I’ve just begun catching up on what Ethereum is doing now that 2.0 upon us. This prompted a short reflection on smart contracts, and how smart contracts put a huge premium on clear metrics to make them resolvable automatically.
This is prime Goodhart territory. It’s already a common strategy to hedge your bets by dividing investment up among less and more risky payoffs in order to stabilize your expected benefit; I cannot for the life of me think of any reason the exact same reasoning (maybe even quantitatively) cannot be applied to proxies for a goal, where risk is replaced with accuracy or some other form of goodness-of-proxy.
This would complicate smart contracts a bit, because it adds another layer to what you want to consider in the resolution mechanisms: before you needed to agree on the metric and then agree on a source to resolve the metric and now you need to agree on a suite of metrics with sources for each.
However, I feel like it weighs heavily in favor of the more modular approach to assembling software, so that people could publish open source contract resolution mechanisms which specify Metric X resolved by Source Y, and then these would be laying around ready to be added to any given smart contract you are designing. We would be able to rate the goodness of the metric-source combination independently of the contracts in which they are included; it would be easy to compare sources or metrics as a group; individuals or businesses could pre-commit to using libraries of specific mechanisms.
Then assembling a suite of metrics to hedge your Goodhart vulnerability is as simple as adding more of them to the contract, and you can get your smart contract resolution reasoned-rule-style by weighing the mechanisms equally.
If anyone knows where in the smart contract literature this is discussed, I would love to hear about it, because with the possible exception of saying Goodhart this can’t possibly be a new thought.
Could we hedge Goodhart?
I’ve just begun catching up on what Ethereum is doing now that 2.0 upon us. This prompted a short reflection on smart contracts, and how smart contracts put a huge premium on clear metrics to make them resolvable automatically.
This is prime Goodhart territory. It’s already a common strategy to hedge your bets by dividing investment up among less and more risky payoffs in order to stabilize your expected benefit; I cannot for the life of me think of any reason the exact same reasoning (maybe even quantitatively) cannot be applied to proxies for a goal, where risk is replaced with accuracy or some other form of goodness-of-proxy.
This would complicate smart contracts a bit, because it adds another layer to what you want to consider in the resolution mechanisms: before you needed to agree on the metric and then agree on a source to resolve the metric and now you need to agree on a suite of metrics with sources for each.
However, I feel like it weighs heavily in favor of the more modular approach to assembling software, so that people could publish open source contract resolution mechanisms which specify Metric X resolved by Source Y, and then these would be laying around ready to be added to any given smart contract you are designing. We would be able to rate the goodness of the metric-source combination independently of the contracts in which they are included; it would be easy to compare sources or metrics as a group; individuals or businesses could pre-commit to using libraries of specific mechanisms.
Then assembling a suite of metrics to hedge your Goodhart vulnerability is as simple as adding more of them to the contract, and you can get your smart contract resolution reasoned-rule-style by weighing the mechanisms equally.
If anyone knows where in the smart contract literature this is discussed, I would love to hear about it, because with the possible exception of saying Goodhart this can’t possibly be a new thought.