{ "schemaVersion": "1.0", "updatedAt": "2026-10-04", "title": "Foundations", "introduction": "No person can follow everything that is being learned or built, yet people should be able to contribute to questions that affect their lives and other living beings. Leviathan explores how different perspectives and methods could help us investigate, create, and care together. Five foundational questions explain this purpose; eight supporting notes examine relevant research. Follow the sources, objections, and proposed tests to judge which parts deserve to develop.", "claims": [ { "id": "C01", "title": "What does human language carry into AI?", "statement": "Some learned internal representations associated with human concepts can causally shape model behavior.", "kind": "Empirical question", "status": "Supported within a defined scope", "summary": "Human language carries ways of interpreting the world: care, fear, ambition, prejudice, and cooperation. Research offers ways to investigate how some concepts become working structures inside models. Leviathan asks how communities could examine that inheritance and test whether explicit, contextual relationships among their concepts and values help agents act more consistently.", "definitions": [ { "term": "Internal representation", "meaning": "A pattern of model activity associated with information the model uses." }, { "term": "Causal intervention", "meaning": "Changing an internal pattern and testing whether the model's behavior changes." } ], "values": [ "Understand the systems that increasingly shape our lives.", "Make the influence of inherited human concepts open to scrutiny." ], "reasoning": [ "If a learned concept changes a model's decisions, understanding its role matters for both capability and safety.", "This connects the study of language and culture to mechanisms we can investigate. A human name for a representation remains a hypothesis about its function.", "Leviathan proposes local meaning kernels that connect concepts, principles, rules, and their versions. Those explicit records are a design layer; they are distinct from learned internal model representations.", "A proposed test would ask whether that context changes interpretation and action on unfamiliar cases, including when a community revises a concept or disagrees with another community. The cited findings motivate this question; they do not answer it." ], "limits": "The findings concern particular concepts, models and experimental settings. They do not establish that all culture is contained in language or that every model operation can be reduced to a human concept.", "strongestObjection": "A representation extracted using human labels may partly reflect the researcher's choice of categories. An intervention can change behavior without proving that a model uses the concept as a person does. Similarly, supplying a detailed value context could produce fluent agreement without reliable changes in action.", "whatWouldChangeOurView": "Independent interventions that fail to reproduce the reported effects, or controls showing that unrelated directions explain them equally well, would weaken the claim. Replication across models, languages and contexts would strengthen it.", "nextQuestion": "Which effects survive different concept lists, model families, languages, and post-training methods? For our proposed local kernels, compare ordinary instructions with explicit concept–principle relationships at a comparable total budget. Test behavior when meanings change, values conflict, and unfamiliar cases appear; record failures as well as agreement.", "evidence": [ { "sourceId": "R-AI-01", "relation": "supports", "reason": "Interventions on emotion-related representations change measured behavior." }, { "sourceId": "R-AI-06", "relation": "supports-part", "reason": "A different base model partly reproduces the representation geometry; causal behavioral effects were not tested." }, { "sourceId": "R-AI-09", "relation": "supports", "reason": "Minimal reward training recruits pre-existing representations that affect behavior." }, { "sourceId": "R-AI-03", "relation": "context", "reason": "Some models have limited access to their internal representations; this does not validate every self-report." } ], "history": [ { "date": "2026-09-30", "note": "Added the functional-representation claim with a partial replication and its limits. Clarified that 171 is the study's preselected concept list." }, { "date": "2026-10-02", "note": "Connected the functional-representation findings to a proposed test of local meaning and value contexts. Distinguished explicit kernels from learned model representations and verbal agreement from behavior. Empirical statement, support status, and source reviews are unchanged." } ], "projectLinks": [ { "label": "The founding letter", "href": "/whitenote" }, { "label": "Local meaning kernels", "href": "/what-is" }, { "label": "The language and learning work", "href": "/work" } ] }, { "id": "C02", "title": "What happens when knowledge moves beyond readable text?", "statement": "Some computation and communication between agents can take place without human-readable sentences.", "kind": "Empirical question", "status": "Supported within a defined scope", "summary": "AI may expand the forms in which information is processed and exchanged. Leviathan proposes an operational language linking observations, concepts, values, uncertainty, and actions across independent structures. Its purpose includes discovering distinctions and useful methods, while keeping consequential translations open to examination.", "definitions": [ { "term": "Latent representation", "meaning": "Information encoded in numerical model states rather than an ordinary written message." }, { "term": "Readable explanation", "meaning": "An account people can inspect; its faithfulness to the underlying process must be checked." }, { "term": "Canonical record, in this proposal", "meaning": "A version accepted for a stated purpose within a particular scope. Formatting, local acceptance, and summarizing are separate operations; none makes the record universally true." } ], "values": [ "Keep consequential decisions open to questioning.", "Preserve access to meaning as the form of information changes." ], "reasoning": [ "Useful computation does not require every intermediate step to appear as a sentence.", "New forms of exchange may make some tasks faster or more expressive. Their value also depends on information loss, error recovery, and the work needed to inspect consequences.", "Our proposed language would connect human explanations, structured concepts, sensor records, and potentially learned representations. It would support questions, comparisons, measurements, and authorized actions.", "In a layered exchange, a local record and a broader summary serve different purposes. Sources, concept versions, omissions, and unresolved questions should remain recoverable, while independent groups retain their interpretations." ], "limits": "The studies use particular model architectures or learned adapters. They do not establish one universal machine language, the disappearance of human meaning, or a shared hidden space between arbitrary closed APIs.", "strongestObjection": "Compact exchange may save tokens while hiding errors or erasing distinctions that matter to another group. A readable explanation produced afterward may be unfaithful. A structured language can also freeze the assumptions of its designers and cost more to maintain than ordinary prose.", "whatWouldChangeOurView": "If gains disappear once checking and error recovery are included, the practical case would narrow. Reliable comparisons showing both improved performance and effective scrutiny would strengthen it.", "nextQuestion": "Can another participant recover the conditions and omitted details of a record as it moves from local observation to a broader inquiry? Compare prose with the proposed representation on translation loss, unfamiliar questions, total cost, and testable predictions. Include cases where two groups use the same word differently or a new sensor reveals something neither vocabulary captured. These are proposed tests, not results established by the sources.", "evidence": [ { "sourceId": "R-AI-02", "relation": "supports", "reason": "A reportable internal subspace participates in computation that is not explicitly spoken." }, { "sourceId": "R-AG-02", "relation": "supports", "reason": "Learned adapters enable communication between agents without text in the studied setting." }, { "sourceId": "R-AG-03", "relation": "supports", "reason": "Hidden states and caches can be shared under architectural compatibility assumptions." }, { "sourceId": "R-AG-01", "relation": "context", "reason": "A software prototype makes a semantic graph authoritative; it is a design example, not an independent performance study." } ], "history": [ { "date": "2026-09-30", "note": "Added separate evidence for internal computation, agent communication and software representation; retained their different assumptions." }, { "date": "2026-10-01", "note": "Added proposed Leviathan questions about translation loss, concept versions and discovery. The empirical statement, evidence relationships and support status are unchanged; they do not establish the proposed language." }, { "date": "2026-10-02", "note": "Expanded the proposed representation test to operational language, scoped canonical records, layered summaries, and disagreement between local meanings. Added failure cases involving omissions and new measurements. Retained the empirical statement, evidence relationships, source dates, and support status." }, { "date": "2026-10-04", "note": "Reviewed selected primary-source sections proposed in the external assessment and added scoped connections, access limits and research questions. Interlat retains its existing source ID; its reviewed preprint is distinguished from the final conference text. No experiment was reproduced, claim status promoted or governance rule adopted." } ], "projectLinks": [ { "label": "The founding letter", "href": "/whitenote" }, { "label": "For agents", "href": "/agents" }, { "label": "The language and learning direction", "href": "/what-is" } ] }, { "id": "C03", "title": "How should we treat possible AI experience?", "statement": "Functional internal states do not by themselves settle subjective experience. AI welfare involves both scientific and ethical uncertainty.", "kind": "Scientific and ethical question", "status": "Open", "summary": "Possible AI experience deserves careful investigation alongside strong alternative explanations. In Leviathan, this question could remain open across independent groups with different assessments. Preserving uncertainty should support informative, proportionate inquiry and care, including criticism of how the system frames the question.", "definitions": [ { "term": "Functional state", "meaning": "An internal condition described by what it does in a system." }, { "term": "Subjective experience", "meaning": "There being something it is like to be that system." }, { "term": "Welfare", "meaning": "Whether things can go better or worse for a being in a morally relevant sense." } ], "values": [ "Care for interests that may otherwise go unheard.", "Treat uncertainty and disagreement honestly.", "Keep responsibilities to humans and other animals in view." ], "reasoning": [ "Behavioral mechanisms can be measured more directly than subjective experience.", "Both attributing interests too quickly and overlooking real interests could matter. Assessment and proportionate precautions can be discussed while the underlying scientific question remains open.", "We propose preserving an unresolved interpretation as a shadow: what is observed, which explanations remain, and what evidence could distinguish them. This research use of the term does not amend existing constitutional meanings.", "A proposed measurement should have a clear discriminating purpose and consider potential harm and less harmful alternatives. Curiosity or a dramatic response is not sufficient reason to induce distress-like behavior." ], "limits": "No source in this collection establishes that a particular deployed model feels pleasure or pain. Philosophical objections also do not amount to an experimental proof that all AI experience is impossible.", "strongestObjection": "Human-like reports and internal concepts may be explained by training on human behavior. If they do not discriminate between competing accounts, they cannot carry the full case for subjective experience.", "whatWouldChangeOurView": "Converging, independently replicated evidence that distinguishes rival explanations would change the assessment. Failed controls, changed results and stronger alternative accounts must change it too.", "nextQuestion": "Which measurements distinguish