Oceans Yet to Burn

[Oceans Yet to Burn]

mitigatedchaos asked:

What is your opinion of me? And what do you think of my plot to replace congress with a new legislature of political party/think tank hybrids that bet on outcomes of legislation (as outlined in my "The National Delegation" posts, where I still owe squid another post but I won't have sit down access to a computer for a while)?

xhxhxhx answered:

on The National Delegation:

• I think it’d be hard to legitimate such a goofy system ex nihilo and I don’t think its performance would be enough to legitimate it in action
• I have a bunch of cavils about specification: I don’t think you could define policy outcomes or contract conditions with sufficient precision to make the market deep or efficient under most outcomes, and I don’t think you can define values with sufficient precision to make that constraint binding 
• I don’t think the policy outcomes would be much better, except to the extent that you might fix the system by excluding ‘incoherent’ or ‘imprecise’ values – which would mean the system wouldn’t really be doing what it promises
• I’m deeply skeptical of any regulatory system that’s this hard and finicky – it’s like you’re putting the FDA or the FTC in charge of the whole system of government, and I don’t think market discipline is enough to get federal agencies to behave themselves
• I think its worse than electoral democracy, although that’s not an especially strong belief, and I think it’d underperform purer technocracies or a purer liberal states

on you:

• you put effort and thoughtfulness into your work – that’s good and that’s rare
• I’m too much of a liberal to appreciate your commentary
  
• you’re very kind and thoughtful, and I haven’t done enough to return the favor

[mitigatedchaos]

Oh, there appears to be a point of confusion - values are informal, not rigidly specified. And values are not explicitly what is bet on. It’s more along the lines of both parties claiming their legislation would reduce gun crime, as they often do, and betting against each other and with amounts, and “gun crime” is defined as a bundle of metrics to prevent min-maxing. It occurs to me that because so much policy is just flat out wrong at achieving its supposed aims, some fairly large improvements should be possible with even that much of a check on whether it works.

Ofc, there is also the question of what a purer technocracy would look like.

[Oceans Yet to Burn]

The National Delegation

[mitigatedchaos]

In case you haven’t noticed recently, democracy has major issues.  Every major developed state is strewn with dysfunction and programs that are actively at odds with their intended purposes.  Our politicians are either incompetent idiots or shrewd operators working against our interests.

Policies routinely have reasonable stated values, but terrible efficacy.

Organizations such as the RAND Corporation knew the Iraq War would be a lot tougher than the Bush administration said it would be.  Policy plans coming out of think tanks seem to be better than the actual policies we get.

If we didn’t know they’d immediately get subverted, we’d almost be better off with think tanks running the country.

Better results are necessarily different results, and systems produce the outcomes they incentivize, so to change the results it is necessary to change the system.

The truth is, it may be possible to get something like think tanks in charge of the government, a hybrid between them and political parties, but we will have to add selection pressure to ensure they work towards correctness.

I propose a new legislature, composed of a new kind of corporate entity, the Delegate Candidate Organization (DCO).  

Every three years, at election time, each voter delegates their vote to a DCO.  The top 50 Delegate Candidate Organizations then form the legislature, becoming that term’s Delegate Organizations.  This legislature is known as the National Delegation.

In a second election, those DCOs that did not make the cut delegate their votes to members of the top 50.

(In an optional alternative, the vote could be split between DCOs by categories by voters, allowing a truly innovative level of representation.  Bills would have to pass on all categories to pass, and the tax category would determine how funding is obtained, but not total expenditures.  Sadly, this is probably too complex for typical voters.)

A Delegate Candidate Organization receives its funding exclusively from the State.  For each delegated vote it receives, the DCO receives $5 in annual funding, and an additional $5 times its percentile standing in a legislative outcome prediction market.

(That might sound like a lot.  America has around 300 million people, so you could potentially be looking at three billion dollars.  I would answer that the 2016 Presidential election cost $2.6 billion by itself, and that money had to come from somewhere and is already influencing our political process.  The size of the US economy is $18,570 billion dollars.  The real question is whether better policy by the DCOs could improve that by 0.016% or more, which would make the National Delegation pay for itself.  I believe that it would.)

The key factor that makes DCOs behave more like think tanks is that a significant chunk of their funding depends on correctly estimating the outcomes of legislation.  What keeps them honest?  First, competition with other DCOs that will pressure them against spoiling the metrics.  Second, voters.

When a piece of legislation is to be passed, DCOs make predictions on outcomes and bet on them in a virtual currency called Credibility Score (or just “Cred”).  Each outcome must be represented by a basket of multiple metrics, to prevent min-maxing.

