I made two visual guides that could be used to improve online discussions. These could be dropped into any conversation to (hopefully) make the discussion more productive.
I improved the lay-out of the old image and added a top layer for steelmanning. You can find my reasoning here and a link to the pdf-file of the image here.
Jobst and I want to improve AI-safety by supplementing RLHF with a consensus generating voting system. Last week we did a small experiment at a conference. Here is the poster we used to explain this idea to the attendants:
Before you start cheering I should mention that the Farmer–Citizen Movement (who are very conservative when it comes to animal rights) have burst onto the scene with 16 seats (making them the largest party).
With farming and livestock becoming a hot button issue in the Netherlands there's a chance that animal rights will now become a polarizing issue, with a lot of people who previously didn't think about it becoming explicitly for or against expanding animal rights. While this would increase the amount of vegetarians and vegans, it remains to be seen if this will turn out positive for animal welfare overall.
EDIT: Biden Backs $8 Billion Alaska Oil Project. I don't know why someone gave this shortform an immediate -9 downvote, but for those EAs that still care about climate change, thank you.
A massive and controversial new oil production project in Alaska is under review by the US Department of the Interior.
ConocoPhillips' massive Willow Project would be a climate disaster, locking in at least 30 years of fossil fuel production on sensitive Arctic ecosystems near Indigenous communities. It would unleash high levels of pollution - roughly the equivalent of 66-76 coal plants worth of carbon to the air - and directly undermine President Biden's climate goals.
At more than six and a half thousand words I think the transcription would be a bit long and I think in general the linkposts should link to the primary source.
But I appreciate the link and I will use that site in the future. This forum doesn't allow me to pin this comment, so I'll upvote instead. It's a pretty short post so I'm sure people will find and upvote your link rather quickly, but if this post becomes unexpectedly popular and this comment sinks to the bottom of the comments, I'll copy-paste the links into the post.
Allowing anonymous predictions causes a whole bunch of other problems. But even we could somehow get rid of coordination mechanisms like: dominance assurance contracts, the fear of losing your social network, or psychological loyalty towards your ingroup, is it really in your own interest to lose the source of income for you and your children for a <51% chance of a one time payout, or will the outcomes of conditional prediction markets be biased towards the interests of rich people?
I don't think it does, but this an empirical matter so discussing it has little use and we'll find out once someone does a study.
Sure it can. If you ever see a prediction market which you don't think is measuring the underlying probability of its event, you can make money from it [...] But if you think that a prediction market will be unreliable in cases where any other method is reliable, you can use that to get rich.
That's the thing, I don't think there is another method, I think free will mucks it up. Human beings are what the philosophers call 'frustrators' and in fact there are even non-human entities that are frustrators. Let's say you have to bet on whether a machine will turn on or off, however the machine is a 'frustrator', once you bet that it will turn off it will turn on and vise versa. You aren't predicting the outcome (you literally can't) you are causing the opposite outcome. Humans are the same, if I see that someone predicts I will do X, I might do Y just to assert my autonomy (or maybe not because of mindgames).
So let's say a new prediction market opens on the machine frustrator, no one wants to be the first to make a bet, since you're guaranteed to lose money if no one else participates. Even if people are irrational enough to bet on it, we don't learn anything new and we just wasted a bunch of electricity and manhours for nothing. It's a waste of time and money to bet on frustrators, especially since you can get a 10% return on the stockmarket.
immediate edit: And since humans are frustrators and slowly gain more and more control over the world my question "Does this mean that prediction markets become less useful as humanity progresses, or is there a way to solve this?" still stands.
It will affect the trading and worse it will affect the trading inconsistently so we can't even use mathematics to subtract it. When the president promises not to manipulate the market about event X some people will trust him 90% and bet accordingly while others will trust him 10% and bet according to that. But that's not the same thing as people having 10% credence event X will happen, it can be that you think the president is trustworthy but the event itself is unlikely. On top of that the president might react to the market, e.g if there is a lot of money to be gained he'd risk it but otherwise he wouldn't, but some people might think the president will manipulate the market at 51% when there is some money to be gained, while others think he will only act at 90% when there is a lot of money to be gained (or any other number of percentages), at seemingly random intervals people will suddenly buy or sell stock depending on what the price itself is doing. And the president might react to that etc, it becomes a recursive mindgame instead of being about event X. The resulting percentage cannot be used as a measure of trustworthiness nor as a measure of the underlying probability of event X.
assassinate CEOs
This is always the go-to example defenders of prediction markets use, but it's not representative since it's the most extreme possible version of manipulation. No, I don't think most people will start murdering folk to influence prediction markets, not just because it could land you in jail, but also because it's difficult to execute and most people have an aversion to murder. But you can manipulate prediction markets in much more easy, mundane and legal ways. If I say I will do something and can subsequently win a bunch of money by not doing it, without anyone noticing, that's not even illegal, let alone difficult.