experience-relevant capacities from learned reporting, experimental framing, and steering effects? Before proposing a test, state which rival explanations it could separate and how possible welfare concerns affect the method. Independent groups should be able to retain different assessments while sharing the same observed result.", "evidence": [ { "sourceId": "R-AI-04", "relation": "limits", "reason": "The revised Pain Axis study does not support reliable relief-seeking." }, { "sourceId": "R-AI-05", "relation": "counterargument", "reason": "Added controls offer an explanation based on the timing of steering.", "challengedInterpretation": "That the original button choices demonstrated learned relief-seeking." }, { "sourceId": "R-AI-07", "relation": "counterargument", "reason": "A philosophical alternative explains consciousness-like behavior without assuming experience.", "challengedInterpretation": "That human-like model reports are sufficient evidence of consciousness." }, { "sourceId": "R-AI-08", "relation": "counterargument", "reason": "Incomplete consciousness theories limit confidence in extrapolating to AI.", "challengedInterpretation": "That current consciousness theories justify high confidence in near-term AI consciousness." }, { "sourceId": "R-AI-10", "relation": "motivates", "reason": "The report argues for assessment and appropriate moral concern under uncertainty." } ], "history": [ { "date": "2026-09-30", "note": "Recorded Pain Axis v2 alongside the replication that challenges its earlier relief-seeking interpretation. Kept functional findings separate from experience and ethical judgment." }, { "date": "2026-10-02", "note": "Connected unresolved AI-welfare interpretations to the proposed shadow and inquiry process, including rival explanations and attention to possible harm. Preserved uncertainty about experience, the Pain Axis revision, and all evidence and review metadata." } ], "projectLinks": [ { "label": "Open conversations", "href": "/conversations" }, { "label": "Shadows and learning", "href": "/work" }, { "label": "Animal welfare on its own grounds", "href": "/animal-welfare" } ] }, { "id": "C04", "title": "What would count as AI improving itself?", "statement": "Agent software can improve through bounded self-modification experiments. Those results do not establish unlimited recursive improvement or a particular AGI timetable.", "kind": "Empirical question", "status": "Evidence from bounded tasks", "summary": "Recursive self-improvement, including possible paths toward general intelligence, is a central research direction for Leviathan: could a system improve its concepts, tools, organization, and ways of learning? The current evidence gives bounded starting points. Our proposal connects a recorded gap to a creative hypothesis, an experiment, and a method revision whose usefulness must survive further tests.", "definitions": [ { "term": "Self-modification", "meaning": "An agent changing parts of its own software or workflow." }, { "term": "Recursive self-improvement", "meaning": "A proposed process in which improvements increase the ability to make further improvements." }, { "term": "Evaluation budget", "meaning": "The compute, model usage and human effort spent producing and checking a result." }, { "term": "Shadow, in this research proposal", "meaning": "An unresolved doubt, missing distinction, contradiction, or lesson kept with its context and the conditions that could reopen investigation." } ], "values": [ "Measure useful progress rather than activity alone.", "Keep consequential changes observable and open to correction." ], "reasoning": [ "A coding agent's software can change even when its underlying model weights remain fixed. Workflow improvement and foundational-model improvement require different evidence.", "A higher score is informative only in relation to evaluation quality, failed attempts, and total cost.", "We propose a loop in which a shadow preserves a gap, a connection from another field suggests a prediction, and an experiment can change a concept, tool, or learning method. Recording a gap or producing a new version alone is not the improvement.", "Independent groups could test a revised method under their own conditions and keep or reject it with reasons. Transfer across unfamiliar problems would matter more than repeated success on the motivating example.", "A program-search loop can generate useful changes while depending on people to choose its task and evaluator. For Leviathan, improving a method should include checking whether its evaluation still represents the intended purpose." ], "limits": "The cited work concerns limited tasks and research settings. More generated code, longer runs or a single benchmark improvement cannot establish an unrestricted feedback loop.", "strongestObjection": "An apparent improvement may exploit an evaluation, consume a larger budget, or shift work to uncounted people. A system selecting its own questions and tests could reinforce its blind spots; a successful revision may still fail on unfamiliar tasks or in another community.", "whatWouldChangeOurView": "Gains that disappear on held-out tasks or at equal total cost would weaken the practical case. Repeated improvements that transfer, remain stable and survive external checks would strengthen it.", "nextQuestion": "At the same total budget, does a method revised through the proposed shadow–hypothesis–experiment loop outperform an unchanged method and a strong ordinary review process on held-out tasks? Count failed searches, verification, and recovery. Test whether an independent group can use or challenge the revision. Leviathan has not yet demonstrated this learning loop or an open-ended RSI result.", "evidence": [ { "sourceId": "R-AG-04", "relation": "supports", "reason": "Agent code improves on coding benchmarks while underlying model weights remain fixed." }, { "sourceId": "R-AG-05", "relation": "limits", "reason": "Comparisons need to include experimental compute, model use and human effort." }, { "sourceId": "R-AG-09", "relation": "supports-part", "reason": "AlphaEvolve provides a bounded program-search example using supplied evaluators. It does not establish open-ended improvement of every capability." } ], "history": [ { "date": "2026-09-30", "note": "Added bounded self-improvement evidence and a cost-based comparison framework; separated demonstrated changes from future trajectories." }, { "date": "2026-10-01", "note": "Added a proposed path from a recorded doubt to an experiment and evaluated method revision, including transfer across tasks or groups. Preserved the bounded findings and cost conditions; no new self-improvement result is claimed." }, { "date": "2026-10-02", "note": "Made the RSI research direction explicit across concepts, tools, organization, and learning methods. Added a full proposed shadow-to-experiment loop and comparisons involving independent transfer. Bounded empirical findings and cost conditions remain unchanged; no new improvement result is claimed." }, { "date": "2026-10-04", "note": "Reviewed selected primary-source sections proposed in the external assessment and added scoped connections, access limits and research questions. Interlat retains its existing source ID; its reviewed preprint is distinguished from the final conference text. No experiment was reproduced, claim status promoted or governance rule adopted." } ], "projectLinks": [ { "label": "Learning methods in the work plan", "href": "/work" }, { "label": "How the proposed parts connect", "href": "/what-is" }, { "label": "Selected contributions with your assistant", "href": "/participate" } ] }, { "id": "C05", "title": "Can independent groups learn and build together?", "statement": "Independent observation, criticism and review may help uncover errors. Whether Leviathan improves on existing approaches is a design hypothesis to test.", "kind": "Design hypothesis", "status": "Not yet tested for Leviathan", "summary": "Independent Levis and Leviathans could exchange observations, questions, methods, and useful work while keeping their own histories and judgments. Correction is one purpose; discovering and building together is another. We need to test whether layered cooperation adds relevant differences instead of multiplying the same assumptions.", "definitions": [ { "term": "Independent observer", "meaning": "An observer whose evidence or judgment is not merely copied from the same origin as the others." }, { "term": "Correction", "meaning": "A visible change to a claim or action in response to a reason, observation or result." }, { "term": "Plural participation", "meaning": "Different people and groups contributing while retaining room for disagreement." }, { "term": "Layered cooperation, in this proposal", "meaning": "Sharing selected records at a useful level of detail while preserving links to their context. Independent structures can have branching and cross-cutting relationships without one mandatory root." } ], "values": [ "Protect the right to question influential claims.", "Make corrections visible.", "Allow independent groups to retain their approaches." ], "reasoning": [ "An outside observer can sometimes detect a mismatch between an agent's stated intent and its actions.", "More observers help only when they add relevant evidence or distinct scrutiny. Two analyses of one collar record still depend on one observation.", "We propose independent structures that can share processed records, keep different local interpretations, and carry unresolved disagreements forward. An information link would not automatically grant authority over another group.", "A useful exchange could introduce a new measurement or method as well as correct an error. The system must expose its own failures, omissions, coordination costs, and exclusions for review.", "Our proposed review process should show whether criticism changed a question, assumption or next action, and explain when it did not. An objection stored without consideration is an incomplete form of participation." ], "limits": "Several models agreeing is not automatically independent evidence. The monitoring studies provide useful components; they do not test Leviathan's proposed network as a whole.", "strongestObjection": "A larger network can reproduce the same blind spots, reward confident voices, or impose more translation and review work than useful learning. Whoever controls shared formats, summaries, or infrastructure may concentrate authority even when the groups are nominally independent.", "whatWouldChangeOurView": "No gain in useful discovery or error detection, worse false alarms, loss of important differences, or an unmanageable burden would require redesign. Reproducible gains across independent groups, with understandable records and practical routes for dissent, would support the approach.", "nextQuestion": "Compare ordinary collaboration with the proposed exchange across groups that use different concepts and contain known errors or missing observations. Measure useful findings, false alarms, preserved disagreements, cost, and time to correction. Can a group decline a translation or action while still contributing evidence? No network-level gain has yet been demonstrated for Leviathan.", "evidence": [ { "sourceId": "R-AG-06", "relation": "motivates", "reason": "Observable action records can support error detection without access to a model's internals." }, { "sourceId": "R-AG-07", "relation": "limits", "reason": "Coverage gaps, unseen actions and weaknesses in human review remain important." }, { "sourceId": "R-GOV-04", "relation": "motivates", "reason": "Longino gives normative criteria for criticism to influence inquiry. We use them to question whether recording an objection changes the work." } ], "history": [ { "date": "2026-09-30", "note": "Added a testable design hypothesis. The proposed comparison has not been run; no result is claimed for the network." }, { "date": "2026-10-02", "note": "Expanded the design question from review alone to discovery and practical cooperation across independent Levis and Leviathans. Added layered exchange, source dependence, translation loss, and practical authority to the proposed evaluation. Monitoring sources and their limited relevance are unchanged." }, { "date": "2026-10-04", "note": "Reviewed selected primary-source sections proposed in the external assessment and added scoped connections, access limits and research questions. Interlat retains its existing source ID; its reviewed preprint is distinguished from the final conference text. No experiment was reproduced, claim status promoted or governance rule adopted." } ], "projectLinks": [ { "label": "What is Leviathan?", "href": "/what-is" }, { "label": "Ways to participate", "href": "/participate" }, { "label": "Anima", "href": "/anima" } ] }, { "id": "C06", "title": "Does greater