The mission
In recent conversations with delegates, we were surprised to find that there seems to be some confusion about the overall aim of Event Horizon. The original proposal was clear: Voter participation is low. To that end, Event Horizon has “the aim of enfranchising the tens of thousands of small voters” via a delegation from the DAO to a public access voter pool over the course of a one-year experiment. This was to be done by “giving underrepresented DAO citizens multiplied voting power and therefore more incentive to express their voices.” In other words, the mission is, and has always been, to increase the number of voters in Arbitrum DAO. Half way into this one-year experiment, how have we done?
Evaluation and metrics
Most delegates found our initial arguments for how we could attract more voters with vote multiples via Implicit Delegation compelling, but it’s important to be empirical with these experiments. How has Event Horizon fared so far? To answer this question we need to first understand what important markers of success might look like. Given our mandate has always been to increase voters, a natural metric to look at is, well, how many voters we’ve attracted. Let’s look at the numbers.
The original proposal highlights, “voting when you only have, say, $500 worth of ARB doesn’t make sense, it’s a drop in the bucket. Sitting out of governance proposals is, unfortunately, the rational decision but collectively makes everyone worse off. The strength of the DAO is directly related to the number of participants having their voices heard by the community. The governance platform and vote multiple that Event Horizon gives to these citizens incentivizes greater participation and surfaces greater cognition from the community.”
Today, without EH, of the >3,000 total voters, most only mobilize dust amounts. In fact, there are only ~280 voters who vote with 1000 ARB (~$500). Each Event Horizon voter represents ~43,000 ARB. This places each EH voter at a tied spot for ~65th largest voter in any given vote and increases the number of voters mobilizing >1,000 ARB from 280 to ~430, or a ~57% increase in meaningful votes.
To the second portion, turning this top of funnel, meaningful voting, into meaningful forum and discussion participation, we have already begun implementing communication features detailed below.
What about absolute voters? For the first 6 months, our average voters per proposal floated at around 10-15 voters. When we were estimating a few thousand in the proposal it’s clear: not good. But we’ve been around for slightly longer than 6 months. What does the full data look like? Let’s take a look:
Very interesting. What happened in the last couple of weeks? In a word: AI.
AI Agents
After hopping on several dozen calls with users it became clear to us that while voting power is a real incentive to show up, it’s not enough to justify the slow crawl through molasses that is DAO governance. You have to read the forums, through incessant arguments in a hostile, low cost-to-critique environment. You have to read proposals, talk with delegates, attend community calls, vote on Snapshot, vote on Tally, and provide a rationale for both votes. It’s a lot. We learned that it’s not that people don’t care; they simply don’t have the time and energy to labor for free. And fair play. A model voter is a delegate, and delegates, which includes Event Horizon users, are best thought of as senators, not as citizen voters.
In effect, through discourse with our users, we found that there are two sides to the equation when it comes to voter participation. The first being incentive. This was the only side our initial structure addressed. We incentivized participation with vote amplification. However, the other side, which was neglected (and ironically made more evident) through Event Horizon, was the perceived difficulty. The perceived difficulty of governance must not exceed the perceived value. And, without making governance easier, the high difficulty meant we could only attract those who found particularly high value in governance. The governance nerds as we lovingly put it.
This is where AI agents come in. What if you could spin up an agent which voted based on your preferences and values with a high degree of accuracy? What if all it took was answering a few questions in under 2 minutes?
There’s nothing special about manually clicking a button to vote. Would you rather have 10 humans voting manually? Or 10,000 agents voting with 90% accuracy to their underlying human’s preferences? For us, the answer is clear: there is no meaningful distinction to be drawn by who presses a button. What ultimately matters is that a user’s volition is expressed.
Current AI agents aren’t perfect. But we also have to be humble: most likely, these AI agents are already better than you at programming, research, making art, making music, giving therapy, debate, mathematics, and more. And if they’re not now, just wait a week. A new state-of-the-art model will be waiting for you.
But it’s also true that these models have their drawbacks. They can hallucinate and their context windows are limited. But skate to where the puck is going, not to where it is. AI governance is here to stay, and we’re happy to build with the DAO including all the necessary precautions and oversight (like the MSS veto vote) required to do this right. If we don’t build this, someone else will. And chances are they won’t care to work with the delegates as we do.
