A genuinely elegant and precise framing of how AI can influence aggregation and formation as the two ends of civic life! I fully agree that we should focus more on the role of formation, not just aggregation, for guarding against Tocqueville’s tutelary state.
This post dovetails nicely, I think, with some of the recent discussion about the role of AI in mathematics. If the whole point of mathematics is proving theorems, then human mathematicians are probably obsolete. But if the important part is actually more like, "the things we learn while trying to prove theorems", as (I think) Terence Tao has suggested, then it's not so clear.
AI can do a lot of things, but it cannot learn our lessons for us.
Distinction between aggregation and formation, interesting! I like that. I would worry, though, that you are unconsciously accepting one of the core premises that has been used against lower case "d" democracy for centuries, that "modern" (the same things have been said again and again across the centuries) complexity necessarily requires governance by specialists operating at great scale.
because when your write that today's system is partly a function of "the highly complex nature of the modern world" and its dependence on "large-scale coordination and deep technical specialization." But almost every major expansion of democratic participation, from Athens, to medieval civic republics, to the American localist traditions Tocqueville admired, to the Jacksonian era, was opposed on just those grounds!
The recurring argument was always that society had become too large, too interconnected, too technically demanding, and therefore had to be managed by experts. Yet the great strength of democratic and federated systems was never that they eliminated expertise; it was that they accessed vastly more expertise, local knowledge, practical experience, and error-correction than centralized alternatives. The problem is not complexity itself, but where decision-making authority is located, how it is distributed, how its functions are designed, who has access to those functions, and by what means do they access them.
A highly federated system can be extraordinarily sophisticated while still distributing responsibility across thousands of local institutions, jurisdictions, associations, firms, schools, parties, civic bodies, etc.
The United States once had genuinely democratic governance structures, however imperfect and limited, fundamentally based around decentralized and publicly accessible mass-member parties. The Democratic Party, as a small "d" democratic institution, and the Republican Party, as a small "r" republican institution, were honest in their naming and functioned within a politically, economically, governmentally, financially, and scientifically decentralized and pluralized system that had legal and regulatory variability, policy variability, an intentionally diffused and pluralized private sector, and local fiscal dominance.
These parties, while far from flawless, allowed for real representation, genuinely participatory governance structures even for very serious policy matters by serving as an organizational interface and coordinating mesh for a wide array of organizational linked publicly accessible decision making bodies that spanned economy-schooling/training-culture-science-government, and a level of public accountability in political, economic, governmental, financial, and scientific decision making.
However, after WW2 a long multi decadal transformation began due to the dirty deeds of a convergence of several interests and an assortment of powerful special interest groups, and then our parties were transformed into centralized, exclusionary membership organizations. The so called Democratic Party has become a technocracy party, and the so called Republican Party became a conservative party. Neither really represents their original principles of democracy or republicanism, and they don't offer meaningful access or representation to the public. This transformation of the parties has been accompanied by a broader centralization of political, economic, and scientific decision making, which has caused the effective loss of most democratic governance structures.
Great article, glad someone wrote it up! For the record, we worked on a AI-enabled democratic process with OpenAI where we measured the effect of participants as a design criteria (and saw that people learned something new about what was important to them in the process, and gained respect for their fellows). https://arxiv.org/abs/2404.10636
The formation-versus-aggregation split explains why almost every AI-for-democracy tool drifts toward the same shape. Aggregation is measurable, so it gets built and funded; formation has no clean metric, so it quietly falls out of scope. The same bias shows up in ed-tech, where completion rates crowded out actual learning because one number was countable and the other wasn't. Civic AI looks set to inherit that blind spot unless formation gets a measure worth optimizing for.
No mention of how Switzerland's model of political participation fits into... or responds to... the problems discussed in this piece. Switzerland is small, yes... but has 3-4 quite distinctive cultural/language units linked to major European neighbors... the country still a model of possibly best democratic governance.
Switzerland is a goods example here but thats easy to do in such a small Land. It was so hard in the continent spanning and very ethnically diverse land of the USA.
So the for them, the truly hard fight wasnt just keeping things Fair, it was keeping lower case "d" democratic governance *real* across a vast continental-scale territory.
That is why Men like Jackson, Van Buren, Benton, the Locofocos, Polk, and many more thought of a federated architecture of states, municipalities, civic organizations, locally anchored financial institutions, mass-member parties, and other people-to-state bodies that distributed decision-making throughout society.
I loved this piece! It reminded me of the AGIL model, used to be very popular -- maybe time for a comeback. Says everything, politics included aggregates, sets goals and integrates, the does it again, and the "black box" between was like us not knowing how AI turns inputs to outputs... This piece does serious theory!