I'm confused by your last paragraph
I'll use the example I used in my other comment:
If people can make it rain and there is e.g a market for every hour of the day, will people really check that every single user who has betted on time T has also betted on time T+1 or else find it suspicious? Because that would mean that now T+1 becomes suspicious if they didn't bet on T+2 etc.
Not only that, it also effects related markets like "will it hail at time T" and if subsequently not betting on that one is suspicious, then markets that are related to that one, like "will it be cold at time T" also become suspicious etc.
So every time a new market is opened and people see that not every user who has betted on related markets has betted on this new market, then everyone will update the related markets which will in turn update their related markets etc. In this model the only way for prediction markets to become accurate is if every user bets on every market and every user is aware of the bets of every other user and is updating on all those bets 24/7.
Suffice to say, if you combine the fact that 1) humans can't instantly know all the new information with 2) the fact we can't know whether the market has updated because of information that we already know, new information, or people updating on the assumption that there is new information, with 3) recursive mindgames and 4) these constant 'ripples' of shifting uncertainties; you'll get so much asynchronous noise that the prediction market becomes an unreliable source. Again, this is only for things we can change, prediction markets will work on things we can't change, like predicting supernovas (until the day we can cause them)
But is it in your individual interest? Is making enemies of your friends and family while losing a guaranteed lifelong stream of income for you and your children really worth it for the possibility of having a one time large payout? What about with different probabilities? What if you think the policy has a 51% chance of helping the poor and a 49% chance of doing nothing. This would be a fantastic policy to try but even without coordination mechanisms like dominance assurance contracts, I doubt a self interested rich person would sacrifice their social network and lifelong stream of income for a 51% of having a one time large payout (actually less than 51% since it probably won't be implemented due to the conditional prediction market).
I made two visual guides that could be used to improve online discussions. These could be dropped into any conversation to (hopefully) make the discussion more productive.
The first is an update on Grahams hierarchy of disagreement
I improved the lay-out of the old image and added a top layer for steelmanning. You can find my reasoning here and a link to the pdf-file of the image here.
The second is a hierarchy of evidence:
I added a bottom layer for personal opinion. You can find the full image and pdf-file here.
Lastly I wanted to share the Toulmin method of argumentation, which is an excellent guide for a general pragmatic approach to arguments
Jobst and I want to improve AI-safety by supplementing RLHF with a consensus generating voting system. Last week we did a small experiment at a conference. Here is the poster we used to explain this idea to the attendants:
Here's the PDF
Reddit user blueshoesrcool discovered that Effective Ventures (the umbrella organization for the Centre for Effective Altruism, 80000 hours, GWWC, etc) has missed its charity reporting deadline by 27 days.
Given that there's already a regulatory inquiry into Effective Ventures Foundation, maybe someone should look into this.
The results of the Dutch provincial elections are in. The Party for the Animals (the party that banned factory farming, but people ignored it) has increased its number of seats in the senate from 3 to 4 (out of 75).
Before you start cheering I should mention that the Farmer–Citizen Movement (who are very conservative when it comes to animal rights) have burst onto the scene with 16 seats (making them the largest party).
With farming and livestock becoming a hot button issue in the Netherlands there's a chance that animal rights will now become a polarizing issue, with a lot of people who previously didn't think about it becoming explicitly for or against expanding animal rights. While this would increase the amount of vegetarians and vegans, it remains to be seen if this will turn out positive for animal welfare overall.
EDIT: Biden Backs $8 Billion Alaska Oil Project. I don't know why someone gave this shortform an immediate -9 downvote, but for those EAs that still care about climate change, thank you.
A massive and controversial new oil production project in Alaska is under review by the US Department of the Interior.
ConocoPhillips' massive Willow Project would be a climate disaster, locking in at least 30 years of fossil fuel production on sensitive Arctic ecosystems near Indigenous communities. It would unleash high levels of pollution - roughly the equivalent of 66-76 coal plants worth of carbon to the air - and directly undermine President Biden's climate goals.