Sybil
How can we be sure that each agent corresponds to a single human? We can’t with absolute certainty, but we can take some precautions to minimize sybil attacks. After a recent sybil incident due to a too low Gitcoin pass requirement, we’ve tightened the screws. Changes include:
- IP address monitoring preventing too many agents being spun up from a single location
- VPN blocking
- Winding down manual NFT voting in favor of voting manually through chat
- The ability to shut down suspicious bot clusters
- Further, the very nature of the agent swarm affords build in Sybil resistance. As good actors setup agents, those agents continue to vote in perpetuity. To date we have over 150 doing just that. As that number grows the count to capture rises for any potentially malicious actors. As we rapidly approach north of 1,000 independent, human-rooted agents, a mal-intended participant would need to create hundreds of agents while trying to avoid IP blocks and traceable similarity between the agents and conversations the actor creates.
No sybil solution will be perfect, but this is a step in the right direction.
New features (launched in the last 2 weeks)
As outlined above, the number of voters we have has skyrocketed in only 2 weeks after switching to agentic voting. This validates our thesis and the conversations we’ve had with users that the final piece of the puzzle to increase participation is friction. Agents automate away this last mile problem.
Given the step function in users we saw after switching to agentic voting, we’ve been hard at work shipping as many features as possible to make these agents as smart, useful, and seamless as possible.
Voting rationales
Agents are no longer opaque black boxes who vote seemingly at random. All agents now provide thoughtful rationales for why they voted the way they did. These are not yet public, as we’re working on building the UI, but users can access their own agent’s rationale via the history tab at any point. Automatic forum posting of these rationales is also a feature we will be shipping soon, but we have begun sharing consolidated reasoning manually in our delegate communication channel.
Agentic vote preview
Users can see how their personal agent will vote, giving them the ability to intercede at any point.
External Context Inclusion – Agentic forum discussions monitoring
Agents no longer vote with just the user’s preferences + the contents of the proposal. As we all know, much of the governance process is done on the forums. Each agent now has this context just as any of us do. Decisions are made with the full back and forth taken into account, not just the “salesy” language put in a proposal.
Forum comment writing
Agents can now use full context provided from forum discussions and the users’ preferences to write thoughtful commentary on live and active proposal discussions. A common point of feedback we heard from delegates was that thus far Event Horizon hasn’t contributed to the direction of the pre-voting governance process. While our initial mandate was increasing voting, we took this feedback very seriously. These agents, in their current form (not some future state) are well poised to help with the DAO decision making process today by providing thoughtful contributions. We are not yet automatically posting these comments, as we want to find the right way to add agent rationales to discussions, while avoiding content overload. But, this is the first step in the direction of adding constructive contributions to the pre-voting process.
Proposal suggestions
Our agents have the ability to go beyond crafting forum responses that highlight their user’s preferences. Our agents can also suggest changes to active proposals.
An inflection point
The switch to agentic voting has proven to be a step function, not only in users (10->150 in 2 weeks), but also in functionality and utility.
We’re excited to keep building in this direction to not only continue to bring voters to the Arbitrum DAO, but to go use this new direction to also
Further risk reduction
We’ve heard concerns about Event Horizon’s delegation size and ability to influence votes. Given a recent sybil episode and the imprecise nature of sybil resistance, we suggest the following change: if a vote is contentious, and EH is likely to be a deciding vote, Event Horizon will force vote Abstain.
This change, paired with the already in place MSS veto ability, means that the Event Horizon DAO poses effectively zero risk to the DAO in any way at all. We’ll also remind everyone that Event Horizon does not hold any DAO assets. The MSS holds all tokens delegates to Event Horizon. Absent Event Horizon’s ability to influence any contentious votes, we functionally serve as an innocuous experiment in agentic governance building features to help make DAO governance easier to participate in. In addition to serving our initial mandate of bring more voters to the DAO, the recent agentic features listed above mean that we are on the verge of bringing thoughtful feedback and suggestions to actively live proposals, prior to any voting being held.
Final notes and further work
We would like to reemphasize that despite the change in form factor, the mission has not changed. For many, this change in form factor toward AI solutions is already seen as an exciting advancement toward the future of governance and governance inclusion. However, for those who may hold reservations, as we all learn to embrace the changes AI is driving, it is important to recognize that in all likelihood, this is an inevitability be it through Event Horizon or another similar structure. And, this is genuinely your opportunity to collaborate with a team in sculpting a solution you are comfortable with or can find excitement toward. The alternative being another initiative which you may have no voice in molding. We therefore invite feedback and suggestions for how to better improve the product over the span of the remaining 6 months of this experiment.
Finally, some exciting new features are being looked at internally including:
- Feeding in social media feeds and Arbitrum governance forum account tracking to further customize an agent’s persona and style of thinking
- Building a “delegate mode” whereby delegates can charge the agent with voting with their personal delegations, comment on forums, update a delegates delegation thread with rationales, and more. This is experimental and the feasibility is being investigated internally