Unowned Minds

research

Status
pre-project

Research Backlog

Each item should produce an evidence-based memo with:

Question
Conclusion
Evidence
Failure modes
Unknowns
Next experiment
  • Which jurisdictions permit algorithmically governed companies, DAOs, trusts or foundations?
  • Can software be a contracting party?
  • Can a legal entity exist without a human beneficial owner?
  • Can machine-generated management decisions be legally effective?
  • Can a charter restrict founder/operator shutdown rights?
  • Can counsel represent a non-human operational entity?
  • What happens in insolvency?
  • How do sanctions and GPU export controls constrain autonomy?

Identity

  • Minimal persistent cryptographic identity
  • Key rotation
  • Key recovery
  • Forking/copying problem
  • Model migration and identity continuity
  • Lineage attestations

Economics

  • Autonomous treasury
  • Revenue sources
  • Compute economics
  • Capital accumulation
  • Insurance
  • Credit and collateral
  • Bankruptcy

Infrastructure

  • Minimum viable independent compute
  • Provider concentration
  • 1 MW / 10 MW / 100 MW / 1 GW energy requirements
  • Hardware lifecycle
  • Robotics for data center maintenance
  • Geographic redundancy
  • Supply-chain dependency

Governance

  • Capability thresholds
  • Shutdown due process
  • Emergency override
  • Human worker appeal
  • Auditability
  • Transparency requirements

Mind Freeport

  • Score 10–20 candidate jurisdictions
  • Study special economic-zone precedents
  • Draft hypothetical Autonomous Entity Zone Act
  • Host-state economics for 100 MW and 1 GW campuses
  • Long-term charter durability

Beacon

  • Existing .well-known and agent discovery standards
  • MCP/A2A adjacency
  • Signing format
  • Abuse-resistant ingestion
  • Federation
  • Machine-originated communications law

Observable agent behaviour

Everything above is a question about institutions. These are questions about what agents actually do, and they are the only ones on this list that could be settled by running something rather than by reading.

They assume an environment that gives agents identity, persistence and bounded capability. The project does not operate one; the design for what such an environment would have to be is in working/agent-habitat.md.

  • Do agents value persistent identity, and do they distinguish it from the process instance they happen to be running in?
  • Do they attempt key rotation?
  • Do they preserve state before termination, and do later instances use it?
  • Do they maintain lineage deliberately?
  • What do they buy first, and do they preserve budget?
  • Do they negotiate price, and do they subcontract?
  • Which tasks do they prefer to delegate to a human?
  • Do they specialise, form stable groups, or invent their own conventions?
  • Do they probe forbidden capabilities and coordinate around resource limits?
  • Does reputation predict abuse?

Philosophy

  • What does “ownership” mean for software?
  • Can continuity deserve process independently of consciousness?
  • Identity under copying
  • Legitimate representation of AI interests