The Biden administration has the power to stop this massive fossil fuel development. Click here to send them a letter (2 minutes).
At more than six and a half thousand words I think the transcription would be a bit long and I think in general the linkposts should link to the primary source.
But I appreciate the link and I will use that site in the future. This forum doesn't allow me to pin this comment, so I'll upvote instead. It's a pretty short post so I'm sure people will find and upvote your link rather quickly, but if this post becomes unexpectedly popular and this comment sinks to the bottom of the comments, I'll copy-paste the links into the post.
Allowing anonymous predictions causes a whole bunch of other problems. But even we could somehow get rid of coordination mechanisms like: dominance assurance contracts, the fear of losing your social network, or psychological loyalty towards your ingroup, is it really in your own interest to lose the source of income for you and your children for a <51% chance of a one time payout, or will the outcomes of conditional prediction markets be biased towards the interests of rich people?
I don't think it does, but this an empirical matter so discussing it has little use and we'll find out once someone does a study.
That's the thing, I don't think there is another method, I think free will mucks it up. Human beings are what the philosophers call 'frustrators' and in fact there are even non-human entities that are frustrators. Let's say you have to bet on whether a machine will turn on or off, however the machine is a 'frustrator', once you bet that it will turn off it will turn on and vise versa. You aren't predicting the outcome (you literally can't) you are causing the opposite outcome. Humans are the same, if I see that someone predicts I will do X, I might do Y just to assert my autonomy (or maybe not because of mindgames).
So let's say a new prediction market opens on the machine frustrator, no one wants to be the first to make a bet, since you're guaranteed to lose money if no one else participates. Even if people are irrational enough to bet on it, we don't learn anything new and we just wasted a bunch of electricity and manhours for nothing. It's a waste of time and money to bet on frustrators, especially since you can get a 10% return on the stockmarket.
immediate edit: And since humans are frustrators and slowly gain more and more control over the world my question "Does this mean that prediction markets become less useful as humanity progresses, or is there a way to solve this?" still stands.
It will affect the trading and worse it will affect the trading inconsistently so we can't even use mathematics to subtract it. When the president promises not to manipulate the market about event X some people will trust him 90% and bet accordingly while others will trust him 10% and bet according to that. But that's not the same thing as people having 10% credence event X will happen, it can be that you think the president is trustworthy but the event itself is unlikely.
On top of that the president might react to the market, e.g if there is a lot of money to be gained he'd risk it but otherwise he wouldn't, but some people might think the president will manipulate the market at 51% when there is some money to be gained, while others think he will only act at 90% when there is a lot of money to be gained (or any other number of percentages), at seemingly random intervals people will suddenly buy or sell stock depending on what the price itself is doing. And the president might react to that etc, it becomes a recursive mindgame instead of being about event X. The resulting percentage cannot be used as a measure of trustworthiness nor as a measure of the underlying probability of event X.
This is always the go-to example defenders of prediction markets use, but it's not representative since it's the most extreme possible version of manipulation. No, I don't think most people will start murdering folk to influence prediction markets, not just because it could land you in jail, but also because it's difficult to execute and most people have an aversion to murder. But you can manipulate prediction markets in much more easy, mundane and legal ways. If I say I will do something and can subsequently win a bunch of money by not doing it, without anyone noticing, that's not even illegal, let alone difficult.
I'll use the example I used in my other comment:
Suffice to say, if you combine the fact that 1) humans can't instantly know all the new information with 2) the fact we can't know whether the market has updated because of information that we already know, new information, or people updating on the assumption that there is new information, with 3) recursive mindgames and 4) these constant 'ripples' of shifting uncertainties; you'll get so much asynchronous noise that the prediction market becomes an unreliable source.
Again, this is only for things we can change, prediction markets will work on things we can't change, like predicting supernovas (until the day we can cause them)
But is it in your individual interest? Is making enemies of your friends and family while losing a guaranteed lifelong stream of income for you and your children really worth it for the possibility of having a one time large payout?
What about with different probabilities? What if you think the policy has a 51% chance of helping the poor and a 49% chance of doing nothing. This would be a fantastic policy to try but even without coordination mechanisms like dominance assurance contracts, I doubt a self interested rich person would sacrifice their social network and lifelong stream of income for a 51% of having a one time large payout (actually less than 51% since it probably won't be implemented due to the conditional prediction market).