0
Constitutional AI Infrastructure
Not fundedGrant
$0raised
The Wisdom Forcing Function (WFF) is a Constitutional AI system that enforces democratic principles in AI-assisted governance. In 36 experimental trials, standard LLMs generated non-viable governance proposals in 100% of cases; with WFF, compliance reached 100%. We are seeking $75,000 to open-source, validate, and deploy this work.
What are this project's goals? How will you achieve them?
Primary Goal
Release an open-source Constitutional AI framework that prevents AI systems from producing governance proposals that violate democratic principles.
The Problem
AI tools (ChatGPT, Claude, etc.) are rapidly entering municipal planning, budgeting, and resource-allocation workflows. These models optimise for efficiency but lack constitutional constraints.
In our experimental trials (N=36), unconstrained LLMs violated core democratic principles 100% of the time when asked to generate governance proposals.
Our Solution
WFF is a neurosymbolic constitutional enforcement layer that validates governance outputs from LLMs. It:
Accepts proposals from any AI model (e.g., “Allocate $1M to housing”).
Tests them against 11 constitutional principles (wholeness, distributed authority, temporal justice, reciprocity, transparency, and others).
Issues a binary verdict:
VIABLE – proposal respects constitutional constraints
NON-VIABLE – proposal violates one or more principles
Produces a human-readable diagnostic describing the exact violations.
Performs autonomous self-repair: when a proposal is non-viable, WFF can generate corrected alternatives (78% success rate across trials).
Experimental Validation
We compared outputs from:
Standard LLMs (0% compliance)
LLMs with WFF enforcement (100% compliance)
How we will achieve the goal
Months 1–3 — Scientific Validation
Expand experimental trials from N=36 → N=100
Submit for peer-review publication
Release all experimental datasets
Months 3–6 — Open-Source Release
Refactor the 21,300-line TypeScript/Python codebase
Build clear deployment documentation
Create a practitioner-facing GUI
Release under Apache 2.0 on GitHub
Months 4–9 — Pilot Deployments
Deploy with 3–5 UK Community Land Trusts (10 CLTs already expressed interest)
Run participatory workshops
Gather user feedback and iterate
Publish real-world case studies
Deliverables
Expanded peer-reviewed research
Fully open-source, documented codebase
Practitioner toolkit (GUI + guides)
3–5 pilot deployments with case studies
Training materials for 50+ practitioners
How will this funding be used?
Budget ($75,000)
Platform Development — $25,000
Production-grade refactor
Practitioner GUI
Deployment documentation
Test infrastructure
Technical debt elimination
Pilot Deployments — $20,000
Server infrastructure
On-site technical support & workshops
Integration with CLT systems
Participatory constitution-design sessions
Scientific Validation — $18,000
Expand trials to N=100
Data analysis + replication studies
Open-access publication fees
Training & Dissemination — $10,000
Training materials
Tutorials and documentation
Practitioner workshops
Conference presentations
Project Management — $2,000
Reporting and administration
Minimum Viable Scope ($25,000)
If only partial funding is available, we will deliver:
Full open-source release (code + documentation)
Expanded scientific validation (N=100, published results)
This guarantees delivery of the core public good independent of pilot deployment funding.
Who is on your team? What’s your track record?
Principal Investigator
Carlos Arleo
Independent Researcher
15 years experience in participatory governance & urban planning
Relevant Experience
15 years in participatory governance & urban planning (OMA, HASSELL, Atelier Terre)
Developed WFF in 12 months (21,300 lines of production code)
Published research on constitutional AI
Network & Collaborators
Active connections with UK Community Land Trusts (National CLT Network)
Links with European Urban Initiative
Track Record
To date, the project has produced:
A working, validated prototype
Published experimental results
Demonstrated self-repair capability
Strong practitioner interest (10 CLTs)
This funding turns a proven prototype into deployable public infrastructure.
What are the most likely causes and outcomes if this project fails?
Risk 1 — Complexity Limits Adoption
Mitigation: Practitioner GUI, training materials, user testing.
Probability: Medium.
Risk 2 — Slow Academic Peer Review
Mitigation: Open-source release is independent of publication timing.
Probability: Medium.
Risk 3 — Pilot Deployment Delays
Mitigation: $25k minimum scope ensures success even without pilots.
Probability: Medium-High.
Risk 4 — Computational Overhead
Current validation time is 4.9 seconds — well within operational needs.
Probability: Low.
If the Project Fails
Governments will continue adopting commercial AI systems optimised for efficiency rather than democratic principles. Without open-source safeguards, constitutional violations become structurally embedded in governance workflows.
This project creates the first public, transparent alternative.
How much money have you raised in the last 12 months?
$0.
The entire system has been self-funded through a PhD stipend.
Development, experiments, and publication have been completed without external support.
Why this request?
Transitioning from Research to Public Infrastructure
Current status:
✓ Working prototype
✓ Published validation
✓ Demonstrated full compliance
✓ Strong practitioner interest
What’s missing is the work required to make it accessible to non-technical users, scientifically expanded, and openly deployable.
Why Manifund
Manifund supports public-good research that communities can own.
Constitutional AI for democratic governance belongs in the public domain, not locked in proprietary models.
Why Now
AI is already shaping governance decisions. Constitutional safeguards must exist before proprietary systems become entrenched as default.
The research is complete.
The system works.
This funding makes it public infrastructure.