Artificial intelligence assistant for a more efficient continuous tokenomics management

I am applying for a $5k Developer Grant.

Please review my application and discuss! See DFINITY’s directions for becoming a registered reviewer here. They will be collected by DFINITY. When one week passes, DFINITY will release them and they will appear as a new section on this post.

Please review my application and discuss! If you would like to submit an official review, then please use the links below to register, see the rubric, and submit a review.

I’m looking forward to everyone’s input!

Reviewer Registration | Rubric for Evaluating | Submit a Review



1 Like

How would people access the data that this AI generates? It doesn’t sound like its going to be on chain AI, or is it?

With a model like this, how reliable can it be? It doesn’t seem like its making the decisions, but rather informing the decisions of the DAO members, but i think if people were to use this, they would need to be confident its not going to be unreliable

1 Like

Thank you for the question/comment.

  1. Once the planner and inestor agents are trained, those are just neural networks that can be uploaded as WASM models. So simulations could run on chain. However, by the time being the training must be done off chain.
  2. About the reliability, I claim that tokenomics updates generated with this system could be more reliable than those made from pure intuiition or based on experience, and no less reliable than those tokenomics updates leaned by running multi-agent simulations where agents behavior has been learned from historical data. In this second case, I claim in my proposal that modelling agents behavior according to historical data is very limiting and an architecture as proposed based on 2 levels RL agents would offer the possibility of taking into account many cases that never occured in real life (historical data)
  3. Concerning confidence in predictions, let me elaborate a bit more: Once both planner and investor agents are trained, the users of the simulated system (responsible person inside the DAO to use the system) would set initial conditions corresponding to the current market state and let planner and investor agents to run. During the runs, the planner agent would try to optimize the tokenomics updates to maximize the utility function of the DAO under different possible evolutions of the market. Once the simulations end, the users of the simulated system (responsible person inside the DAO to use the system) would analyze the most successful tokenomics updates suggested by the planner. So, the planner suggestions would be backed up by full simulations traces consisting of time series of market conditions, investor behavior and tokenomics updates by the planner. confidence metrics should be extracted from these traces.
1 Like