Why this program exists
MIR already governs state, authority, tools, memory, recovery, and evidence. This program asks a narrower question: Can a runtime monitor its own operating condition, allocate adaptation effort across timescales, and propose bounded structural or goal changes while preserving external policy and reviewability?
That is a research question, not a product capability claim. Teleodynamics is used here as a source of testable engineering hypotheses about coupled constraints and end-directed organization—not as proof that software is alive or conscious. Source: Deacon Source: Teleodynamic Learning
Key takeaways
- Internal control objectives remain subordinate to authorized external objectives and immutable policy.
- Structural adaptation is proposal-driven, costed, checkpointed, and evidenced.
- No-op and termination are valid outcomes when a change is not justified.
Research map
Six bounded lines of inquiry
Teleodynamics and Machine Intelligence
Separate biological theory, editorial analogy, and explicit runtime mechanism.
Adaptation
Learning and Structural Adaptation
Compare fast parameter updates with slower, governed changes to structure.
Control
Goal Systems and Self-Maintenance
Represent bounded operating objectives without open-ended self-interest.
Recovery
Reflective Control and Runtime Recovery
Let a critic diagnose and propose without granting it unchecked authority.
Evaluation
Runtime Benchmarks and Evaluation
Measure recovery, policy, evidence, cost, stability, and operator control.
Evidence
Reproducibility and Open Questions
Define replay levels, evidence packages, claim registers, and null results.
Core distinctions
| Distinction | Side A | Side B | Why it matters |
|---|---|---|---|
| External objective vs internal control objective | User or application intent | Runtime health, evidence, and resource constraints | Prevents internal controls from silently replacing user intent. |
| Parameter adaptation vs structural adaptation | Change values inside a fixed model | Change components, routes, rules, or topology | Requires different approvals, rollback, and evidence. |
| Self-maintenance vs self-preservation | Preserve valid runtime operation | Open-ended drive to preserve the system | The former can be engineered; the latter creates alignment and anthropomorphism risks. |
| Reflection vs authority | Diagnose and propose | Authorize and execute | Reflection must not grant itself power. |
| Resource budget vs literal thermodynamics | Compute, time, memory, energy, and cost | Physical thermodynamic claims | Keeps engineering accounting separate from unsupported physics. |
Evidence ladder
- Foundational theoryAccurate source interpretation and explicit domain boundaries.
- Formal or computational proposalDefined state, update rules, assumptions, and falsifiable predictions.
- PrototypeExecutable implementation, versioned artifacts, and documented failure modes.
- Reproducible benchmarkBaselines, repeated trials, environment details, and independent replication.
- Production patternOperational evidence across workloads, incidents, and maintenance cycles.
- Standardized contractMultiple implementers, governance process, compatibility tests, and stable semantics.
A claim should not move upward because its vocabulary sounds formal. It moves when evidence, replication, and operational limits support the stronger status.
Research boundary
What this program does not claim
- Machine consciousness, emotion, subjective experience, or intrinsic moral status.
- A universal theory of intelligence, quantum cognition, or cosmological explanation.
- Solved alignment or a self-preservation drive that may override authorized termination.
- A production-ready MiRuntime teleodynamic SDK, certification, or standardized contract.
Open questions
- Which viability variables are measurable, difficult to game, and portable across workloads?
- How should goal hierarchies preserve external authority while permitting bounded recovery?
- What cost model makes structural editing preferable to retries, rerouting, or no-op?
- Which evaluations require an independent verifier rather than the proposing model?
- How are operator override, termination, checkpoints, replay, and provenance guaranteed?
- What benchmark baselines distinguish useful adaptation from added complexity?
- How do self-maintenance controls avoid competing with human authority?
Source record
References
- Incomplete Nature: How Mind Emerged from Matter Primary source
Terrence W. Deacon. W. W. Norton & Company. Published 2011; last reviewed 2026-06-24 UTC. Foundational book.
- A thermodynamic basis for teleological causality Primary source
Terrence W. Deacon and Miguel García-Valdecasas. Philosophical Transactions of the Royal Society A. Published 2023; last reviewed 2026-06-24 UTC. Peer-reviewed research paper.
Enrique ter Horst and Juan Diego Zambrano. arXiv. Published 2026-03-11; last reviewed 2026-06-24 UTC. Research preprint.
National Institute of Standards and Technology. NIST. Published 2023-01-26; last reviewed 2026-06-20 UTC. Government framework.