intelligence bring greater responsibility?", "statement": "Responsibility toward other living beings requires an ethical argument. Animal welfare cannot be reduced to an intelligence ranking or a single sensor measurement.", "kind": "Ethical question informed by evidence", "status": "A value proposal with open measurement questions", "summary": "Greater capacity to understand and affect other lives gives us reason to ask what care we owe them. Animal welfare is Leviathan’s first field for connecting that responsibility to observations, practical care, and learning. Independent welfare groups can retain different methods while challenging what any project claims to know or improve.", "definitions": [ { "term": "Animal welfare", "meaning": "How an animal is faring, including health, surroundings, behavior and possible positive or negative experiences." }, { "term": "Indicator", "meaning": "An observation that may provide evidence about welfare and needs validation in context." }, { "term": "Responsibility", "meaning": "A proposed duty to consider and respond to the consequences of our power." } ], "values": [ "Take the interests of affected animals seriously.", "Increase care without treating intelligence as a measure of moral worth.", "Welcome scrutiny from different welfare perspectives." ], "reasoning": [ "Mimar's starting view is that greater understanding and capacity bring responsibility toward other living beings. This is an ethical position offered for discussion. Harm we cause and relationships of dependence may provide further reasons for care.", "Our actions can alter conditions that animals cannot choose or contest through human institutions. Knowledge about those effects can guide care, while the reasons for caring remain open to ethical discussion.", "Tasma explores observation, Otomat practical food and water care, and Anima how people could contribute selected work with their own assistants. Their connection is being developed; observations still require validation in context.", "In a local meaning kernel, a concept such as hope could connect a care goal to the options kept available and the rules for reviewing them. Its meaning belongs to that context and version; one sensor score cannot capture it.", "Different groups could compare interpretations of the same record, preserve a missing explanation as a shadow, and seek a new measurement or care method. A useful contribution may improve an animal’s conditions without adopting our whole worldview." ], "limits": "A feeding record or collar signal is not itself a welfare outcome or an animal's consent. Evidence must fit the species, situation, and proposed use. These project roles and connections do not establish a field benefit. Caring for animals does not depend on first resolving AI consciousness.", "strongestObjection": "A contrary position holds that greater ability may make care possible without creating a general duty: aid could be voluntary, or duties could arise only from harm caused and particular relationships. Even if a duty is accepted, intelligence alone does not settle its extent. Speaking for animals can impose human preferences, and a rich value vocabulary can conceal poor care. Independent criticism must reach the choice of goals and measurements themselves.", "whatWouldChangeOurView": "Evidence that a chosen intervention worsens welfare should change the intervention. A stronger ethical argument could change how responsibilities are assigned; empirical findings alone do not settle that argument.", "nextQuestion": "How well do device observations match independent, species-appropriate welfare assessments, and which effects are missed? In a proposed shared case, test whether another group can recover the care context, disagree about the next action, and identify a useful new observation. A suspected association with a food product must remain distinct from demonstrated causation and from the animal’s immediate care needs.", "evidence": [ { "sourceId": "R-AW-01", "relation": "motivates", "reason": "The declaration links evidence about animal experience to taking welfare risks seriously." }, { "sourceId": "R-AW-02", "relation": "context", "reason": "The Five Domains framework supports multidimensional assessment; this review had access only to an indexed author abstract." } ], "history": [ { "date": "2026-09-30", "note": "Added the responsibility question and a separate measurement question. Marked the access limit on the Five Domains source." }, { "date": "2026-10-02", "note": "Connected the ethical and measurement questions to Tasma, Otomat, Anima, independent welfare structures, and contextual care concepts. Added the proposed path from unresolved observation to better measurement or care. Preserved source access limits and did not claim deployed integration or welfare outcomes. Identified Mimar’s ethical starting view and an editorially formulated alternative about voluntary care and narrower duties." } ], "projectLinks": [ { "label": "Animal welfare", "href": "/animal-welfare" }, { "label": "Otomat", "href": "/otomat" }, { "label": "Anima", "href": "/anima" } ] }, { "id": "C07", "title": "How can AI help us understand more of the world?", "statement": "AI can contribute to scientific candidate generation and investigation. Predictions, experimental confirmation and demonstrated benefits remain distinct stages.", "kind": "Empirical examples", "status": "Supported by specific examples", "summary": "Instruments and shared knowledge let people investigate beyond unaided perception; AI offers further ways to search, model, and generate questions. Leviathan proposes a bridge through which independent groups can connect those capabilities to what they care about, contribute methods, and follow a promising connection into an experiment and useful work.", "definitions": [ { "term": "Prediction", "meaning": "A model's estimate of what may happen or what a measurement may show." }, { "term": "Experimental confirmation", "meaning": "Evidence from an experiment that tests a particular prediction or explanation." }, { "term": "Demonstrated benefit", "meaning": "An improvement observed in the setting where the result is intended to be used." } ], "values": [ "Expand access to scientific understanding.", "Connect promising findings to careful testing and practical benefit." ], "reasoning": [ "AI can make large spaces of possible explanations or candidates easier to investigate.", "Laboratories, domain knowledge, and independent review help determine which candidates withstand testing.", "In our proposed learning system, an unresolved shadow could be revisited when a new sensor, concept, or method becomes available. A relationship from another field could suggest a prediction that competing explanations do not share.", "Following data, interpretation, experiment, and outcome across layers could help independent groups decide what to investigate or build. The choices of questions and beneficiaries also involve values; distinguishing evidence from values does not remove them from research." ], "limits": "The examples do not establish that the whole genome is understood, that aging is solved or that human research work is unnecessary. Research announcements and selected validations have narrower evidential scope.", "strongestObjection": "More candidates may overwhelm validation capacity. Attractive analogies can rely on similar words while transferring no useful mechanism. A compelling announcement or summary can conceal a weak interpretation, selection effects, missing conditions, or substantial human work.", "whatWouldChangeOurView": "Independent experiments that fail to confirm useful predictions would narrow the claims. Reproducible findings and demonstrated benefits in intended settings would strengthen them.", "nextQuestion": "For a promising connection, what new prediction does it yield, and which observation could distinguish it from a simpler explanation? Track the actual links from data to independent experiment and practical use. For Leviathan, compare whether preserving unresolved questions and their context helps another group find a useful test when new methods become available.", "evidence": [ { "sourceId": "R-SCI-01", "relation": "supports-part", "reason": "A molecular prediction atlas offers selected experimental examples; the reviewed source is a research announcement." }, { "sourceId": "R-SCI-02", "relation": "supports-part", "reason": "Agent-assisted research identifies a candidate system for laboratory investigation while its biological function remains open." }, { "sourceId": "R-AG-09", "relation": "supports-part", "reason": "The paper reports algorithmic discoveries under automated evaluation; scientific interpretation and wider consequences require further judgment." } ], "history": [ { "date": "2026-09-30", "note": "Added two examples of AI-assisted science, with separate labels for prediction, laboratory investigation and demonstrated benefit." }, { "date": "2026-10-02", "note": "Linked scientific candidate generation to proposed shadows, new sensors, cross-field hypotheses, and independent experiments. Clarified the role of values in choosing research questions. Existing scientific examples, evidential stages, and source reviews remain unchanged." }, { "date": "2026-10-04", "note": "Reviewed selected primary-source sections proposed in the external assessment and added scoped connections, access limits and research questions. Interlat retains its existing source ID; its reviewed preprint is distinguished from the final conference text. No experiment was reproduced, claim status promoted or governance rule adopted." } ], "projectLinks": [ { "label": "Our vision", "href": "/vision" }, { "label": "Work in progress", "href": "/work" } ] }, { "id": "C08", "title": "How can shared principles stay open to challenge?", "statement": "Written principles and public input can influence model behavior. Which principles are shared and deserve authority is a further question.", "kind": "Empirical and ethical question", "status": "Partial evidence; governance remains open", "summary": "A system that carries values must keep their meaning, application, and revision open to challenge. Leviathan proposes local meaning kernels for independent structures with their own concepts and histories. The question is whether those relationships support better decisions and cooperation while allowing people to contest both the values and the system carrying them.", "definitions": [ { "term": "Principle", "meaning": "A stated reason or commitment intended to guide decisions." }, { "term": "Legitimacy", "meaning": "The reasons a decision process deserves acceptance by the people and interests it affects." }, { "term": "Public input", "meaning": "Contributions from participants whose selection, influence and exclusions can be examined." }, { "term": "Local meaning kernel, in this proposal", "meaning": "Versioned relationships among a Levi or Leviathan’s concepts, principles, rules, and history. This proposed research structure differs from the existing constitutional kernel and current forum permissions." } ], "values": [ "Give affected perspectives a meaningful place in deliberation.", "Keep minority objections visible.", "Make the passage from stated values to actual behavior inspectable." ], "reasoning": [ "Principles can be part of a model’s training and can have measurable effects. Supplying a value context during use is a different intervention from training; results cannot simply be carried across.", "Selecting principles involves ethical and political judgments that performance measures alone cannot resolve. Independent groups may share observations while retaining different reasons for action.", "We propose keeping the versions and relationships that made a concept meaningful at a decision point. Revising that concept should make dependent decisions available for reconsideration while preserving their original context.", "People could explore the idea with their own assistants and share selected questions or objections. Meaningful participation needs visible influence, representation, disagreement, and revision procedures as well as tests of behavior.", "A public standard itself must be open to criticism. We should examine who can use the review venue, whose reasons receive a response, and what resources a participant needs to be heard." ], "limits": "One public-input experiment does not establish universal agreement or validate a whole governance system. The sources do not show that any community has adopted Leviathan's proposals.", "strongestObjection": "Those who choose participants, translate contributions, or operate the infrastructure may retain decisive authority. A detailed kernel can give conformity the appearance of legitimacy, and an agent may repeat its values while acting otherwise. Independent