea at all. If the political power boiled down to picking paper over scissors, the loosing side may very well ascribed no legitimacy to the outcome. That is almost certainly going to end far worse than actually counting all the votes (i am looking at you, supreme court).

The same problem happens with your exam example. Why is it fair that someone less competent than you got in? A better solution would be to let everyone over the cut off point in.

I can't disagree with your example of the stock market, so long as the trades were still executed with a few seconds.

Re: The "overlearning the game" problem

#27

Post by jacques_chester » Mon, Aug 29, 2011, 7:24 AM UTC

drewcrawford wrote: post #24

jacques_chester wrote: post #23

Earlier quoted context omitted.

The idea of directly polling everybody is not new -- that 'everyone' could vote on everything was at the core of Athenian government. The US founders specifically rejected that system because it leads to majoritarianism. And it would do so again, if you let it.

> The US founders specifically rejected that system because it leads to majoritarianism. It would be wonderful if you could cite a few primary sources discussing a direct vs indirect legislative branch as leading to majoritarianism, whether a longitudinal study or quotation of the appropriate historical person. You can still have separation of powers, judicial oversight, bills of rights, local governments, federalism…

> You can still have separation of powers, judicial oversight, bills of rights, local governments, federalism, and all the checks we have today against majoritarianism within a direct legislative system.

This doesn't change the fact that legislation would still take on a majoritarian tone. You'd need even more circumscription of the powers of the legislature for it to work at all; history suggests that it's at best a partial solution.

> I would prefer majority rule to what we have.

I saw majority rule in high school. No thanks.

Re: The "overlearning the game" problem

#28

Post by jacques_chester » Mon, Aug 29, 2011, 7:27 AM UTC

jowiar wrote: post #25

jacques_chester wrote: post #23

Earlier quoted context omitted.

The idea of directly polling everybody is not new -- that 'everyone' could vote on everything was at the core of Athenian government. The US founders specifically rejected that system because it leads to majoritarianism. And it would do so again, if you let it.

Further along these lines, modern communication has caused the US political system to more closely approach direct democracy, causing problems along the lines of those foreseen.

I think polling is what has done it, because it shortcuts the feedback cycle between representative and represented.

Edmund Burke made the point eloquently in his speech to the voters of Bristol:

"Certainly, gentlemen, it ought to be the happiness and glory of a representative to live in the strictest union, the closest correspondence, and the most unreserved communication with his constituents. Their wishes ought to have great weight with him; their opinion, high respect; their business, unremitted attention. It is his duty to sacrifice his repose, his pleasures, his satisfactions, to theirs; and above all, ever, and in all cases, to prefer their interest to his own.

But his unbiassed opinion, his mature judgment, his enlightened conscience, he ought not to sacrifice to you, to any man, or to any set of men living. These he does not derive from your pleasure; no, nor from the law and the constitution. They are a trust from Providence, for the abuse of which he is deeply answerable. Your representative owes you, not his industry only, but his judgment; and he betrays, instead of serving you, if he sacrifices it to your opinion."

[emph added]

http://press-pubs.uchicago.edu/founders/documents/v1ch13s7.h...

Burke is already presaging the exact topic we're discussing: whether democracy is about the direct will of the people, or whether an elected representative has duties other than merely voting as his electors wish him to.

Re: The "overlearning the game" problem

#29

Post by david927 » Mon, Aug 29, 2011, 7:32 AM UTC

drewcrawford wrote: post #24

jacques_chester wrote: post #23

Earlier quoted context omitted.

The idea of directly polling everybody is not new -- that 'everyone' could vote on everything was at the core of Athenian government. The US founders specifically rejected that system because it leads to majoritarianism. And it would do so again, if you let it.

> The US founders specifically rejected that system because it leads to majoritarianism. It would be wonderful if you could cite a few primary sources discussing a direct vs indirect legislative branch as leading to majoritarianism, whether a longitudinal study or quotation of the appropriate historical person. You can still have separation of powers, judicial oversight, bills of rights, local governments, federalism…

That's a fantastic response. I'm shocked at how many people have unsubstantiated fears of simply eliminating an unnecessary proxy. The reason we didn't have direct democracy before is because it wasn't practical before -- not because those elected are superior and can ignore what the people wish for.

It is now practical to have a direct democracy: If we can bank online, we can vote online. In contrast, there are 26 lobbyists for every member of congress in the US. They are not there to support the public's will but to contravene it by paying for the opposite. Over 80% of Americans (statistically almost everyone) are dissatisfied with Congress, as they should be. The representational form of democracy is both no longer needed and horribly broken.

Re: The "overlearning the game" problem

#30

Post by tomjen3 » Mon, Aug 29, 2011, 7:36 AM UTC

drewcrawford wrote: post #24

jacques_chester wrote: post #23

Earlier quoted context omitted.

The idea of directly polling everybody is not new -- that 'everyone' could vote on everything was at the core of Athenian government. The US founders specifically rejected that system because it leads to majoritarianism. And it would do so again, if you let it.

> The US founders specifically rejected that system because it leads to majoritarianism. It would be wonderful if you could cite a few primary sources discussing a direct vs indirect legislative branch as leading to majoritarianism, whether a longitudinal study or quotation of the appropriate historical person. You can still have separation of powers, judicial oversight, bills of rights, local governments, federalism…

Direct democracy killed Socraties. The current system, flawed as it is, does a reasonable job of prevention the most stupid ideas from getting Mainstream attention.

The people deciding political issues wouldn't be the reasonable and smart people who hang around here. It will be the people who stand around picking funerals, people who never graduated high school, functional illiterates and the drug dealers in the inner cities.

Is it really that kind of people, the ones who can be convinced with a few late night tv commercials, you want to run your country?