Welcome to DeGroot, USA a small town with a decision to make: convert a plot of land into an outdoor playground or sell it to a private developer. The mayor dispatches her staff to hold community meetings, create websites and write articles to collect and present pros and cons for each option. DeGrootians are invited to share their ideas and opinions. Discussion points spread in chance conversations, in-person and online, and between neighbors, family, coworkers and friends. The conversations stay mostly contained in the social circles they start, and in some cases, cross them.
Following Morris DeGroot’s consensus model, during each conversation a person gives weights to other people’s opinions based on how much they know and trust them: Jamie weighs her neighbor Aaron’s opinion 0.3 and Jan her coworker 0.2. Everyone else is a 0. Jamie’s opinion changes based on a weighted average of the opinion of folks she’s talked to. Everyone doing this over and over starts to point to an average, or a consensus. DeGroot is large enough to dilute bad reasoning and trusted folks are amplified across conversations. How nice.
Now let’s imagine that instead of these conversations, people talk to an AI, and the AI gathers and synthesizes all the askers’ positions. It determines the consensus based on collected responses and is open about the consensus to anyone that asks.
Even if the consensus ends up the same in both scenarios, and the AI is transparent about how it grouped, weighed and compared the opinions, what’s lost? The weights DeGrootians give each other encode a relationship, connection and history. When AI replaces this, it’s doing something else; it weighs the task, and the one who asks, centrally.
My solar plexus tightens and stomach knots at the thought of DeGrootians getting to a consensus, even the right answer, through a centralized system. The friction in imperfect but decentralized human knowing is freedom. Wrongness is human and debate, disagreement and the friction of imperfect information are the humanist and democratic way to learn and decide without end.
Wisława Szymborska, the Polish poet, in her Nobel lecture (1996) warns of closed answers as the purview of “all sorts of torturers, dictators, fanatics, and demagogues…They know, and whatever they know is enough for them once and for all. They don’t want to find out about anything else, since that might diminish their arguments’ force.” The seducers who know what a woman is for, the revolutionaries who know what friendship is for, and the doctors that decide when nothing can be done. They all have the answers. Their closed answers come with ease often out of a friendly voice with intelligent, clarifying efficiency. I use AI to improve all three in myself all the time because it’s rewarded.
I used Claude to help me research and review what I’ve written here. Claude put a Daron Acemoglu academic paper in plain language, where the DeGroot consensus is mentioned. Claude helped me find articles by Derek Thompson for inspiration on how to write about complex things simply. Claude recommended Chekhov and Turgenev stories to read on the theme of friendly-faced tyrants. Claude reviewed this piece, offered edits, and told me what I wanted to hear: “I notice I’m treating the AI critique sympathetically because I largely share it, which means I’m probably being a softer reader on those passages than I would be if the essay argued the opposite. Worth knowing.”
Wait, who is worth knowing? I think about Jan and her daughter at an imagined playground in an imagined town. Tired after a long day, watching the kids bump into each other, glad Jamie told her about the new park.
Published by: reitmane in ✍︎