A genuinely elegant and precise framing of how AI can influence aggregation and formation as the two ends of civic life! I fully agree that we should focus more on the role of formation, not just aggregation, for guarding against Tocqueville’s tutelary state.
This post dovetails nicely, I think, with some of the recent discussion about the role of AI in mathematics. If the whole point of mathematics is proving theorems, then human mathematicians are probably obsolete. But if the important part is actually more like, "the things we learn while trying to prove theorems", as (I think) Terence Tao has suggested, then it's not so clear.
AI can do a lot of things, but it cannot learn our lessons for us.
Distinction between aggregation and formation, interesting! I like that. I would worry, though, that you are unconsciously accepting one of the core premises that has been used against lower case "d" democracy for centuries, that "modern" (the same things have been said again and again across the centuries) complexity necessarily requires governance by specialists operating at great scale.
because when your write that today's system is partly a function of "the highly complex nature of the modern world" and its dependence on "large-scale coordination and deep technical specialization." But almost every major expansion of democratic participation, from Athens, to medieval civic republics, to the American localist traditions Tocqueville admired, to the Jacksonian era, was opposed on just those grounds!
The recurring argument was always that society had become too large, too interconnected, too technically demanding, and therefore had to be managed by experts. Yet the great strength of democratic and federated systems was never that they eliminated expertise; it was that they accessed vastly more expertise, local knowledge, practical experience, and error-correction than centralized alternatives. The problem is not complexity itself, but where decision-making authority is located, how it is distributed, how its functions are designed, who has access to those functions, and by what means do they access them.
A highly federated system can be extraordinarily sophisticated while still distributing responsibility across thousands of local institutions, jurisdictions, associations, firms, schools, parties, civic bodies, etc.
The United States once had genuinely democratic governance structures, however imperfect and limited, fundamentally based around decentralized and publicly accessible mass-member parties. The Democratic Party, as a small "d" democratic institution, and the Republican Party, as a small "r" republican institution, were honest in their naming and functioned within a politically, economically, governmentally, financially, and scientifically decentralized and pluralized system that had legal and regulatory variability, policy variability, an intentionally diffused and pluralized private sector, and local fiscal dominance.
These parties, while far from flawless, allowed for real representation, genuinely participatory governance structures even for very serious policy matters by serving as an organizational interface and coordinating mesh for a wide array of organizational linked publicly accessible decision making bodies that spanned economy-schooling/training-culture-science-government, and a level of public accountability in political, economic, governmental, financial, and scientific decision making.
However, after WW2 a long multi decadal transformation began due to the dirty deeds of a convergence of several interests and an assortment of powerful special interest groups, and then our parties were transformed into centralized, exclusionary membership organizations. The so called Democratic Party has become a technocracy party, and the so called Republican Party became a conservative party. Neither really represents their original principles of democracy or republicanism, and they don't offer meaningful access or representation to the public. This transformation of the parties has been accompanied by a broader centralization of political, economic, and scientific decision making, which has caused the effective loss of most democratic governance structures.
Great article, glad someone wrote it up! For the record, we worked on a AI-enabled democratic process with OpenAI where we measured the effect of participants as a design criteria (and saw that people learned something new about what was important to them in the process, and gained respect for their fellows). https://arxiv.org/abs/2404.10636
The formation-versus-aggregation split explains why almost every AI-for-democracy tool drifts toward the same shape. Aggregation is measurable, so it gets built and funded; formation has no clean metric, so it quietly falls out of scope. The same bias shows up in ed-tech, where completion rates crowded out actual learning because one number was countable and the other wasn't. Civic AI looks set to inherit that blind spot unless formation gets a measure worth optimizing for.
No mention of how Switzerland's model of political participation fits into... or responds to... the problems discussed in this piece. Switzerland is small, yes... but has 3-4 quite distinctive cultural/language units linked to major European neighbors... the country still a model of possibly best democratic governance.
Switzerland is a goods example here but thats easy to do in such a small Land. It was so hard in the continent spanning and very ethnically diverse land of the USA.
So the for them, the truly hard fight wasnt just keeping things Fair, it was keeping lower case "d" democratic governance *real* across a vast continental-scale territory.
That is why Men like Jackson, Van Buren, Benton, the Locofocos, Polk, and many more thought of a federated architecture of states, municipalities, civic organizations, locally anchored financial institutions, mass-member parties, and other people-to-state bodies that distributed decision-making throughout society.
I loved this piece! It reminded me of the AGIL model, used to be very popular -- maybe time for a comeback. Says everything, politics included aggregates, sets goals and integrates, the does it again, and the "black box" between was like us not knowing how AI turns inputs to outputs... This piece does serious theory!