I’m an American, and in younger and brighter days I used to beat my chest about freedom, democracy, and the American Way. And today, though the veil has been lifted and the masks have been shed on the systemic insufficiency of the American system, I still believe in democracy. So I want to play with some ideas on how we could do it better.

I’m an engineer and I often have to estimate how much data I need to control the dynamics of a system. To control a complex system, you need to take periodic measurements and process that information to inform your next actions. Let’s apply this thinking to United States democracy.

Let’s say you only get to vote once every four years, for the president. In America, you only really have two choices, red team or blue team. This makes it pretty easy to represent: one vote is 1 bit of information. If everyone votes, that’s 350 million bits, or 43.75 MB per 4 years. That’s less than 1 bit per second. Now, not everyone votes and there are many more elections than just the presidential one, so this is a very rough estimate. Nevertheless, this order of magnitude should feel all wrong. Does anyone believe that the will of the people can be represented by a datastream of 1 bit per second?

It’s worth noting that there are other, faster “societal datastreams” than voting, principally “voting with their dollar”: choosing one product over another or choosing what stocks to buy. Nowadays we “vote with our attention” when we decide what media to watch. These alternative voting schemes are much higher throughput but ultimately impotent. It’s like George Carlin said, our “freedoms” are really just about “paper or plastic?” and “do you want fries with that?”

What if, instead of voting, everyone could write an essay? Let’s say 1000 words on how the government should operate. At a modest 5 characters per word and 8 bits per character, that’s 40 kB per citizen. If 350 million people do this every 4 years, that’s 14TB per election, a datastream of 111 kB per second. Now we’re getting somewhere.

Let’s let people submit essays whenever they want, and let’s take off the word count while we’re at it. Let’s say we manage to produce a datastream of 1 MB per second. That’s a datastream consisting of normative, moralizing “should” statements. Theoretically the will of the people is in there.

What are we supposed to do with all that data? Who is going to read all that? We’re meant to synthesize that into coherent thoughts? If only we had a technology that could process truck-loads of data and distill them into meaningful patterns.

You knew it was coming, right?

Artificial intelligence could perfect democracy. Let’s make a national LLM. On a regular schedule, say once a year, we train a new model on the will of the people (about 31 TB per year). It would be like a software update for the government. In an age of a rapidly changing technological landscape, our society needs to be able to adapt fast enough. The Constitution and the Declaration were written with feathers. It’s only prudent that we update our system to the available technologies.

Obviously it’s not easy at all. We’ll have to face the bleak reality of American illiteracy: 21% of adults are illiterate and half are below a 6th grade reading level. We’ll have to figure out how to actually integrate that AI into the day-to-day operations of the government. There’s still an issue of “voter fraud,” for example a bunch of submissions from a bad actor.

It’s a harebrained idea to be sure, but I find it exciting to imagine automating away the jobs of corrupt politicians and lobbyists.