create-more-parties

Lightcone Commons Application (August 2026 round) · Submitted by Carol Calin (carol@think2earn.com), Prague / Romania · Deadline: August 23, 2026 · Recommendations ~October 23, 2026

TL;DR

The best defense against transformative AI is coordinated nation-states. The unit of coordination ordinary people can actually build themselves is the political party. This project pays citizens to found new, legally registered parties, not to join existing ones. Identity is proven with zero-knowledge passport proofs (one reward per human, no database to breach). Rewards are staged so payouts follow real parties with traction, rather than correctly filled paperwork. Romania is the pilot: since 2015 a party there needs only 3 founding members, and a new party law is in public consultation until September 20, 2026. Education is embedded in the process - one learns law and politics by using them, verified through short, content-neutral assessments. In 2024, the Constitutional Court of Romania annulled a presidential election over undeclared foreign money in politics. This program is built to be the opposite pattern by using a public ledger, without supporting specific candidates or campaigns, and with no personal data to seize.

What are you working on?

My high level belief is that the best way to defend yourself against the upcoming superintelligence is to properly coordinate entire nation states of individuals: both to make changes in their own nation, and to make changes at a global one. Coordination at that scale has a common name many have a negative bias for - politics. And the unit of politics that ordinary people can actually build themselves is the political party. Schattschneider said it in 1942: "modern democracy is unthinkable save in terms of political parties."

I want to be honest about the diagnosis underneath this project. It is tempting to say decision-making is slow because the people currently in politics are slow or corrupt, and want to enrich themselves and their future generations. This may be partly true, but the deeper truth is structural: institutions are slow by design — veto points, committee stages, consultation periods are anti-error features, and swapping personnel does not remove them. The same design casts a long shadow — Weber called it the "iron cage": rules accumulating until means become ends, and we find ourselves working for the systems that were meant to work for us. Cheap entry into organized politics is the standing antidote — it is the same discipline Baumol described, applied to institutions: nothing calcifies completely while founding a replacement stays affordable.

What new people and new parties do change is what the institutional machine is pointed at. Party systems have high barriers to issue entry; historically, neglected issues get representation when new parties force their way in. The green parties of the 1980s put climate on agendas decades before incumbents would have; the German Piratenpartei's 2011 surge turned ACTA from a quiet treaty into a defeated one; the Czech Pirate Party — founded 2009, in government by 2021 — made e-government a governing agenda here in Prague. Political scientists call these "movement parties" (Kitschelt 2006): outsiders organized around an issue the system refuses to process. AI governance is currently the largest unprocessed issue in Western politics, and as far as I know there is no movement party for it anywhere. This is the gap that I want to directly attack.

Why new parties — and why that is a feature

Three design decisions define this project, and they interlock:

1. Registration comes first, education second. The standard playbook — inform citizens, hope they organize — fails predictably: Kahan's motivated-reasoning work shows information campaigns bounce off identity, and field experiments (Gerber & Green; more recently Broockman & Kalla's deep-canvassing experiments) consistently find that contact and action beat information for political activation. So we invert it. The act of founding a party is the curriculum: drafting statutes teaches you law, the constitutive act teaches you procedure, negotiating a program teaches you coalition politics. You learn by doing the thing, with short practical assessments embedded at each step to verify process competence — content-neutral, no political litmus tests. Activation first; understanding follows commitment, not the other way around.

2. Rewards attach only to new parties, not to membership in existing ones. Romanian law 14/2003 bars simultaneous membership in multiple parties — each citizen has exactly one political affiliation to allocate. Existing parties receive nothing under this program. That makes exclusive political membership the scarce resource, and the program prices it decisively toward new entrants: joining a new party is rewarded, staying inside an incumbent is not. This is deliberate decentralization pressure. Madison's answer to faction (Federalist No. 10) was plurality: the more distributed power is across many organized groups, the harder it is for any single faction — including, eventually, an AI-concentrated elite — to capture the whole. Many small parties are redundant, anti-capturable infrastructure.

3. Proliferation now, re-coalition later. Phase one maximizes entry: many new parties, diverse programs, low coordination. Phase two — after the ecosystem exists — supports deliberate merging and coalition-building among the new parties, since Romania's 5% electoral threshold for Parliament forces consolidation anyway. And we make that claim testable in advance. Political science has a standard way to summarize a party system with one number — the "effective number of parties" (Laakso & Taagepera 1979) — where ten tiny parties count as less than two big ones. Because every cohort member is zk-verified, we can compute it for our own cohort at zero cost. Our pre-registered expectation: the cohort's number rises as new parties enter, then falls as the threshold forces mergers. If it only rises, we are manufacturing fragmentation, not renewal — and the post-mortem will say so. This is exploration followed by exploitation, and it mirrors how healthy party systems renew themselves. Economists have a name for the discipline effect: contestable markets (Baumol 1982). Even if most new parties stay small, the threat of well-funded new entry changes incumbent behavior on the issues newcomers carry. We do not need our parties to win majorities to move AI governance up the agenda; we need them to exist credibly, to educate themselves, and to push for change.

