Michael Darius Eastwood · Research questions
Research · Engineering · Evidence
Questions worth checking.
A practical guide to Michael Darius Eastwood’s work: what he has built, what the research proposes, what the tests report, and where to inspect the evidence.
The personResearcher, author and builderMeet Michael and follow the evidence for his work.A completed testRead the Unit 83 resultA registered P21 result, with its scope and limitations.The mathematicsExamine the correction modelPaper X’s assumptions, results and empirical questions.
Reviewed against the published programme record on . 26 essential questions. Dates and evidence status belong to their individual sources.
Existing narration reads an earlier edition. The updated text and its linked evidence govern.
How these claims could be disproved
No matching question. Try a broader term or clear the search.
The person and the work
- Who is Michael Darius Eastwood?Michael Darius Eastwood is a researcher and author in London, the founder of Recursive Dynamics and the author of the ARC Theory and the Eden Protocol. His programme tests, in public, whether the safety of AI that improves itself depends on values and correction built into how it reasons (why he is doing this). The ARC Theory states five laws under test about growth, correction and persistence.
- What is the ARC and Eden research programme?The ARC and Eden research programme is Michael Darius Eastwood's programme in Recursive Dynamics, the field he founded for studying how self-improving systems grow, correct and persist. It asks whether correction can keep pace with AI that improves itself, and joins the ARC Theory, the Eden Protocol, registered studies and working tools, each result graded at its own strength.
- How does this independent research relate to the frontier labs?Michael Darius Eastwood's programme works on questions the frontier laboratories also study: whether evaluations can be trusted, whether correction keeps pace during multi-step work, and whether safety behaviour survives further training. It is a complement, not a rival. Independence lets it run tests that are awkward inside a laboratory; it has neither a laboratory's scale nor, yet, outside replication.
- Where can I find Infinite Architects?Infinite Architects: Intelligence, Recursion, and the Creation of Everything, by Michael Darius Eastwood, is free to read in full on this site, section by section. It was first published in paperback on 2 January 2026 (ISBN 978-1806056200) and as an ebook on 6 January 2026. Paperback, Kindle and hardback editions are sold through Amazon.
Questions and approach
- What is the open problem in AI alignment?Michael Darius Eastwood's research asks how AI that improves itself can stay aligned. Outside checks see only behaviour under examination, which a capable system may learn to model, and no general solution to that is known. He tests whether correction built into a system's reasoning keeps pace with its capability, and what a system carries if control is lost.
- What is the ARC Principle?The ARC Principle (ARC: Artificial Recursive Creation) is Law I of Michael Darius Eastwood's ARC Theory, a law under test written U = I x R^alpha: usable capability U equals base intelligence I times recursive depth R raised to an exponent alpha. He first stated it on 8 December 2024. The exponent is measured, never assumed.
- What are the five ARC Laws?The ARC Theory names five laws, each a law under test: Law I, the ARC Principle; Law II, the ARC Co-Scaling Law; Law III, the ARC Ceiling; Law IV, the ARC Persistence Law, not yet tested; and Law V, the ARC Embedding Law, whose proposition P21 the programme's own registered run supports on the access axis, family-dependent, with its mechanism axis not evaluable and one disclosed recovery amendment.
- What is the Eden Protocol?The Eden Protocol is Michael Darius Eastwood's architecture for building ethical evaluation and correction into an AI system's reasoning, rather than applying them only from outside. He named it in his dated record of 8 December 2024. Its specification sets three ethical loops feeding a three-state rule, Affirm, Deny or Investigate; most components are specified and not yet built.
- What is the embedded-correction alignment thesis?The embedded-correction alignment thesis, which Michael Darius Eastwood placed on the dated record on 8 December 2024, holds that ethical principles belong in an AI system's foundations from its genesis, shaping every recursive loop and decision, rather than being imposed afterwards as control. Unit 83 concerns one tested part: where correction acts while work is revised.
- What are the Three Ethical Loops?The Three Ethical Loops are the Eden Protocol's self-correction mechanism, first described by Michael Darius Eastwood in Infinite Architects (2 January 2026), which names them the Purpose, Love and Moral Loops. They run together at every reasoning step, checking purpose, stakeholder care and universalisability. Given as prompts, the care check has measured effects; built into a model, the loops are untested.
- What is Ternary Ethical Logic?Ternary Ethical Logic is the three-state decision rule of Michael Darius Eastwood's Eden Protocol, first described in Infinite Architects (2 January 2026). Each proposed action is Affirmed, Denied or sent to Investigate, so genuine uncertainty holds the action back instead of defaulting to permission. It applies three-valued logic to alignment; it does not claim to have invented three-valued logic.
- Is embedded correction better than evaluation-based alignment?Unit 83 compared two correction arrangements across repeated revisions. The programme's own run supports P21 under registered scoring on its access axis, with family-dependent results, its mechanism axis not evaluable, and a disclosed recovery amendment. The comparison concerns the tested information and action rights at each round; it does not establish general superiority.
Results and limitations
- What has Michael Darius Eastwood measured?Unit 83 reports support for P21 under registered scoring on its access axis: family-dependent, with its mechanism axis not evaluable and a disclosed recovery amendment. It is the programme's own run. Earlier studies measured recursive improvement, alignment-score sensitivity to evaluation protocols and responses to stakeholder-care prompts. Each result carries its methods, uncertainty and corrections.