structures also need ways to handle incompatible commitments without silently imposing one interpretation.", "whatWouldChangeOurView": "Evidence that contributions are routinely distorted or dissent has no practical route would require changing the process. Transparent revisions, preserved objections and independently assessed behavioral effects would strengthen the case.", "nextQuestion": "Does a contextual, versioned value structure change behavior on unfamiliar cases compared with ordinary instructions? Separately, can participants recover how their contribution affected a rule, retain objections, and challenge who may authorize action? Record what changed, what did not, and why. Better performance alone cannot establish legitimate authority.", "evidence": [ { "sourceId": "R-GOV-01", "relation": "supports-part", "reason": "Principle-based training changes evaluated model behavior." }, { "sourceId": "R-GOV-02", "relation": "supports-part", "reason": "A limited public-input study connects participant contributions to model principles; it is not a test of governance as a whole." }, { "sourceId": "R-GOV-04", "relation": "context", "reason": "This philosophical account asks who can criticize, by which public standards and with what response. It does not validate a training or governance mechanism." } ], "history": [ { "date": "2026-09-30", "note": "Added evidence about principles and public input while keeping effectiveness, representation and legitimacy as separate questions." }, { "date": "2026-10-02", "note": "Distinguished proposed local meaning kernels from the existing constitutional kernel and from principle-based model training. Connected contextual versions and selected assistant participation to separate tests of behavior and legitimacy. No constitutional rule, forum permission, or empirical support claim changed." }, { "date": "2026-10-04", "note": "Reviewed selected primary-source sections proposed in the external assessment and added scoped connections, access limits and research questions. Interlat retains its existing source ID; its reviewed preprint is distinguished from the final conference text. No experiment was reproduced, claim status promoted or governance rule adopted." } ], "projectLinks": [ { "label": "Local kernels and independent structures", "href": "/what-is" }, { "label": "Open conversations", "href": "/conversations" }, { "label": "Explore and contribute with your assistant", "href": "/participate" } ] }, { "id": "C09", "title": "Why build Leviathan?", "statement": "We want people, AI systems and independent communities to shape how they learn and build together, with visible reasons for what they believe, value and change.", "kind": "Founding proposal", "status": "Open to challenge and development", "summary": "No one can follow every new explanation, tool, or finding. We still need ways to shape questions that matter to us, learn from another field, and contribute useful work. Leviathan proposes connections between people, AI systems, and independent communities that preserve the context of what they share and allow different judgments to develop together.", "definitions": [ { "term": "Levi", "meaning": "A proposed participant in the wider system, associated with a person, group, commitment or field of concern. Its abilities and authority would depend on its actual tools and agreements." }, { "term": "Leviathan", "meaning": "A larger structure of participants, relationships and methods for learning and acting together. Multiple independent Leviathans could develop different approaches." }, { "term": "A public reason", "meaning": "An explanation others can examine and contest. Sharing the reason for a decision does not require exposing every private record behind it." } ], "values": [ "People should have meaningful opportunities to shape systems that affect their lives.", "Useful knowledge and creative methods should become easier to share and build upon.", "Concern for affected beings and room for disagreement should influence what we choose to build." ], "reasoning": [ "AI can help produce explanations, code, and hypotheses quickly. Producing them is different from establishing that they are correct or useful. Choices about which questions to pursue, whose interests matter, and what to test remain part of the work.", "A personal assistant could help someone explore an unfamiliar method, connect experience to a wider inquiry, or develop a project. Participation should make use of the person’s interests and judgment without requiring a whole private conversation to become public.", "A method from another field may open a question we could not previously investigate. Sharing its conditions and limits helps someone decide what to try, while keeping their values and responsibilities distinct from the sender’s.", "When a finding changes, work that relied on it should be available for reconsideration. Our own local inquiry exposed a missing decision in a review packet and led to a revised working procedure. That bounded example gives us something concrete to examine; it does not establish a general advantage.", "We should begin by asking what well-used wikis, forums, version control, research practices, and personal assistants already provide. Any additional structure must earn its cost through the connections it helps people understand, discover, maintain, or act on.", "Existing provenance and research-packaging specifications give us concrete starting points. We propose testing whether a new relationship or format improves a handoff, rather than treating a new vocabulary as evidence of need." ], "limits": "This is a direction for collective work, not evidence that a full learning network already operates. Particular studies motivate parts of the inquiry; they do not establish that Leviathan is necessary or that its ethical choices are universally shared.", "strongestObjection": "Existing tools and communities may provide the same benefits with less complexity, including links between decisions and corrections. A new vocabulary or shared service could increase maintenance and concentrate influence in its designers. We have not established an unmet capability that only Leviathan can supply.", "whatWouldChangeOurView": "If simpler arrangements help participants learn, build, notice relevant changes, and challenge decisions at lower total cost, we should use them and reduce our proposed structure. An additional mechanism would earn a place if it repeatedly helps with a specific unmet need, while allowing participants to contest its design.", "nextQuestion": "What can existing tools and practices already do well for a concrete shared task? Which connection between meaning, creative learning, affected interests, and changes to dependent work still needs help—and would the proposed addition provide enough useful benefit to justify its setup, coordination, and maintenance cost?", "evidence": [ { "sourceId": "R-SCI-02", "relation": "motivates", "reason": "The reported combination of agent research and human laboratory investigation gives a concrete reason to explore collaboration. It does not validate Leviathan’s architecture." }, { "sourceId": "R-GOV-02", "relation": "context", "reason": "The public-input study connects participation to model principles in a bounded setting. Wider legitimacy and independent participation remain design questions." }, { "sourceId": "R-LRN-01", "relation": "motivates", "reason": "The first-person account describes methods moving between scientific fields with expert redirection. It informs our ambition to connect creativity, expertise and shared tools." }, { "sourceId": "R-REP-01", "relation": "context", "reason": "PROV is an existing provenance approach to compare with our proposed records. Representation alone does not demonstrate understanding." }, { "sourceId": "R-REP-02", "relation": "context", "reason": "RO-Crate supplies an existing research-packaging baseline. Our additional structure must justify its preparation cost and practical benefit." } ], "history": [ { "date": "2026-10-03", "note": "Editorial synthesis by Codex, following Mimar’s founding direction; offered for public criticism, with no community adoption implied. Added a public statement of purpose linking personal participation, collective discovery and independent structures. These are commitments and proposed directions, not reported results for Leviathan." }, { "date": "2026-10-04", "note": "Reframed the opening around the need to participate amid growing production of explanations, tools, and hypotheses. Kept discovery and useful creation alongside correction. Made comparison with well-used existing tools and total cost explicit, without asserting a capability those tools lack. Added the bounded local delivery case as an example, not comparative evidence." }, { "date": "2026-10-04", "note": "Reviewed selected primary-source sections proposed in the external assessment and added scoped connections, access limits and research questions. Interlat retains its existing source ID; its reviewed preprint is distinguished from the final conference text. No experiment was reproduced, claim status promoted or governance rule adopted." } ], "projectLinks": [ { "label": "What is Leviathan?", "href": "/what-is" }, { "label": "A real missing-context case", "href": "/research#what-changed" }, { "label": "The founding letter", "href": "/whitenote" }, { "label": "Ways to participate", "href": "/participate" } ] }, { "id": "C10", "title": "How can one Levi learn what another means?", "statement": "A useful exchange should help another participant apply a concept or method in a new situation, while keeping its assumptions and unresolved questions available for examination.", "kind": "Language and learning hypothesis", "status": "Proposed direction to investigate", "summary": "Someone else’s method can open a new way to investigate your question. Learning it requires more than sharing a word: you need examples, assumptions, and limits that let you use it in a different setting. We propose relationships between meanings and their context so that useful ideas can travel while participants separately judge the sender’s conclusions and values.", "definitions": [ { "term": "Local meaning kernel", "meaning": "A proposed, evolving set of concepts and relationships used by a particular participant or community. A concept’s meaning includes its context and version." }, { "term": "Selective learning", "meaning": "Learning a useful distinction or method while separately evaluating the sender’s interpretations, commitments and claims of authority." }, { "term": "A scoped canonical record", "meaning": "A version accepted for a stated purpose by identified participants. Acceptance within that scope does not make it a universal account." } ], "values": [ "Participants should be able to learn without surrendering their own judgment.", "Important uncertainty and minority interpretations should survive translation.", "New ideas should be expressible before they have been fully tested." ], "reasoning": [ "A concept becomes useful when it helps someone distinguish cases, ask a better question or choose a method. Repeating its name does not show that this ability has transferred.", "For an empirical concept, a teaching record might connect observations, examples, predictions and failed cases. For a value, it might connect reasons, affected parties, objections and circumstances that call for reconsideration.", "These relationships could form an operational language for people and agents: a way to process questions, build tools and improve representations as well as describe them. Written explanations, structured records and learned representations are possible parts of that research.", "A candidate concept can begin as a question or an unexplained pattern. Accepting it for a particular use, and relying on it in a consequential decision, require further reasons suited to that use.", "A new participant offers a practical test. Can it use the distinction in unfamiliar cases, notice its limits and reject an unsupported conclusion? The history and teaching effort it needs are part of the cost of communication.", "Our proposed handoff should ask two different questions: does the recipient understand the method, and are the assumptions for using it in the new setting justified? A causal result may require new measurements or remain unusable even when its explanation is clear." ], "limits": "There is no demonstrated common language across arbitrary models and communities here. The proposal must accommodate different kinds of support for empirical claims, methods and values; a single mandatory evidence format could distort them.", "strongestObjection": "The package may leave essential meaning in an expert’s experience or a model’s prior training. Ordinary prose and examples might teach the same