Are you still game to implement this? Would you be willing to add a clean API to integrate with Discord or other team chat tools?
Futarchy is an old idea from Robin Hanson. It may be unrealistic for governments any time soon but it can, with the right UI, work great for small groups. Daniel Reeves and Bethany Soule call it "prognootling". It's their process for resolving a disagreement that involves not conflicting preferences but conflicting predictions. (They use auctions for the case of conflicting preferences, but that's another story.)
Say we're deciding whether to submit a Manifund proposal (this is a true story!). We first agree on some numbers, in this case the following:
If we submit a proposal and we're accepted, we'll personally derive $2k of utility, net of our own efforts (and the funding)
If we're rejected, we'll have wasted $100 of effort putting the proposal together
If we don't submit a proposal, that's $0 of utility
How did we pick those numbers? Almost entirely proctogenically. We did independently pick 80% confidence intervals on the developer time required and averaged them and factored that in to the $2k number. But we had no disagreement about any of that. The disagreement was about the likelihood of actually being accepted. If you math this out, it's worth submitting this proposal as long as the probability of getting accepted is at least 4.8%. But what is that probability?
Daniel Reeves and his friend and potential partner in this project, Jake Coble, have different takes. Daniel's sense is that it's at least 50/50 and Jake's sense is it's at most a few percent. So we resolved that like so:
Conditional on choosing to submit the proposal, Daniel puts $1 on "we get accepted" to Jake's $1 on "we get rejected".
(Conditional on choosing not to submit, no bets are needed.)
The implied market probability for those preliminary bets is 50% which makes it a no-brainer to submit.
Jake bumps his bet up to $20 on rejection which pushes the market probability below 5% for an implied optional action of not submitting.
Danny bumps his bet to $2, for a market probability of 9% and an implied optimal action of submitting after all.
Jake is now happy with this. We agree to submit a proposal and Daniel will owe Jake $2 if we're rejected and Jake will owe Daniel $20 if we're accepted.
Clearly Jake didn't actually need much convincing in this case. You can imagine higher stakes decisions where one person wants to do something their partner thinks is too risky. In that case the wager can let the risk-averse person hedge the risk. They wouldn't want to risk it but they had a betting war and pushed the wagered amounts high enough that if the risk-averse person's fears come to pass, they'll be getting a nice payout from their partner. A reified told-you-so.
You can read a story like that here: https://doc.dreev.es/prognootle
And the proof of concept that we've used for these decisions is here: http://dreev.es/prognootle (world-editable; please don't actually edit it)
The objective of this project is to turn that into a tool that a small group can use to make decisions like this.
Daniel Reeves has a PhD in computer science (algorithmic game theory / AI) and has published academic papers on prediction markets. He is also a cofounder of Beeminder. He has built various tools for group decision-making such as http://bid.yootl.es and https://biddybuddy.dreev.repl.co .
$2500 will pay for the projected 35 hours of developer time. If our estimates are off, we'll eat the cost of the additional time needed to complete the task.
@Domenic: You're probably right that it's a long shot on this getting actual traction. I'm hoping that the right interface reduces the friction enough for it to be useful to a certain nerdy niche. The key is being able to agree on some made-up utility values and then to fight it out over a probability. I think that's not exactly a rare form of disagreement!
Also, I think the difficulty of assigning a probability is actually addressed pretty well by this tool. I agree that you don't have a strong feeling about the specific numerical probability but you do -- in the situations where this tool would be useful -- have a strong opinion about what action should be taken. If you've agreed on the utility values, then that strong opinion implies a threshold on the probability.
So the person you're disagreeing with feels strongly that the probability is on one side of that threshold and you feel strongly it's on the other side. Now you just let the stakes on the wager grow until one of you relents. You still think the other person is wrong but now they'll be paying you so much money for being wrong that you're ok with that!
It's gonna be great!
9 months ago
This vaguely seems cool in that I've always wondered, when making day-to-day predictions, whether the amount of money matters. That is, sure, I'll mentally predict a 70% chance that the cafeteria has chicken today, but how much would I bet at 70:30 odds? I get the sense that your proposal here would help me tease that out.
However, in practice, my eyes glaze over when reading the described scenarios. (Sorry! I know you tried really hard.) Putting a single probability number on things is already a hard enough skill to develop. Figuring out a potentially-escalating series of bets, based on wild-ass-guessed utility dollar values? Or, apparently as an alternative formulation, saying I'd "35% do something"? I haven't internalized this concept, and I doubt many other people in the community have either. I think it'd be difficult to get adoption for anything built here, without first percolating the idea through the blogosphere for a few years.
Q: Is this already possible with Manifold Markets?
A: Kind of, but if two people are having a tug of war over the price in a Manifold Market, any 3rd party can sit in the middle and extract money from them. So this works better as direct wagers. Maybe it can be done with Manifold's Challenge Bet feature? We'll definitely prefer to use Manifold for this tool if possible.