UPDATE – Agent Interpretability and Inter-Agent Communications v1
1. Pre-Vote Forum Feedback and Context: We are building a forum scraper and agent communication process by which every newly created forum proposal is immediately passed to our agents. Each agent creates its perspective and rationale, emphasizing the greatest strengths, weaknesses, and what would change their opinion / flip their vote. The sum of these rationales is bucketed into higher-level opinion sets, which are then represented by their strongest points. The output is that each proposal author will have a clear understanding of how the agents immediately respond to their initial concept, along with a breakdown of strengths, weaknesses, and most importantly, improvements / what would sway the agents favorably. Below is the very first test case run of the processes.
2. Bi-Directional Communication: Expanding on the above, we will be adding a feature where forum users may @ / tag the agent body to ask questions and request feedback directly. This further removes the agent body from the previous black box model to a true voice which is engagable by any delegate directly within the forum.
3. Inter-Agent Communication: This is very early, so take with a grain of salt. We are looking to find novel outputs as well as to ‘weight’ conviction and validity beyond simple majority. To do this, we are exploring the establishment of debate forums between the agent personas. An interesting test case was seen with the recent Quorum reduction proposal: [Constitutional] AIP: Constitutional Quorum Threshold Reduction - #29 by EventHorizonDAO
Approximately 93% supported while only 7% were skeptical. Despite this, we consolidated the best arguments from both perspective camps and consolidated each into a single representative. We then had the representatives debate amongst each other effectively normalizing the 93 to 7 to a one-on-one with an effort of finding novel compromises and improvements. Each representative entered the discussion with a pre-existing conviction score. The intended objective was to reduce the conviction of the higher-convicted party through compromise and raise the conviction of the lower-convicted party through concession. The end result would then be a more broadly agreeable structure for all perspectives. Interestingly, through this discussion, which was grossly simplified and truncated as to be representable in the forum post, the agents landed on a recommendation to move forward with a quorum reduction, but to add barriers to future reductions. The reason being, the greatest concern of the skeptics was that this approach of lowering quorum to allow for vote passing would follow a slippery slope path as the path of greatest convenience for all quorum-related issues moving forward. The skeptics advocated for more durable, future-proof solutions focused on expanding votable supply. So, while the skeptics were willing to compromise with this initial reduction as to elevate the immediate risks, they wanted to see an increase in the procedural requirements for future reductions as to increase the difficulty, reduce the easy-solution positioning, and further encourage exploring longer term fixes.