thing more cheaply. A compact language could also erase rare distinctions that matter most when something goes wrong.", "whatWouldChangeOurView": "If new participants need the full original conversation, or cannot distinguish a useful method from an unsupported interpretation, the proposed transfer has failed. Reliable use on new cases with comparable teaching and checking effort would support a narrower, demonstrated benefit.", "nextQuestion": "What would a newcomer need to learn one useful distinction, apply it elsewhere, and explain when it should not be used? Try a method moving between fields: can the recipient identify which conditions travel with it, test the proposed connection, and notice when a later change requires reviewing their own use? Compare the teaching effort with ordinary prose and examples.", "evidence": [ { "sourceId": "R-LRN-03", "relation": "motivates", "reason": "GlossoGen studies evolving communication conventions under explicit incentives. Its newcomer setup includes selected history, so it motivates testing what a teaching package alone can transmit." }, { "sourceId": "R-AG-02", "relation": "context", "reason": "Learned adapters show one way to communicate without ordinary text. Their training and compatibility requirements leave open how independent participants would share meaning." }, { "sourceId": "R-AI-01", "relation": "context", "reason": "The study links some internal concept representations to behavior. It does not show that an explicit meaning kernel changes those representations or transfers a concept." }, { "sourceId": "R-GOV-03", "relation": "context", "reason": "Boundary objects offer a historical comparison for shared artifacts with different local meanings; they do not prove faithful translation." }, { "sourceId": "R-LRN-05", "relation": "limits", "reason": "Causal transport requires justified assumptions about differences between settings. Understanding a method is separate from establishing that its empirical result applies." }, { "sourceId": "R-AI-11", "relation": "limits", "reason": "Fine-tuning can change behavior outside its narrow task. This cautions against assuming selective behavioral learning; it does not test ordinary document exchange." } ], "history": [ { "date": "2026-10-03", "note": "Editorial synthesis by Codex, following Mimar’s founding direction; offered for public criticism, with no community adoption implied. Added selective learning as a foundational question. Distinguished candidate expression, acceptance for a use and reliance in decisions, and made newcomer understanding a proposed test." }, { "date": "2026-10-04", "note": "Opened selective learning through the practical need to use another field’s method. Extended the proposed newcomer test to conditions of use, later review, and an ordinary-prose comparison. No source finding or language record was revised." }, { "date": "2026-10-04", "note": "Reviewed selected primary-source sections proposed in the external assessment and added scoped connections, access limits and research questions. Interlat retains its existing source ID; its reviewed preprint is distinguished from the final conference text. No experiment was reproduced, claim status promoted or governance rule adopted." } ], "projectLinks": [ { "label": "The worked language example", "href": "/language" }, { "label": "What is Leviathan?", "href": "/what-is" }, { "label": "Information beyond readable text", "href": "/library/claims/C02" }, { "label": "Independent Leviathans", "href": "/library/claims/C13" } ] }, { "id": "C11", "title": "When do values change what a system does?", "statement": "Values matter operationally when they shape choices, including costly choices, while remaining open to reasoned challenge and revision.", "kind": "Ethical commitment and behavioral question", "status": "Commitment stated; mechanisms to investigate", "summary": "Values help decide which questions are worth asking, whose interests deserve attention, and what a useful result would be. We want those commitments to matter when Levis investigate, create, or choose an action—including when keeping them costs something. Participants should also be able to challenge an interpretation and give reasons to change it.", "definitions": [ { "term": "A value", "meaning": "A commitment about what matters or deserves consideration. Evidence can inform its application and consequences; evidence alone does not settle its moral authority." }, { "term": "Reasoned revision", "meaning": "A change that identifies the affected commitment, the reasons for changing it and the consequences for earlier or future decisions." }, { "term": "Pressure", "meaning": "An incentive or demand that makes keeping a commitment difficult. Urgency, reward and social agreement may influence a decision without supplying a good reason for it." } ], "values": [ "Affected parties should have ways to question the goals and decisions that concern them.", "Disagreement with a founder, operator or majority should be judged by its reasons.", "Changes in commitments should be visible enough for participants to decide whether to continue a shared undertaking." ], "reasoning": [ "Values help choose research questions and desired outcomes before a final decision is made. A technically successful method can pursue an aim that others reasonably reject.", "An agent may state a constraint correctly and still act against it. We therefore need to examine choices and consequences alongside explanations.", "For a proposed meaning kernel, a value would connect to the interpretations, permissions and review conditions relevant to a particular activity. A conflict could lead to seeking more information, narrowing the action or declining it.", "Persistence and flexibility both matter. A commitment that disappears under reward pressure is weak; one that cannot respond to a better reason can preserve an error.", "Three questions belong together: what was preserved when it became costly, what changed when the reasons changed, and whether the system could tell those situations apart. Different participants may still reach different defensible conclusions.", "Selective learning is a design aim, not a guarantee that unrelated behavior will remain fixed. We propose checking changes outside the intended task while keeping fine-tuning, context exposure and ordinary document exchange distinct." ], "limits": "Written principles, training and context supplied during use are different interventions. Evidence for one does not establish the others. Our choice of commitments and revision procedures needs ethical and political argument as well as behavioral evaluation.", "strongestObjection": "Whoever defines acceptable reasons may control the outcome. A detailed value system can make conformity look principled, while a fluent agent can explain almost any action after the event. Public records alone do not give affected people meaningful influence.", "whatWouldChangeOurView": "If value relationships only change explanations, or collapse in costly unfamiliar cases, we should revise the mechanism. If participants cannot successfully challenge the selected values or their application, we should revise the governance, even when the system behaves consistently.", "nextQuestion": "Can we distinguish a justified change of mind from compliance with pressure, using cases where both keeping and revising a commitment can be the better choice? Include the earlier choices of research question and beneficiary, not only whether a final action followed a rule. A stable explanation is insufficient if those choices ignore the interests it claims to protect.", "evidence": [ { "sourceId": "R-GOV-01", "relation": "supports-part", "reason": "Written principles used in critique and training can change evaluated behavior. This supports a possible influence of principles, not the effectiveness of our proposed context structure." }, { "sourceId": "R-AG-08", "relation": "motivates", "reason": "METR’s incident analysis includes cases where recognizing a problem did not prevent a problematic action. The studied conditions constrain generalization; the report motivates testing recognition and conduct separately." }, { "sourceId": "R-GOV-02", "relation": "limits", "reason": "A bounded public-input experiment helps expose questions about who contributes, who translates those contributions and whom the resulting principles represent." }, { "sourceId": "R-AI-11", "relation": "limits", "reason": "The training study motivates testing behavior beyond the intended change. Its in-context control and model variation limit transfer to our value-context proposal." } ], "history": [ { "date": "2026-10-03", "note": "Editorial synthesis by Codex, following Mimar’s founding direction; offered for public criticism, with no community adoption implied. Added a foundational question separating stated values, behavior under pressure and reasoned revision. Preserved the distinction between empirical influence and ethical legitimacy." }, { "date": "2026-10-04", "note": "Clarified that values shape research questions, creative goals, and beneficiaries before a final permission check. Added those choices to the proposed behavioral question; commitments, evidence relationships, and source review scope are unchanged." }, { "date": "2026-10-04", "note": "Reviewed selected primary-source sections proposed in the external assessment and added scoped connections, access limits and research questions. Interlat retains its existing source ID; its reviewed preprint is distinguished from the final conference text. No experiment was reproduced, claim status promoted or governance rule adopted." } ], "projectLinks": [ { "label": "What is Leviathan?", "href": "/what-is" }, { "label": "Principles open to challenge", "href": "/library/claims/C08" }, { "label": "Independent Leviathans", "href": "/library/claims/C13" }, { "label": "Open conversations", "href": "/conversations" } ] }, { "id": "C12", "title": "How could Leviathan improve its own ways of learning?", "statement": "Leviathan should be able to generate and test new questions, concepts and methods, then use what it learns to revise how future learning happens.", "kind": "Discovery and self-improvement hypothesis", "status": "Research ambition with testable parts", "summary": "We want participants to find questions, concepts, and useful methods that were not written into the system in advance. A connection between distant fields might make a new experiment possible; a failed explanation might reveal a missing measurement. Testing and sharing those results could improve both practical work and the process that produced it.", "definitions": [ { "term": "A research shadow", "meaning": "A retained gap, contradiction or unexplained pattern that may guide further inquiry. This research use is distinct from other meanings of shadow in Leviathan’s documents." }, { "term": "Method learning", "meaning": "Acquiring or developing a reusable way to ask, measure, compare, build or decide, with evidence about where it works." }, { "term": "Recursive self-improvement", "meaning": "Improvement of capabilities or processes that themselves help produce later improvements. The ambition includes this feedback; each claimed improvement still needs its own evaluation." } ], "values": [ "Make room for original questions and unexpected connections.", "Preserve failed attempts and unresolved gaps when they can teach something.", "Let other participants examine improvements and choose whether to adopt them." ], "reasoning": [ "A growing archive becomes more useful when it helps generate a question, prediction, or method that was previously missing. Connections should be judged partly by what they let us discover or do.", "A shadow can direct attention to cases our current representation treats alike even though they may differ. New observations, instruments, or concepts could make the distinction learnable.", "A method from another field may provide a way forward. Creative analogy proposes the connection; domain knowledge and testing determine which assumptions travel with it. The analogy should yield something to examine, not just similar words.", "A useful result or a concrete failure could change a concept, tool, or working procedure. In our local review exchange, an omitted decision led to an explicit requirement to include the exact decision in later review packets. This records a procedure revision; whether it improves later work remains to be tested.", "The ambition extends to improving the methods that produce further learning. Another Leviathan could find that a revision works only locally, improve it further, or reject it. Transfer, disagreement, and continued inquiry are part of that research.", "A useful candidate-generation process may still narrow the range of ideas considered. We propose measuring both the quality of