The biggest issue (aside from computational cost) is definitely how to reconcile conflicting models, although no one would ever be editing the entire model, only small parts of it. I hope (and I could be wrong) that once the system reaches a certain critical mass, predicting the emergent behaviour from the microscopic details becomes so hard that someone with a political agenda couldn’t easily come up with ways to manipulate the system just by making a few local changes (you can think of this as similar to a cryptographic hashing problem). Other large-scale systems (like cryptocurrencies) derive security from similar ‘strength in numbers’ principles.
One option is to limit input to the system to only peer-reviewed statistical studies. But this isn’t a perfect solution, for various reasons.
Using a connection to prediction markets (so that people have some skin in the game) is a nice idea, but I’m not sure how you’re thinking of implementing that?
Well, models generally rely on parameter values which can be determined empirically or reasoned about more theoretically or the model could be fitted to data and the parameters inferred by some form of optimisation algo such as monte carlo markov chains.
Anyway, suppose two people disagree on the value of a parameter. Running the model with different parameter values would produce different predictions, which they could then bet on.
The biggest issue (aside from computational cost) is definitely how to reconcile conflicting models, although no one would ever be editing the entire model, only small parts of it. I hope (and I could be wrong) that once the system reaches a certain critical mass, predicting the emergent behaviour from the microscopic details becomes so hard that someone with a political agenda couldn’t easily come up with ways to manipulate the system just by making a few local changes (you can think of this as similar to a cryptographic hashing problem). Other large-scale systems (like cryptocurrencies) derive security from similar ‘strength in numbers’ principles.
One option is to limit input to the system to only peer-reviewed statistical studies. But this isn’t a perfect solution, for various reasons.
Using a connection to prediction markets (so that people have some skin in the game) is a nice idea, but I’m not sure how you’re thinking of implementing that?
Well, models generally rely on parameter values which can be determined empirically or reasoned about more theoretically or the model could be fitted to data and the parameters inferred by some form of optimisation algo such as monte carlo markov chains.
Anyway, suppose two people disagree on the value of a parameter. Running the model with different parameter values would produce different predictions, which they could then bet on.