The loop

Why this reduces catastrophic risk

Insiders at frontier labs describe possible recursive self-improvement on a 1–3 year horizon; Jeffrey Ladish told ABC's Four Corners that governments effectively have "no plan," and that political salience is rising on both left and right. Scenarios like AI 2027 (Kokotajlo et al.) sketch transitions measured in years, not decades. Meanwhile the persuasion curve is already bending: an EPFL study found GPT-4, allowed to personalize arguments, shifted debate opponents' positions with roughly 80% higher odds than a matched human debater (Salvi et al., Nature Human Behaviour 2025). Awareness of black swan events without a coordination infrastructure decays into fatalism. Politicians respond to institutions, not petitions — and there are currently no citizen-founded movement parties anywhere focused on how transformative AI is governed.

The supply side of AI politics already exists, and this ecosystem already funds it: the AI Policy Institute (SFF-2025, $1.6M) polls and organizes public opinion on AI policy; GovAI and CLTR produce the governance analysis that becomes law. What nobody funds is the demand side — voters organized into durable institutions that make AI governance politically unavoidable. That is the lane this project occupies, and it is complementary by construction.

This maps onto Jaan Tallinn's stated constructive priority of Collective Intelligence — humanity becoming collectively smarter as a species — and on his observation that more people need to "wake up" and develop informed opinions about AI risk, since everyone under ~60 is personally at stake. It is also, per his philanthropy page, a software project: concrete, shippable, legible milestones. And it fits the bigger arc I care about: we will not beat death, spread to the stars, or steer superintelligence by waiting for elites. Mass movements with the right incentives are the mechanism, and nobody has built the incentive layer yet; widening education and participation is how this project tries.

What a funded party does with power

A fair question: what does a registered party at 3–5% actually do for AI governance? More than its size suggests. A party that never wins a seat can table council motions, file parliamentary questions, make AI governance a named demand in coalition negotiations, and force the issue into electoral debate where incumbents must take a position. We give every cohort party a concrete year-one menu, drawn from the governance agenda this funder community has already published: datacenter safety certifications for high-risk training runs; compute speed limits; liability laws making developers and users accountable for harms, including near-misses; a legal right to know whether one is interacting with a machine; and representative veto committees for large-scale AI deployments. A party does not need to originate this agenda — it needs to force a vote on it. That is a measurable output, and we pre-register it: at least one funded party files a formal position on national AI policy within 90 days of registration.

Restriction is only half the agenda. The constructive counterpart — what these parties are for — is citizens getting a governed share of the intelligence economy: mandated compute/inference dividends funded by frontier developers, and personal agency rights over AI agents acting in your name — the right to know when you are dealing with a machine, to revoke an agent's authority, and standing to sue when it acts against your interests. A movement party needs an affirmative vision to survive contact with voters; "the AI economy pays a public dividend and answers to the people it acts for" is one. This is the collective-intelligence future in policy form: not humans replaced by machines, but every human holding a governed stake in them.

What would you do with funding?

Pilot: Romania

Romania is the right first jurisdiction, for reasons that are legal fact, not preference:

Budget outline

TierAmountWhat changes
Minimum $60k / 6 mo MVP shipped: registration-journey tooling (RO/EN), zkPassport integration for RO documents, embedded assessment engine, staged bounty pool for a 200–500 person pilot cohort, and a formal legal opinion from a Romanian elections/party-finance lawyer (~$5k) validating the reward design — including the new-parties-only restriction.
Target $150k / 12 mo Line items below: full tooling production, part-time ops, bounty pool sized for 1,000 verified members, independent payout-integrity audit, public post-mortem.
Scale $400–600k / 18 mo Generalize the registration playbook to 2–3 jurisdictions selected by measured registration cost, open-source the full stack, run the first cross-country comparison of "does funded civic activation produce durable parties?"

Note on the durability tranche: it pays at roughly month 24 (one year after a party's registration), which falls outside the 12-month Target budget. It is ring-fenced from the $150k above — reserved, not left to future fundraising — and if a follow-on round replaces it, the reserve converts to additional cohort bounties.