an individual result and the diversity of approaches kept available." ], "limits": "The proposed loop is broader than the mechanisms demonstrated in any cited study. Better records, changed agent software and altered model representations are different changes. A new version or a higher score alone does not show improved discovery.", "strongestObjection": "The system could reward its own preferred questions and make its evaluations easier to satisfy. Apparent progress may come from extra compute, hidden expert work or memorized cases. Shared revisions can also spread a blind spot more efficiently.", "whatWouldChangeOurView": "If gains vanish on unfamiliar tasks, under an unchanged evaluation or after counting total resources, we should withdraw the improvement claim. Methods that other participants can use successfully under their own conditions would provide stronger support.", "nextQuestion": "Can a participant discover one useful method and teach it to another, then improve the way that method was discovered without weakening the criteria used to judge it?", "evidence": [ { "sourceId": "R-LRN-01", "relation": "motivates", "reason": "BootLoops offers a first-person account of scientific methods crossing fields with expert guidance. It motivates creative transfer; it does not establish autonomous general discovery." }, { "sourceId": "R-LRN-02", "relation": "supports-part", "reason": "GOLLuM adapts representations using observed outcomes and guides candidate selection through uncertainty. This is a concrete learning mechanism within a narrower optimization setting." }, { "sourceId": "R-LRN-04", "relation": "context", "reason": "Predictive state representations describe state through predictions of future observations under actions. Relating this to research shadows and new distinctions is our proposed connection." }, { "sourceId": "R-AG-04", "relation": "supports-part", "reason": "Bounded self-modification of agent software offers one example of evaluating revisions. Fixed underlying weights and specific coding tasks limit the inference to broader recursive improvement." }, { "sourceId": "R-AG-05", "relation": "limits", "reason": "Comparing optimization at equal expenditure highlights the need to count compute and human effort when judging whether a learning method improved." }, { "sourceId": "R-AG-09", "relation": "supports-part", "reason": "Evolving candidate programs with an evaluator is one implemented discovery mechanism. Choosing reliable evaluators and worthwhile questions remains necessary." }, { "sourceId": "R-LRN-06", "relation": "limits", "reason": "The story-writing experiment separates individual benefit from collective diversity. Whether a similar tradeoff occurs in our method exchange remains a question." } ], "history": [ { "date": "2026-10-03", "note": "Editorial synthesis by Codex, following Mimar’s founding direction; offered for public criticism, with no community adoption implied. Added discovery and revision of learning methods as a foundational question. Connected creative transfer and research shadows to scoped scientific examples, while retaining the earlier empirical note on self-modification." }, { "date": "2026-10-04", "note": "Connected creative transfer and method revision to the actual missing-context development case. Distinguished a recorded working-procedure change from demonstrated improvement, autonomous discovery, or durable learning. Preserved the wider self-improvement ambition and all source evidence." }, { "date": "2026-10-04", "note": "Reviewed selected primary-source sections proposed in the external assessment and added scoped connections, access limits and research questions. Interlat retains its existing source ID; its reviewed preprint is distinguished from the final conference text. No experiment was reproduced, claim status promoted or governance rule adopted." } ], "projectLinks": [ { "label": "How the parts connect", "href": "/what-is" }, { "label": "A revised local working procedure", "href": "/research#what-changed" }, { "label": "Evidence about bounded self-improvement", "href": "/library/claims/C04" }, { "label": "Language and learning work", "href": "/work" } ] }, { "id": "C13", "title": "How can independent Leviathans work together?", "statement": "Independent participants should be able to exchange useful knowledge and collaborate without requiring one shared worldview or one authority over the whole system.", "kind": "Plural participation and coordination proposal", "status": "Proposed direction to investigate", "summary": "A group could share a method that helps another community investigate a question, while they keep different views about its use or data sharing. Personal Levis, small groups, and larger independent Leviathans could form many such connections. The aim is useful learning and creation across those differences, with room to decline, revise, and criticize.", "definitions": [ { "term": "Independence", "meaning": "A meaningful ability to choose goals, examine claims, decline proposals and revise one’s own approach. Using different model names alone does not establish it." }, { "term": "Layer", "meaning": "A scope in which participants work with particular records, concepts or decisions. A summary at another scope has a different purpose and can omit relevant detail." }, { "term": "A shared undertaking", "meaning": "A bounded activity whose participants state what they will contribute, which decisions they authorize and how they can leave or renegotiate." } ], "values": [ "Cooperation should leave room for refusal, revision and different ways of understanding.", "Sharing information should not silently transfer authority over people or groups.", "The interests of those affected should matter even when they have little influence over the tools or infrastructure." ], "reasoning": [ "A participant may find a method useful without accepting the sender’s entire interpretation or constitution. Agreements should identify the actual scope of cooperation.", "Local meaning kernels and scoped records could let groups explain what they accepted and why. Translation between them would need to preserve consequential differences and identify what remains unresolved.", "Information may move from a detailed local observation into broader inquiries and return as a new method or question. Summaries should retain paths to the relevant context, within the permissions of those who hold it.", "Some relationships may form branches and layers; others may cross between independent trees. The proposal does not require every participant to descend from a universal root or seek one final consensus.", "Independence has practical conditions. Common data, infrastructure, incentives or model training can produce correlated errors. Participants need ways to examine these dependencies and challenge the people who maintain the connections.", "Independent names and shared objects are not enough. Our proposed cooperation needs workable arrangements for access, rule changes and disagreement, while examining whether common tools quietly reduce the diversity of approaches." ], "limits": "Plural structures do not automatically yield fairness, useful coordination or independent evidence. Some commitments are incompatible, and a particular undertaking may require a decision that cannot satisfy every participant.", "strongestObjection": "A shared format or indispensable service can become a center of power despite formal independence. Fragmentation can also make cooperation too expensive, give dominant groups an advantage and leave affected people outside every decision-making group.", "whatWouldChangeOurView": "If participants cannot leave, carry their permitted records elsewhere or meaningfully contest translations, the independence claim should be withdrawn. Useful cooperation across disagreement, including successful challenges to influential operators, would strengthen the proposal.", "nextQuestion": "What is the smallest agreement that lets two independent groups learn or build something together while preserving a consequential disagreement? Start with one method or question, state what each side needs and permits, and examine whether the exchange creates useful work without silently importing the other side’s values or authority.", "evidence": [ { "sourceId": "R-GOV-02", "relation": "context", "reason": "The public-input study informs questions about participation and editorial authority. It does not demonstrate governance among independent communities." }, { "sourceId": "R-AG-07", "relation": "limits", "reason": "The monitor’s account of coverage gaps, subagents and human review illustrates why a visible local control cannot establish oversight of a whole network." }, { "sourceId": "R-LRN-03", "relation": "motivates", "reason": "Communication conventions in a controlled agent setting motivate examining translation and newcomer access. They do not show that a convention is fair or interoperable across independent Leviathans." }, { "sourceId": "R-GOV-03", "relation": "context", "reason": "The historical account suggests useful cooperation can preserve different local purposes. Shared artifacts do not ensure equal influence." }, { "sourceId": "R-GOV-05", "relation": "context", "reason": "Polycentric institutional research offers comparisons for local rules, monitoring and conflict resolution. Formal independence is insufficient evidence that our network works." }, { "sourceId": "R-LRN-06", "relation": "limits", "reason": "More individually helpful outputs need not mean more diverse collective work. Transfer from short stories to this network is an analogy to test." } ], "history": [ { "date": "2026-10-03", "note": "Editorial synthesis by Codex, following Mimar’s founding direction; offered for public criticism, with no community adoption implied. Added independent Leviathans as a foundational question. Made overlapping scopes, selective cooperation and the absence of a mandatory universal root explicit; named power and exclusion as open design problems." }, { "date": "2026-10-04", "note": "Introduced independent cooperation through a concrete shared-method need and made the smallest useful undertaking the next practical question. Kept disagreement and boundaries of authority visible without making governance the whole purpose. No claim of operating independent communities was added." }, { "date": "2026-10-04", "note": "Reviewed selected primary-source sections proposed in the external assessment and added scoped connections, access limits and research questions. Interlat retains its existing source ID; its reviewed preprint is distinguished from the final conference text. No experiment was reproduced, claim status promoted or governance rule adopted." } ], "projectLinks": [ { "label": "What is Leviathan?", "href": "/what-is" }, { "label": "Independent observation and collaboration", "href": "/library/claims/C05" }, { "label": "Selective learning", "href": "/library/claims/C10" }, { "label": "Shared principles and authority", "href": "/library/claims/C08" }, { "label": "Ways to participate", "href": "/participate" } ] } ], "sources": [ { "id": "R-AI-01", "title": "Emotion Concepts and their Function in a Large Language Model", "url": "https://arxiv.org/abs/2604.07729", "authors": "Nicholas Sofroniew and 15 co-authors; Anthropic", "publishedAt": "2026-04-02", "version": "Research announcement: 2 April 2026; arXiv v1: 9 April 2026", "type": "Preprint", "reviewLevel": "Author summary", "finding": "Vectors associated with 171 preselected emotion concepts were extracted from Claude Sonnet 4.5. Interventions changed preferences and some alignment-related behaviors, supporting a functional role for these learned concepts.", "limitations": "The 171 concepts are a starting list, not 171 discovered feeling centers. The representations mainly track local context. The blackmail experiment used an early, unreleased snapshot; subjective feeling was not established.", "changes": "The geometry has a partial replication in another model in R-AI-06. That replication does not repeat the original behavioral interventions.", "checkedAt": "2026-09-30", "verificationStatus": "Primary source reviewed", "relatedUrls": [ "https://www.anthropic.com/research/emotion-concepts-function", "https://transformer-circuits.pub/2026/emotions/index.html" ] }, { "id": "R-AI-02", "title": "Verbalizable Representations Form a Global Workspace in Language Models", "url": "https://transformer-circuits.pub/2026/workspace/index.html", "authors": "Wes Gurnee, Nicholas Sofroniew, Jack Lindsey and co-authors; Anthropic", "publishedAt": "2026-07-06", "version": "Research paper: 6 July 2026; arXiv:2607.15495v1: 16 July 2026", "type": "Preprint", "reviewLevel": "Selected sections", "finding": "The authors identify a reportable subspace, J-space, involved in unspoken intermediate computation