This structure allows us to build a differentiation between a policy’s values and its efficacy.  Previous discourse has often viewed policy as solely a matter of efficacy, but of course in practice people have different preferences and are not a unified mass just waiting for enlightenment into [your political ideology].  Preserving the values component (in part through voting) also allows bits of efficacy that have slipped through to be represented on the other side of the equation.

The bets serve two purposes.  The first is to reward policymakers that are actively effective at achieving their stated objectives, and punish policymakers that are too unaligned with reality.  The second is to effectively tell voters what the plans will actually do, not just wishy washy language pols want people to hear.

“This bill will reduce gun crime.”
“By how much?”
“Uh… a, uh, lot.”

Not only can the DCO specify what its % estimate for a decrease in gun crime is, but it can also communicate its level of certainty - by how much it bets on the outcome as a percentage of its current Cred reserves, data that can be mined by political scientists and journalists.

DCOs must be able to amend predictions when new legislation is passed.  A court will also be required to punish those who tamper with metrics, and resolve other disputes.  The details of that are a challenge in themselves, but should be feasible to work out.

Each DO has as many votes in the legislature as have been delegated to it.  A majority is required to pass legislation.

The accumulated Credibility Score/Cred across all bets is used to determine the percentile standing of all DCOs, used to determine funding (as above).  Percentile standing is listed on the ballot next to the DCO’s name, but to simplify things for voters, DCOs are listed in the order of votes received in the previous election.

---

Practical experiments will be necessary to assess the viability of this model, but I have high hopes for it.  If we want to advance as a civilization, then we must develop new organizational technologies.

[mitigatedchaos]

Think you need to take a closer look at Robin Hanson, something I thought I’d never say

Specifically, the problem is that predicting the results isn’t the issue, it’s predicting the change in results given some policy change

I think Hanson has people bet on outcome both with and without policy

I may have to look into that, but it doesn’t sound unreasonable. Betting for outcomes based on whether the bill passes or fails to pass certainly provides more information for our voters/etc.

One big problem is that people are going to use this not to predict, but to hedge

It will be financialized

If you believe Hanson that markets are perfect, that’s not a problem it will all work out

if you haven’t had your skull smashed with a brick every day for the past 20 years or worked in the econ dept at GMU, you should be skeptical.

Sorry, I guess I should have been more clear in my intentions earlier.

While the probability estimates produced by the prediction market are interesting, the real purposes are more like: 

1. Punish politicians that are actively at odds with the truth/reward those who have some idea what they’re doing, so that eventually the system is dominated by more clueful politicals who spend less time huffing ideology.  Hopefully, this will result in more effective policy which is more aligned with reality.

(I’m of the opinion that there are many policies that it’s said you can’t do, because markets etc, but which you could do if you were smart about it.  So I want those to come up, actually testing some of these policies before they come up, etc.)

2. Make politicians be more specific and truthful about the outcomes of policies in measurable ways, making it more difficult to do one thing and say another.

3. Track the effectiveness of policies over time so that better policy can be created in the future (through the metrics gathered to feed the market, not the market itself).

Would hedging interfere with those?  I’m not so sure.  It is, itself, information.  It may also depend on the market’s design itself.

[mitigatedchaos]

@collapsedsquid

Alright, then you’re gonna have the problem of “who gets to decide what comes up for prediction and how?” with the various possibilities for manipulation.

Yes, a challenge in itself.  My opinion is that it must be easier to get stuff into the prediction pool than it is to pass the legislation.  Otherwise, it just degrades to normal legislature with some fluff on top.

So, off the top of my head, it may require 30-40% approval to get an item into the prediction pool, perhaps with a limit on the number of items each DCO can put into the pool.

Second and related is that you can basically rewarding people who are connected rather than accurate

To some extent, this doesn’t matter, connections are a part of effective policy too, much as I wish they were not

But it comes down to who can manipulate the outcomes and who has the inside track on what people will do.

- court will be needed so they can sue each other when they cheat

- baskets of metrics harder to game than single metric, so all metrics must be baskets

- hard to actually game some of the more challenging ones by outside interference if metric collection is at all accurate, simply too costly, borders on cost of actually fixing the problem

I’ll expand on this when I have access to an actual computer, which will be a while.

[Oceans Yet to Burn]

Anonymous asked:

i really don't understand your focus on percentile rankings of prediction market performance rather than the raw results. it adds a distortion to the incentives of whatever underlying prediction market system you use and LMSR and such are designed to have incentives for accuracy already.

I want the DCOs hungry enough for money to [edit: make better policy], but not so ludicrously hungry that they’ll work hard to sabotage all the metrics and cause the state itself to become delusional.

