Really super panel that was well moderated by raam! What i loved were not just the thought provoking questions we had time to prepare for in advance but also that each panelist could choose whether to answer or not Here are 5 questions and my answers to them for this panel on "AI governed DAOs" to help those that didn't make it in person 1. When we talk about AI in governance, what do we mean? There are 2 versions of this; a. AI as an assistant providing context on proposals, drafting help etc. Quite common in DAOs and web 2 organizations b. AI as a decision maker, much rarer. The only 2 cases coming to mind are deep fundings allocation to repos based on the winning AI and only dusts distribution of $1 million through an AI trained on their data Governance itself can be broken into approving product changes (Y/N) and who to give how much money. Humans mostly make better decisions so AI can only do better atm when there are too many decisions for executives to make 2. What are the promising and concerning aspects of AI agents as delegates? The biggest issue i see here is that we put the cart before the horse, with delegates simply assuming that their past comments will be taken as training data Whereas we should instead have AI predict how a delegate would respond to a proposal, let the delegate actually respond, and then store in its long term vector embedding databases the difference between its prediction and the actual delegates action. But i dont know anyone doing this at the moment 3. How can AI improve the quality of governance? Humans are generally bad at quantification, so ideally AI comes up with the first draft of fund division that we can then deliberate upon 4. DAOs need transparency so how do we reconcile with black box AI systems? If the AI runs on the same server and doesn't have a random seed, it's response can actually be the same every time its asked. But as tanisha mentioned the non determinism of AI responses is actually a good thing because it's much harder to game 5. What should be the governance mechanisms of the AI itself? Rather than trying to govern a single AI whose diktat we accept, we should have clear rules in place deciding which of the many AIs gets chosen to make a particular decision Finally, what role do we see AIs play in DAOs 5 years from now? In the optimistic scenario, it helps networks connect their revenue and cost centers more seamlessly so we can finally outcompete command & control organizations. Googles AI division doesn't ask for grants from its ad division, so why should vyper have to do that to the revenue generating protocols built on top of it?