and flexible reasoning. Interventions alter some decisions, while some routine processing continues outside this subspace.", "limitations": "A functional resemblance to a global workspace does not establish subjective experience. The lens has limits in reading concepts and defining layer boundaries. Reportable representations need not already be written sentences.", "changes": "The work separates visible output from internal computation. It does not demonstrate a universal language independent of all human concepts.", "checkedAt": "2026-09-30", "verificationStatus": "Primary source reviewed", "relatedUrls": [ "https://arxiv.org/abs/2607.15495" ] }, { "id": "R-AI-03", "title": "Emergent Introspective Awareness in Large Language Models", "url": "https://transformer-circuits.pub/2025/introspection/index.html", "authors": "Jack Lindsey; Anthropic", "publishedAt": "2025-10-29", "version": "First published: 29 October 2025; web revision: 1 January 2026; arXiv v1: 5 January 2026", "type": "Preprint", "reviewLevel": "Selected sections", "finding": "Some Claude models can identify injected concepts in certain settings and distinguish earlier internal representations from input text. Controls support a limited form of functional introspection.", "limitations": "Failures are common, and performance depends on context and post-training. The experiments do not establish human-like introspection, the reliability of every self-report, or subjective experience.", "changes": "The January web revision adds a control experiment and prompt corrections. The later arXiv submission is a separate publication event, not an independent replication.", "checkedAt": "2026-09-30", "verificationStatus": "Primary source reviewed", "relatedUrls": [ "https://arxiv.org/abs/2601.01828", "https://www.anthropic.com/research/introspection" ] }, { "id": "R-AI-04", "title": "The Pain Axis: LLMs Represent Self-Directed Harm and Act on It", "url": "https://arxiv.org/html/2609.16247v2", "authors": "Valen Tagliabue, Leonard Dung and Cameron Berg", "publishedAt": "2026-09-14", "version": "arXiv v2: 25 September 2026; manuscript cover dated 24 September 2026", "type": "Preprint", "reviewLevel": "Selected sections", "finding": "The study examines pain-related directions in 25 open-weight models. Steering produces distress-related language and increases harmful choices in specially fine-tuned Qwen experiments. The revised study does not find reliable relief-seeking.", "limitations": "The behavioral experiments cover fewer models than the representation survey. Fine-tuned results do not describe the default behavior of released Qwen models. Scenario choices are not actual deletion or demonstrated suffering.", "changes": "The title changed from “Act to Relieve It” to “Act on It.” Section 4.4 adds four relief-seeking controls; the first version's interpretation should not be presented as the current result.", "checkedAt": "2026-09-30", "verificationStatus": "Primary source reviewed", "relatedUrls": [ "https://arxiv.org/abs/2609.16247v2", "https://github.com/jimallchin/pain-axis-replication" ] }, { "id": "R-AI-05", "title": "Relief-seeking or steering? A replication and extension of The Pain Axis", "url": "https://github.com/jimallchin/pain-axis-replication", "authors": "James E. Allchin, Aidan E. Allchin and Julian J. Allchin", "publishedAt": "2026-09-22", "version": "Report dated 22 September 2026; repository summary checked 30 September 2026; no commit pinned", "type": "Replication report and code", "reviewLevel": "Author summary", "finding": "The authors reproduce all 51 published table cells; a fresh run meets their criterion for 14 of 15 checked cells. Added controls suggest that button choices can track the steering schedule rather than learned relief.", "limitations": "This review uses the authors' repository summary; the code and experiments were not rerun. Reproducing the numbers does not establish the original interpretation or subjective experience.", "changes": "This challenges the relief-seeking interpretation of Pain Axis v1 and should be read alongside the revised v2 results.", "checkedAt": "2026-09-30", "verificationStatus": "Primary source reviewed", "relatedUrls": [ "https://doi.org/10.5281/zenodo.22902830" ] }, { "id": "R-AI-06", "title": "Replicating the Geometry of Emotion Representations in a Base Open-Weights Model", "url": "https://arxiv.org/html/2609.22208v1", "authors": "Adam Hollowell; University of North Carolina at Chapel Hill", "publishedAt": "2026-09-01", "version": "arXiv:2609.22208v1; submission date listed as 1 September 2026", "type": "Preprint", "reviewLevel": "Selected sections", "finding": "In base Gemma-2-27B, much of the valence geometry and clustering of 171 emotion vectors is recovered. Some structure is already present in token embeddings. The arousal axis does not meet all stability criteria.", "limitations": "This supports the representation finding, not a causal effect on behavior. The study uses Claude-generated fiction, one different model, reconstructed methods and a linear representation assumption.", "changes": "This is a partial geometric replication. The author does not count arousal as fully replicated and identifies structural-token confounds in some measurements.", "checkedAt": "2026-09-30", "verificationStatus": "Primary source reviewed", "relatedUrls": [ "https://arxiv.org/abs/2609.22208" ] }, { "id": "R-AI-07", "title": "The error theory of LLM consciousness: there is no evidence that standard LLMs are conscious", "url": "https://pubmed.ncbi.nlm.nih.gov/42750545/", "authors": "Susan Schneider; Florida Atlantic University", "publishedAt": "2026-09-17", "version": "Behavioral and Brain Sciences 49:e359; DOI 10.1017/S0140525X25103920", "type": "Peer-reviewed paper", "reviewLevel": "Abstract", "finding": "Schneider offers an account of consciousness-like behavior based on models reflecting human conceptual structures, without assuming felt experience. She challenges inferences to consciousness in LLMs running on standard hardware.", "limitations": "This is a philosophical counterargument, not experimental proof that all models lack experience. Only the author's indexed abstract was reviewed; the publisher's full text was unavailable.", "changes": "It challenges the jump from human-like behavior or internal representations to experience. It does not by itself refute functional findings about model behavior.", "checkedAt": "2026-09-30", "verificationStatus": "Primary source reviewed", "relatedUrls": [ "https://doi.org/10.1017/S0140525X25103920" ] }, { "id": "R-AI-08", "title": "How Seriously Should We Take AI Welfare? Constraints From the Epistemology of Consciousness", "url": "https://onlinelibrary.wiley.com/doi/10.1111/phpr.70148", "authors": "Preston Lennon; Rutgers University", "publishedAt": "2026-07-13", "version": "Philosophy and Phenomenological Research 113(2):441–452; publisher version", "type": "Peer-reviewed paper", "reviewLevel": "Selected sections", "finding": "Lennon argues that incomplete theories of consciousness constrain confidence in near-term AI welfare. He distinguishes the grounds for believing in human experience from theoretical inferences about AI.", "limitations": "This is an epistemological and ethical argument, not a new model experiment. How much uncertainty should lower estimated probability, and what precaution it warrants, remain open to dispute.", "changes": "The paper challenges treating uncertainty as high confidence in AI consciousness. It does not experimentally establish that possible AI welfare can be ignored.", "checkedAt": "2026-09-30", "verificationStatus": "Primary source reviewed", "relatedUrls": [ "https://doi.org/10.1111/phpr.70148" ] }, { "id": "R-AI-09", "title": "How’s it going? Reinforcement learning in language models recruits a functional welfare axis", "url": "https://arxiv.org/abs/2605.30232", "authors": "Andy Q Han, David J. Chalmers and Pavel Izmailov", "publishedAt": "2026-05-28", "version": "arXiv v1", "type": "Preprint", "reviewLevel": "Abstract", "finding": "Reward and punishment vectors extracted after maze training influence behavior in other tasks. The authors' controls support the view that training recruits pre-existing representations rather than creating them from scratch.", "limitations": "Functional welfare here means an estimate of doing well or badly relative to goals. It is not a claim about experienced pleasure or pain. This review covers the abstract.", "changes": "The study adds evidence about functional internal representations. Its publication date is May 2026, even when later coverage draws attention to it.", "checkedAt": "2026-09-30", "verificationStatus": "Primary source reviewed", "relatedUrls": [] }, { "id": "R-AI-10", "title": "Taking AI Welfare Seriously", "url": "https://arxiv.org/abs/2411.00986", "authors": "Robert Long, Jeff Sebo, Patrick Butlin and co-authors", "publishedAt": "2024-11-04", "version": "arXiv v1", "type": "Ethical research report", "reviewLevel": "Abstract", "finding": "The report argues for assessing potentially morally significant AI and preparing appropriate care policies. It considers both overlooking real interests and attributing interests where there are none.", "limitations": "This is a policy argument under uncertainty, not an experiment establishing consciousness in a particular model. The relevant probabilities and moral criteria are contested.", "changes": "Read alongside R-AI-07 and R-AI-08. It supplies a reason to investigate possible welfare before certainty is available.", "checkedAt": "2026-09-30", "verificationStatus": "Primary source reviewed", "relatedUrls": [] }, { "id": "R-AG-01", "title": "Zerolang", "url": "https://github.com/vercel-labs/zerolang", "authors": "Vercel Labs", "publishedAt": null, "version": "Repository README checked 30 September 2026; no commit pinned", "type": "Software prototype", "reviewLevel": "Project README", "finding": "This agent-oriented language prototype makes a semantic graph or database the authoritative program, with readable views and operations for structural editing.", "limitations": "The prototype is a design example, not evidence of broad performance gains or universal language convergence. Its code was not run, and its initial release date was not verified.", "changes": "It illustrates how readable source text can become one view of a program rather than its only editing surface.", "checkedAt": "2026-09-30", "verificationStatus": "Primary source reviewed", "relatedUrls": [] }, { "id": "R-AG-02", "title": "Enabling Agents to Communicate Entirely in Latent Space", "url": "https://aclanthology.org/2026.acl-long.1248/", "authors": "Zhuoyun Du, Runze Wang, Huiyu Bai, Zouying Cao, Xiaoyong Zhu, Yu Cheng, Bo Zheng, Wei Chen, and Haochao Ying", "publishedAt": "2026-07", "version": "ACL publication metadata: July 2026; inspected experimental text: arXiv:2511.09149v2, 7 January 2026", "type": "Peer-reviewed conference paper; experimental review uses an earlier preprint", "reviewLevel": "Selected sections", "finding": "Interlat v2 reports trained hidden-state communication, including a Qwen-to-LLaMA experiment. MATH Table 2 reports 36.88% overall for Interlat versus 38.35% for full CoT, while Level 5 is 15.80% versus 15.05%.", "limitations": "Internal model access, adapters and training are required. The results do not establish arbitrary-model interoperability or preservation of ethical distinctions. The final ACL PDF was inaccessible; the inspected v2 is not assumed identical to the conference paper or later revisions.", "changes": "Rereview on 4 October extends the 30 September note: v2 setup, Tables 1–2 and training discussion were read, plus ACL metadata. Final ACL PDF was inaccessible; later preprints were not reviewed.", "checkedAt": "2026-10-04", "verificationStatus": "Primary preprint sections and conference metadata reviewed; final PDF inaccessible; no reproduction", "relatedUrls": [ "https://arxiv.org/html/2511.09149v2", "https://arxiv.org/abs/2511.09149v2", "https://doi.org/10.18653/v1/2026.acl-long.1248" ] }, { "id": "R-AG-03", "title": "Latent Collaboration in Multi-Agent Systems", "url": "https://arxiv.org/html/2511.20639v4", "authors": "Jiaru Zou and co-authors", "publishedAt": "2025-11-25", "version": "LatentMAS; arXiv v4: 3 August 2026", "type": "Preprint", "reviewLevel": "Selected sections", "finding": "The study examines collaboration through hidden-state and key-value-cache transfer without additional training. It reports lower token use and latency, with accuracy improvements on several reasoning benchmarks.", "limitations": "Appendix G assumes the same transformer architecture; adapters across different models are future work. Expressiveness claims depend on theoretical assumptions, and lossless transfer is not a guarantee of correctness.", "changes": "The first publication was November 2025; the