Does that make sense?

Therefore, while their pay must be coupled to their performance, it needs a layer of indirection.  They bet with a fake currency that can’t be directly converted into real money, and they are paid based on their overall performance over multiple bets, which is unlikely to rise or fall quite so sharply that one of their personnel will freak out and cause an incident of corruption.  

There may be other ways to accomplish this goal.

[Oceans Yet to Burn]

The National Delegation

In case you haven’t noticed recently, democracy has major issues.  Every major developed state is strewn with dysfunction and programs that are actively at odds with their intended purposes.  Our politicians are either incompetent idiots or shrewd operators working against our interests.

Policies routinely have reasonable stated values, but terrible efficacy.

Organizations such as the RAND Corporation knew the Iraq War would be a lot tougher than the Bush administration said it would be.  Policy plans coming out of think tanks seem to be better than the actual policies we get.

If we didn’t know they’d immediately get subverted, we’d almost be better off with think tanks running the country.

Better results are necessarily different results, and systems produce the outcomes they incentivize, so to change the results it is necessary to change the system.

The truth is, it may be possible to get something like think tanks in charge of the government, a hybrid between them and political parties, but we will have to add selection pressure to ensure they work towards correctness.

I propose a new legislature, composed of a new kind of corporate entity, the Delegate Candidate Organization (DCO).  

Every three years, at election time, each voter delegates their vote to a DCO.  The top 50 Delegate Candidate Organizations then form the legislature, becoming that term’s Delegate Organizations.  This legislature is known as the National Delegation.

In a second election, those DCOs that did not make the cut delegate their votes to members of the top 50.

(In an optional alternative, the vote could be split between DCOs by categories by voters, allowing a truly innovative level of representation.  Bills would have to pass on all categories to pass, and the tax category would determine how funding is obtained, but not total expenditures.  Sadly, this is probably too complex for typical voters.)

A Delegate Candidate Organization receives its funding exclusively from the State.  For each delegated vote it receives, the DCO receives $5 in annual funding, and an additional $5 times its percentile standing in a legislative outcome prediction market.

(That might sound like a lot.  America has around 300 million people, so you could potentially be looking at three billion dollars.  I would answer that the 2016 Presidential election cost $2.6 billion by itself, and that money had to come from somewhere and is already influencing our political process.  The size of the US economy is $18,570 billion dollars.  The real question is whether better policy by the DCOs could improve that by 0.016% or more, which would make the National Delegation pay for itself.  I believe that it would.)

The key factor that makes DCOs behave more like think tanks is that a significant chunk of their funding depends on correctly estimating the outcomes of legislation.  What keeps them honest?  First, competition with other DCOs that will pressure them against spoiling the metrics.  Second, voters.

When a piece of legislation is to be passed, DCOs make predictions on outcomes and bet on them in a virtual currency called Credibility Score (or just “Cred”).  Each outcome must be represented by a basket of multiple metrics, to prevent min-maxing.

This structure allows us to build a differentiation between a policy’s values and its efficacy.  Previous discourse has often viewed policy as solely a matter of efficacy, but of course in practice people have different preferences and are not a unified mass just waiting for enlightenment into [your political ideology].  Preserving the values component (in part through voting) also allows bits of efficacy that have slipped through to be represented on the other side of the equation.

The bets serve two purposes.  The first is to reward policymakers that are actively effective at achieving their stated objectives, and punish policymakers that are too unaligned with reality.  The second is to effectively tell voters what the plans will actually do, not just wishy washy language pols want people to hear.

“This bill will reduce gun crime.”
“By how much?”
“Uh… a, uh, lot.”

Not only can the DCO specify what its % estimate for a decrease in gun crime is, but it can also communicate its level of certainty - by how much it bets on the outcome as a percentage of its current Cred reserves, data that can be mined by political scientists and journalists.

DCOs must be able to amend predictions when new legislation is passed.  A court will also be required to punish those who tamper with metrics, and resolve other disputes.  The details of that are a challenge in themselves, but should be feasible to work out.

Each DO has as many votes in the legislature as have been delegated to it.  A majority is required to pass legislation.

The accumulated Credibility Score/Cred across all bets is used to determine the percentile standing of all DCOs, used to determine funding (as above).  Percentile standing is listed on the ballot next to the DCO’s name, but to simplify things for voters, DCOs are listed in the order of votes received in the previous election.

---

Practical experiments will be necessary to assess the viability of this model, but I have high hopes for it.  If we want to advance as a civilization, then we must develop new organizational technologies.