Target-tier breakdown:

ItemAmount
Engineering & product (founder stipend + contract help)$60k
Legal: Romanian elections & party-finance opinion + GDPR review$15k
Tooling & assessment production (video, translations, item bank)$25k
Participant bounties (staged: registration / viability / durability — durability ring-fenced per note above)$35k
Independent integrity audit + pre-registered evaluation (includes matched comparison arm)$10k
Ops & contingency$5k

Current funding: none. Pre-seed; I have been working unpaid; I have already built a first draft of the zkPassport verification layer. No matching pledges outstanding.

Two notes on the bounty line. First, payouts are made in RON/EUR through the fiscal sponsor, with the full ledger public; see Technical build for why. Second, the unit economics: at ~10 registered parties, $35k is roughly $3,500 per party across all three tranches. Our pre-registered efficiency metric is cost per durable party, and we will report it either way the experiment lands.

On bounty sizing: $35k covers ~1,000 assessed participants at modest per-person amounts plus pooled, staged bonuses per party. Payment-for-learning has a mixed record in schools — Fryer (2011) found mostly null effects across several US city programs — and three differences matter here: participants are self-selected adults taking a consequential civic action, payment follows demonstrated process mastery rather than attendance, and the largest rewards attach to collective outputs whose value compounds (a durable party), which changes the dynamics from test-taking to institution-building.

Milestones (falsifiable, reported publicly)

Evaluation design, pre-registered before launch: a matched comparison cohort receives the tooling without bounties (recruited through the same channels); primary outcomes are registration completion, viability at month 6, and durability at month 12, reported for both arms. This is how we answer "are you buying parties, or activating citizens?" with data instead of assertion. We also track the cohort's effective number of parties (Laakso & Taagepera) over time. Pre-registration means the post-mortem cannot be gamed in hindsight.

Stop rule: if fewer than 15 founding groups are active and fewer than 200 participants are zk-verified at ≤$40k spent by month 4, we pause acquisition, publish what we learned, and re-plan before spending the remaining budget.

Technical build

The MVP is boring on purpose, because I already run similar stacks: Linux servers, automated TLS provisioning, deployments on AWS with Lambda, Firecracker microVMs, and Fargate. The product is a web app, an assessment engine, the zkPassport SDK, and formalized payout contracts.

zkPassport does the heavy lifting of verification: the passport's NFC chip is read on-device, the signature is verified against the issuing country's certificate chain and there is zero-knowledge selective disclosure of exactly the attributes we ask for (nationality, uniqueness, age), plus a set-membership check against the Tribunal's public founder record. We never see a passport number, and we never learn who the individual is. There is no database to breach. Zero-knowledge proofs go back to Goldwasser, Micali and Rackoff in 1985; what is new is doing them on a phone against sovereign public key infrastructure, which is precisely the trust anchor a state-registration workflow deserves.

Payments - assuming a solid legal clearance - go out in RON/EUR through Lightcone's fiscal sponsor, with every payment on a public ledger. I considered token-based payouts for their audit trail and programmable conditions. However, in Romania's current political climate where a presidential election was annulled over opaque money in politics, I think fiat with a transparent ledger is the defensible choice, and it removes an entire category of headline risk.

Programmable conditions are enforced by milestone verification instead of token logic; tokens remain an option only if counsel explicitly clears them. I have shipped smart contracts in Rust on NEAR and Solidity on EVM, so the payout layer is work I have done before. Mechanism-design work on funding public goods (e.g., quadratic funding, Buterin–Hitzig–Weyl 2018) shows payouts can be engineered; ours is deliberately simpler: fixed, capped, staged milestone bounties.

Who is involved?

Carol Calin — founder and engineer (carolcalin.com · github.com/rainbowpuffpuff · chat.think2earn.com). I wrote this application myself, and the project sits on the intersection of my degrees and my last two years of work:

I am from Romania, based in Prague, and I work well alone and with other people. At the NFI I had to write so that people who are not engineers could still use the result. I want to help, to learn, and to teach when I know something useful, and I think this project will be the culmination of it. After my brother passed away in Romania from a rare heart genetic disease in 2015, and the ambulance that came had no defibrillator, I knew that corruption and inefficiency kills, and that if I don't do something meaningful about it, I just have second-hand blood on my hands.

I believe in a high degree of individual responsibility — history shows what passivity costs — I studied exactly that in my Law, Justice and Morality courses. Although I think "selling" with fear is unethical, I genuinely believe that the dangers of AI can be but are not necessarily species-level eradication, so we must act. It's a black swan event, and even if it's low odds (although many people attribute a high probability to "doom") we must prepare to the best of our ability. I think that those aware of it have an even higher degree of responsibility to do so.

We need to enact laws, and to compete with lobbying organizations whose interest is a monopoly over how superintelligence is governed, we need far more apolitical individuals inside politics. This is the incentive layer that this project builds. (See "What a funded party does with power" above for what those parties push once they exist.)

