Editing Dumb contracts (section)

= Other problems with DAOs =

The bigger problem with the idea of investing in some kind of artificial, decentralized agent is the fact that a human being is virtually never going to be able to audit all of the code. Even a simple neural network is hard for a human to understand so if someone does manage to make some kind of business-like AI how is anyone going to be able to trust it?

You would need to put in some kind of VETO power or kill switch which is kind of what the Bitcoin alert key does. Furthermore, in a decentralized system everyone is going to need to see the exact code used for that agent … which may undermine any competitive advantage that might have been produced by running it.

Business is, after-all, heavily about timing as well as having some kind of edge over your competition. So how useful would it be to invest in an AI where those details all needed to be public? I have seen black box obfuscation / indistinguishability obfuscation thrown around as potential use-cases where the code behind an agent could be hidden while still being able to be run securely on untrusted systems, but this would again mean blindly trusting a black box when investing.

My conclusion on the notion of general-purpose AIs being used to control autonomous systems is that investing in them will end up being extremely risky if they do catch on so I think more specific and limited autonomous systems like my conception of [http://roberts.pm/datachain datachains] will likely fare much better.

<span id="where-were-at-today"></span>