A true direct democracy is an idealized form of self-governance where individuals hold the power to directly vote on issues. This is mostly discarded as a viable form of governance, instead individuals vote for representatives. Each representative is supposed to represent a cohort of the voting blocks interests. This system works well enough, but could it be better? Can we move towards a more direct approach, without succumbing to the tyranny of the majority? Let’s throw off the shackles of the past and imagine a more fair, non-tyrannical, and most importantly “two-way” form of governance.

Direct AI Democracy or DAID, as a form of government where individuals are represented by their AI counterpart. The premise for this is:

AI is sufficiently intelligent to function as a viable proxy for a whole human person. Essentially, AGI.
The above AI agent, henceforth known as the proxy, must be cheap enough to scale for an entire voting population (intelligence too cheap to meter).
In order to prevent an arms race, all proxies must be on the same compute and model. The wealthiest individuals cannot buy a smarter agent that will subsequently argue for their case more convincingly.

As each voter is paired with their AI proxy, there is a learning period where the voter and proxy synchronize on core ideals, opinions, needs, and aspirations. Once this initial phase is over, the proxy can begin to serve its legislative functions.

As the second phase begins, voters are in communication with their proxy as ideological theory meets on-the-ground reality. Edge cases must be worked out, thoughts fleshed out. This is done in concert with the proxy, and at this point the influence is “two-way”. The education of the voter as well as the proxy, is key here. This phase is permanent, the proxy keeps synchronizing with the voter, and now the voter also synchronizes with the proxy. Opinions are changed through arguments. How much you engage with your proxy is up to you, but more engagement means better representation, similar to today’s representative democracies.

Your proxy will argue for your case on every issue, tirelessly. It will build coalitions in real-time, build bridges with ideologically familiar groups, practice intersectionality. Compromises, negotations, brinksmanship, conflicts. All of this, multiplied by billions, quietly happening in the background. If democracy could ever be manifest in its purest form, surely this is as good as it gets.

Why This Is a Terrible Idea

Now that I’ve hopefully riled you up, let’s break down why this would be a horrible, potentially species-ending idea.

Dangerous Centralization of Power

Seeing as compute and model is tightly controlled, someone will control the centralized AI system, and that someone is essentially in control of you. Representative democracy always dangles the threat of elimination in front of the representative, as they can be replaced by vote. This system cannot be voted out. DAID means that you forfeit your future, leaving whatever power you as a voter have, in the hands of technocrats (and they already have quite a bit).

Loss of Human Agency

By removing humans from their own governance, the natural endpoint of which is complete detachment and apathy from the process, we reduce our status to passive consumers. Human involvement, as messy as it is, is preferable to oblivion.

Attack Vectors

Democracy today has some large attack vectors, manufacturing consent is already far too easy, as we are susceptible to mis- and disinformation campaigns. AI systems are and will be vulnerable at a similar level, and the centralization makes the point of failure too large. Centralizing power is always dangerous. On the other hand we can decentralize, allowing voters to bring their own models and compute. As a consequence wealthy voters will simply buy up more compute time and argue for their opinions by convincing other “poorer” proxies, essentially buying votes. So we’re left with two shitty choices, both which lead to imbalance in their own way.

Legislative, Judicial, Executive?

Let’s say you’ve convinced your tiny network state to adopt DAID. The proxies have reached consensus on some set of laws, which are written into smart contracts. But what about those pesky human judges, legislating from the bench? A terrible choice must be made to replace them with proxies as well, and you know what, while you’re at it let’s get rid of those corrupt cops and get some nice aligned proxy-cops. Congratulations, you’ve built the leviathan guillotine with which you can decapitate the vestigial human agency. I know, this is a slippery slope argument, but some slopes are really slippery.

Conclusion

Sometimes technology brothers want to build solutions that are inherently problematic. I fear DAID would be quite robust, self-healing and anti-fragile. It wouldn’t fail, we would. Piece by piece, what is left of humanity would erode, in the name of efficiency. If we must build something, could we instead of replacing human political participation with AI proxies, focus on using AI to enhance human democratic participation. Whatever that looks like, I hope it differs from the black mirror future that DAID would almost certainly result in.