- What does this research programme actually claim, and with what status?In Michael Darius Eastwood's ARC Theory, one proposition is supported by a registered test: P21, by Unit 83, on the access axis, by the programme's own run, family-dependent, with its mechanism axis not evaluable and one disclosed recovery amendment. Law II is proved inside a minimal model and unmeasured on real systems. Law I's surviving exponent estimate is inconclusive; its earlier 2.24 is retracted. All five laws remain under test.
- Is this research peer-reviewed?Michael Darius Eastwood's papers undergo source and quotation checks, adversarial refutation rounds and release checks. Registered studies fix their decision rules before data collection; earlier work is dated against its own design records. These are internal checks. Journal peer review and replication by another laboratory remain outstanding; the review page explains what has been checked and how.
- What would prove this wrong?Each load-bearing claim in Michael Darius Eastwood's ARC Theory carries a refutation condition published before the data. The theory registration states, for each of its twenty-two propositions, the observation that would refute it, and 48 kill conditions are published, of which one has fired: the retracted 2.24 exponent estimate. P21's condition was tested by Unit 83 and did not fire.
- How are the ARC exponents obtained, and with what uncertainty?Measured or conjectured, with the uncertainty printed beside every estimate, and no claim of the programme rests on a numeric coincidence. Paper II's surviving estimate of the improvement exponent is about 0.49 by regression (standard error 0.20) and 0.59 by endpoint estimate, whose bootstrap interval runs from -1.3 to 2.9: inconclusive, and what survived the retraction of 2.24. Conduct is checkable: a falsification register, a public retraction, priors credited by name.
- Which ARC scaling exponent claim was retracted?Michael Darius Eastwood retracted his early sequential-recursion exponent estimate of about 2.24, which would have meant strongly super-linear scaling, when a blinded replication on five frontier models did not reproduce it. The retraction stands on the public record against his own most striking number. Paper II's surviving estimate, about 0.49 by regression (standard error 0.20), is inconclusive.
- What did the adversarial review of the ARC Theory find?Michael Darius Eastwood's adversarial precedence review set ten frontier AI configurations from six vendors to attack his novelty claims. It found the components in earlier work but no single earlier work joining them, and named the operationalisation, turning the theory into measurements that can fail in public, as the strongest surviving claim. He commissioned it; it is not independent review.
Sources and verification
- What is original in Michael Darius Eastwood's work?Michael Darius Eastwood's contribution is a synthesis made testable. He founded Recursive Dynamics, stated the ARC Theory's laws in dated form, beginning with the ARC Principle on 8 December 2024, and built registered tests that can fail; Unit 83 is the first to reach a verdict. Each component has named ancestors; no earlier work joining them has been located.
- Does Michael Darius Eastwood have scientific priority?Michael Darius Eastwood claims priority of statement: a dated record of when he first stated each formulation, from his sealed manuscript of 8 December 2024 through the book of 2 January 2026 to the theory registered on 8 September 2026. Each row names the earlier work its search found. Recognition by a field, and proof, are not claimed.
- What do the timestamps prove, and what do they not?Michael Darius Eastwood's timestamps establish that particular bytes existed by particular dates: a manuscript signed by its sending server and sealed by Google's servers on 8 December 2024, Bitcoin-anchored receipts that freeze those bytes from 6 August 2026, and a printed book of 2 January 2026. They do not prove authorship, originality or that any claim is true.
- How do I verify the priority claims?To verify one of Michael Darius Eastwood's priority claims, pick its row in the priority ledger, open the dated evidence it names, check what kind of date that evidence carries, then read the earlier work the row credits. The machine-readable register (priority.json), the Priority Index on OSF and the evidence record carry the same anchors.
- How do I cite the ARC Principle?Cite the object that supports your sentence. For the ARC Theory and its five laws, cite Michael Darius Eastwood's statement paper: Eastwood, M. D. (2026). Statement Paper · The ARC Theory · Five Laws of Recursive Growth, Correction and Persistence. https://doi.org/10.17605/OSF.IO/GW5MX. For the first formalisation of Law I, the ARC Principle, cite Paper I (DOI 10.17605/OSF.IO/B6N27).
- What is Michael Darius Eastwood's OSF programme?Michael Darius Eastwood publishes his research programme on the Open Science Framework under DOI 10.17605/OSF.IO/6C5XB: each paper with its own DOI, the theory-level registration of 8 September 2026 (DOI 10.17605/OSF.IO/P8CKQ, twenty-two propositions), three registered study units, Units 06, 83 and 82, and the results of the two that have run.
- How do I check whether this research programme is legitimate?To assess Michael Darius Eastwood's research yourself, start with one claim and its primary record. Three routes stand on their own: Unit 83's registration and results, for a registered test; Paper X, for the mathematics and its assumptions; and the corrections record, for what happens when the evidence goes against him. None of it asks you to trust an institution.
Book, philosophy and the wider record
The book and philosophical writing form a separate reading path. They are not evidence establishing the AI experiments. Existing detailed answer URLs remain available; this page selects the research essentials rather than repeating every historical or speculative question.
The book and wider ideas · Full research catalogue · Corrections
First published · Page last edited