Cool read, how do you convert something like logloss into something understandable by a ML contest noob? For instance it'd be cool to know the % of the time one of these algos predicted correctly, how do you get that from the logloss?
Smart contracts can use oracle systems to get consensus on info from the outside world. So a smart contract normally can't communicate with anything besides its internal state, but with Augur they can get info about outside events.
One example of how this could be used (and our primary focus) is on prediction markets using the oracles to report on the outcomes of events after they occur. (Imagine a decentralized InTrade)
1). Good question. Paul Sztorc -- the inventor of the Truthcoin concept, on which Augur is based -- has been helping us a lot with game theoretic calculations. The Reputation concept's goal is to create a Schelling point on the consensus. You can construct a thought experiment, where you look at how the Reputation payoffs align with a perfect coordination game (with Schelling points). The result was that there is around a 90% chance that the Reputation mechanism creates a Schelling point. I don't have the details of this calculation, but I'm sure Paul and/or Jack (our project lead) could provide them if you're interested!
As for emptying the funds, we have the B parameter in LMSR vary according to volume. However, we're currently talking with an expert in automated market making (PhD @ CMU in CS w/ thesis in market makers) and he's recommending that we do not use LMSR (because new, advanced, market making algorithms have came out that don't allow issues like the one you mentioned come up).
2). I'm not exactly sure what you're asking here. We currently plan to encrypt votes using time lapse cryptography and then decrypt them via distributed key sharing.
3). SVD 0 and 1 work fine for binary outcomes, however when you include scalar predictions you have to fit the scalar values between 0 and 1 (because that's what svd operates on), hence floating point decimals. We're currently writing and implementation using fixed point, however, and that's probably what we'll end up using.
4). Yes we have, that was the simplest way to do it in a bitcoin transaction, however...
5) Absolutely, we're currently looking at Serpent contracts as a way to implement this (it looks like the biggest hurdle will be keeping SVD costs low, but so far it's progressing nicely)!
Perhaps you remember InTrade? Or there's also a few UK websites like that. The problem is they're centralized, so they can go out of business or a person who judges / shows the outcomes of events could lie to get a payout.
As for capitalism, I would argue that greed is one of the most common elements. We rely on people being greedy and, generally, willing to do the least work for the most possible results - aka laziness. It's trustless in the sense that bitcoin is trustless, we're building a platform that unlike the traditional financial system, doesn't require you to trust others.
The way the trustless aspect is used is that people are incentivized to vote on the outcomes of events with the consensus, because if you don't, you will lose reputation in the system. This is explained in further detail in the game-theoretical explanation of this project, available here: https://github.com/psztorc/Truthcoin/blob/master/docs/Truthc...