Recruiting/advisory targets

(in conversation, or to be engaged with funding in hand):

We are deliberately small. The money mostly converts into (a) engineering time, (b) legal certainty, (c) citizen-facing bounties.

Anything else evaluators should know?

Legality and optics — we designed for the hardest objection first.

The hardest objection is domestic, and here it is named: in December 2024, Romania's Constitutional Court annulled a presidential election over undeclared foreign money in politics (CCR Decision 32/2024). Any program that moves foreign funds toward citizens around politics must expect that reflex from regulators, press, and readers. This program is designed to be the opposite pattern on every axis that mattered in that case: foreign-sourced funds flow to a public program, never to candidates, campaigns, or parties. Every payment, milestone, and funded party is published to a public ledger. Rewards are content-neutral and never conditioned on voting behavior or platform content. Identity is zero-knowledge, so there is no dossier to subpoena or breach. And the program's first public act is participating in the Permanent Electoral Authority's (AEP) own consultation on the draft party law. We will brief the AEP before launch, and if counsel or the AEP sees risk, the program pauses — the kill switch is written into the charter.

Beyond that, paying people around politics invites two failure modes: vote-buying and astroturfing. Our constraints: rewards are paid for (1) completing the lawful, content-neutral civic act of registering a new party and (2) passing process-competence assessments along the way — never for voting behavior, never for platform content, never routed through the parties themselves. Payments go to individuals via the program entity, not to parties, keeping us outside party-financing regimes such as Romania's 334/2006 Law on restrictions to contributions to parties. On GDPR, the cleanest compliance story is architectural, zkPassport means we store no personal data at all, which is data minimization by design (Art. 5(1)(c)). We still assume a regulator could disagree; that is why the legal opinion is a gating early deliverable. If counsel says the registration bounty is unsafe, the fallback splits the design: open process-learning bounties for everyone + non-monetary recognition (verified badges, public registry, valueless soulbound tokens) for completed registrations. The closest precedent for public money flowing to individual political participation is Seattle's Democracy Vouchers (2017–); the distinction is that vouchers subsidize campaign support, while we subsidize institution creation — and we say so explicitly rather than letting a reader discover the analogy themselves.

On abuse, incumbents, and zombies — stated plainly.

Three attack surfaces, three answers:

Impact honesty.

We cannot guarantee that parties founded through the program prioritize AI safety. Cohort 1 is an experiment with pre-registered metrics; if funded civic activation does not produce durable organizations, we will say so publicly and the playbook gets adapted or stopped. Month 12 is the built-in decision point: funders receive the comparison-arm results, the cost-per-durable-party figure, and the payout-integrity audit — everything needed to scale it, pivot it to tooling-only distribution, or shut it down, on evidence rather than narrative.

Precedents in this ecosystem.

Looking at the public SFF grant record, funders here have already paid for the pieces: Meaning Alignment Institute ($145k SFF-2024, $149k SFF-2025) for democratic inputs to AI; Existential Risk Observatory ($44k, SFF-2022) for public x-risk communication; Plurality Institute ($100k, SFF-2025); Collective Intelligence Project ($300k SFF-2023, $500k Initiative Committee 2024); Center for Election Science across multiple rounds; the AI Policy Institute ($1.635M, SFF-2025) for the supply side of AI politics; even the All-Party Parliamentary Group for Future Generations ($171k, SFF-2021). Nobody has funded the incentive layer that turns individual citizens into durable political institutions. That is the gap create-more-parties fills, and why this is complementary rather than redundant with existing portfolio.

Urgency.

Three clocks are running: (1) the AEP draft-law consultation closes September 20, 2026 — engaging now means shaping the regime we will operate under; (2) the current 3-member registration window may close if the 100-member threshold passes, raising the cost of every future cohort; (3) lab insiders describe possible recursive self-improvement on a 1–3 year horizon, and machine persuasion is already measurably superhuman in debates. Democratic coordination infrastructure takes years to matter; the time to start it is while it can still influence the transition.

Logistics.

I split my time between Prague, where my partner studies, and Romania, where my family lives. I don't have a 501(c)(3); happy to receive via Lightcone's fiscal sponsorship or LLC contracting route. Previously created an LLC in Wyoming. I don't have restrictions on which funders we can accept from. Individual grants welcome; we can start with the minimum tier (and even less) immediately, and the legal opinion is the highest-leverage first spend — a direct grantor can insure that step alone. I genuinely believe that working on this is the number one best thing I can do with my life.

Thank you for reading. I would be happy to discuss — and if a direct grantor wants to insure the legal-opinion step specifically, that is the highest-leverage $10k in this application.

Reading that shaped this proposal

Sources