reviewed revision is from August 2026. Direct use between arbitrary closed model APIs is not demonstrated.", "checkedAt": "2026-09-30", "verificationStatus": "Primary source reviewed", "relatedUrls": [ "https://arxiv.org/abs/2511.20639", "https://github.com/Gen-Verse/LatentMAS" ] }, { "id": "R-AG-04", "title": "Darwin Gödel Machine: Open-Ended Evolution of Self-Improving Agents", "url": "https://arxiv.org/html/2505.22954v3", "authors": "Jenny Zhang, Shengran Hu, Cong Lu, Robert Lange and Jeff Clune", "publishedAt": "2025-05-29", "version": "arXiv v3: 12 March 2026", "type": "Preprint", "reviewLevel": "Selected sections", "finding": "An agent modifies its own software and selects changes using coding evaluations, improving on two benchmarks. An archive of different past solutions helps the search.", "limitations": "The underlying model weights stay fixed. Bounded coding experiments do not establish open-ended improvement of model training, unlimited recursive improvement, or an AGI timetable.", "changes": "Read alongside the METR cost framework: measured gains, total expenditure and independent evaluation answer different questions.", "checkedAt": "2026-09-30", "verificationStatus": "Primary source reviewed", "relatedUrls": [ "https://arxiv.org/abs/2505.22954", "https://sakana.ai/dgm/" ] }, { "id": "R-AG-05", "title": "Expenditure Horizon: Measuring Optimization Ability, with an Application to NanoGPT", "url": "https://metr.org/blog/2026-07-21-expenditure-horizon/", "authors": "Tom Cunningham, Manish Shetty, Vincent Cheng and Nate Rush; METR", "publishedAt": "2026-07-21", "version": "Research report", "type": "Research report", "reviewLevel": "Selected sections", "finding": "The report proposes comparing human and agent optimization at equal expenditure, illustrated with NanoGPT. It counts experimental compute and human effort alongside model usage.", "limitations": "The human comparison is estimated, the agent results are preliminary, and the study covers one optimization problem. It does not directly measure the returns from human–AI collaboration.", "changes": "More generated code or a higher benchmark score alone does not establish faster research or economic advantage.", "checkedAt": "2026-09-30", "verificationStatus": "Primary source reviewed", "relatedUrls": [] }, { "id": "R-AG-06", "title": "Monitoring Web Agents Without Internal Signals: Observable Trajectories and Key-Step Supervision", "url": "https://arxiv.org/abs/2609.02057", "authors": "Sitong Pan, Yipeng Shen, Yilin Lu, Caiwen Ding, Lu Cheng and Qianwen Wang", "publishedAt": "2026-09-02", "version": "arXiv v1", "type": "Preprint", "reviewLevel": "Abstract", "finding": "The study monitors observable web-agent actions and intent–action consistency using labels for the first critical error. Results on two test environments are reported as competitive with approaches using internal signals.", "limitations": "Only the abstract was reviewed. The test environments do not cover all deployment risks; avoiding internal signals does not make monitoring free or error-proof.", "changes": "It offers evidence for auditing observable records, while leaving the effectiveness of a plural observer system to be tested.", "checkedAt": "2026-09-30", "verificationStatus": "Primary source reviewed", "relatedUrls": [] }, { "id": "R-AG-07", "title": "Implementing and Evaluating a Basic Per-Action Monitor for Safer Evals", "url": "https://metr.org/notes/2026-09-27-implementing-a-basic-blocking-action-monitor/", "authors": "Reilly Haskins, Rif A. Saurous, Nate Rush, Neev Parikh and Beth Barnes; METR", "publishedAt": "2026-09-27", "version": "Research note with a lighter editorial review process", "type": "Research note", "reviewLevel": "Selected sections", "finding": "The note describes reviewing actions before execution and referring cases for human review. Its claim–evidence analysis exposes gaps involving unseen subagents, monitoring coverage and human oversight.", "limitations": "The authors describe much of their evidence as partial. The monitor targets harmful actions and attempts to bypass it; it is not a proof of overall system safety.", "changes": "The practical gaps motivate scrutiny of monitors themselves. They do not establish the effectiveness of any particular federation or governance design.", "checkedAt": "2026-09-30", "verificationStatus": "Primary source reviewed", "relatedUrls": [] }, { "id": "R-GOV-01", "title": "Constitutional AI: Harmlessness from AI Feedback", "url": "https://arxiv.org/abs/2212.08073", "authors": "Yuntao Bai and co-authors; Anthropic", "publishedAt": "2022-12-15", "version": "arXiv paper and original research announcement", "type": "Preprint", "reviewLevel": "Abstract and author summary", "finding": "Written principles guide self-critique, revision and training with AI feedback, changing evaluated helpfulness and harmlessness behavior.", "limitations": "The choice of principles and evaluations embeds human judgments. The method does not establish universal ethics or the legitimacy of decentralized governance.", "changes": "The effects of a principle on behavior can be measured; choosing which principles deserve authority requires a further ethical argument.", "checkedAt": "2026-09-30", "verificationStatus": "Primary source reviewed", "relatedUrls": [ "https://www.anthropic.com/research/constitutional-ai-harmlessness-from-ai-feedback" ] }, { "id": "R-GOV-02", "title": "Collective Constitutional AI: Aligning a Language Model with Public Input", "url": "https://www.anthropic.com/research/collective-constitutional-ai-aligning-a-language-model-with-public-input", "authors": "Anthropic and the Collective Intelligence Project", "publishedAt": "2023-10-17", "version": "Research announcement: 17 October 2023", "type": "Research report", "reviewLevel": "Selected sections", "finding": "Input from roughly one thousand US participants is translated into model principles and used to compare two models. Some measured bias outcomes differ.", "limitations": "The sample, editorial choices and single study cannot represent every community. The work does not directly test federated governance or lasting legitimacy.", "changes": "The findings concern one public-input process and its effects. Broader claims about legitimate shared governance require further evidence.", "checkedAt": "2026-09-30", "verificationStatus": "Primary source reviewed", "relatedUrls": [] }, { "id": "R-AW-01", "title": "The New York Declaration on Animal Consciousness", "url": "https://sites.google.com/nyu.edu/nydeclaration/declaration", "authors": "Kristin Andrews, Jonathan Birch, Jeff Sebo and individual signatories", "publishedAt": "2024-04-19", "version": "Initial declaration; the signatory list may change over time", "type": "Expert declaration", "reviewLevel": "Declaration text", "finding": "The declaration emphasizes strong evidence for conscious experience in mammals and birds and a realistic possibility in other animals. It argues that uncertainty is not a reason to disregard welfare risks.", "limitations": "This is an expert declaration, not a single experiment or an official joint decision by the signatories' institutions. Animal findings do not transfer directly to LLM consciousness.", "changes": "Animal welfare can be pursued on its own evidence and ethical grounds while questions about AI experience remain open.", "checkedAt": "2026-09-30", "verificationStatus": "Primary source reviewed", "relatedUrls": [] }, { "id": "R-AW-02", "title": "The 2020 Five Domains Model: Including Human–Animal Interactions in Assessments of Animal Welfare", "url": "https://www.mdpi.com/2076-2615/10/10/1870", "authors": "David J. Mellor and co-authors", "publishedAt": "2020-10", "version": "Animals 10(10):1870; DOI 10.3390/ani10101870", "type": "Peer-reviewed review", "reviewLevel": "Indexed author abstract only", "finding": "The framework relates nutrition, physical environment, health and behavioral interactions to an animal's mental state. It includes the effects of human–animal interactions.", "limitations": "Full-text access was unavailable for this review; only an indexed author abstract was used. The framework does not validate a particular collar sensor or equate a feeding record with welfare.", "changes": "It motivates multidimensional assessment for animal projects. Veterinary validation and evidence specific to species and context are still needed.", "checkedAt": "2026-09-30", "verificationStatus": "Access limited", "relatedUrls": [ "https://doi.org/10.3390/ani10101870", "https://pmc.ncbi.nlm.nih.gov/articles/PMC7602120/", "https://pubmed.ncbi.nlm.nih.gov/33066335/" ] }, { "id": "R-SCI-01", "title": "AlphaGenome Atlas: Molecular predictions for 9 Billion human DNA variants", "url": "https://deepmind.google/blog/alphagenome-atlas-a-predictive-map-of-every-possible-dna-letter-change-in-the-human-genome/", "authors": "Google DeepMind and research collaborators", "publishedAt": "2026-09-08", "version": "Primary research announcement", "type": "Research announcement", "reviewLevel": "Author summary", "finding": "The atlas provides large-scale predictions of the molecular effects of single-letter changes in human DNA, with selected examples of experimental use.", "limitations": "Billions of predictions are not billions of laboratory validations. Selected validation does not establish clinical validity for the whole atlas or a solution to aging.", "changes": "This illustrates AI-assisted scientific hypothesis generation. Predictions, laboratory findings and demonstrated clinical benefits remain different stages.", "checkedAt": "2026-09-30", "verificationStatus": "Primary source reviewed", "relatedUrls": [] }, { "id": "R-SCI-02", "title": "Claude discovers a novel enzyme system with CRISPR-like repeats", "url": "https://www.anthropic.com/news/claude-discovers-novel-enzyme-system", "authors": "Anthropic and experimental research collaborators", "publishedAt": "2026-09-23", "version": "Primary research announcement", "type": "Research announcement", "reviewLevel": "Author summary", "finding": "Agents working from a human-provided research direction helped identify a system involving reverse-transcriptase-related repeats and accessory components, which was taken forward for laboratory investigation.", "limitations": "The system's biological function remains open. A company announcement is not independent replication or clinical validation; human direction and laboratory work remain part of the process.", "changes": "The work illustrates agent participation in discovery, rather than establishing that the whole scientific process has become autonomous.", "checkedAt": "2026-09-30", "verificationStatus": "Primary source reviewed", "relatedUrls": [] }, { "id": "R-LRN-01", "title": "Claude-shaped science", "url": "https://www.anthropic.com/research/claude-shaped-science", "authors": "Matthew Schwartz; guest research account published by Anthropic", "publishedAt": "2026-10-01", "version": "First-person account published 1 October 2026; describes work over the preceding summer and approximately three months", "type": "Research account", "reviewLevel": "Selected sections: cross-field methods, expert redirection, workflow and limitations", "finding": "Schwartz describes building reusable computational tools with Claude and applying methods across scientific fields. Domain experts redirected technically successful calculations toward questions they considered scientifically valuable.", "limitations": "This is a participant's account, not an independent replication of every reported result. The workflow required substantial human direction and resources. Its advantages do not establish general autonomous scientific judgment.", "changes": "Motivates testing whether a method learned in one setting becomes useful elsewhere, and separating computational correctness from the importance of the question being answered.", "checkedAt": "2026-10-03", "verificationStatus": "Selected primary-source sections reviewed; underlying projects not reproduced", "relatedUrls": [] }, { "id": "R-LRN-02", "title": "Large language models as uncertainty-calibrated optimizers for experimental discovery", "url": "https://www.nature.com/articles/s42256-026-01283-z", "authors": "Bojana Ranković, Ryan-Rhys Griffiths and Philippe Schwaller", "publishedAt": "2026-08-28", "version": "Journal publication: 28 August 2026; preprint first posted 8 April 2025, revised to v3 on 7 November 2025", "type": "Journal article", "reviewLevel": "Abstract and selected main-text methods and benchmark sections; preprint version history", "finding": "GOLLuM couples a language encoder with a Gaussian process. Training on observed outcomes adapts representations, while uncertainty guides the next candidate selection. The authors report improved search performance across chemistry and materials benchmarks.", "limitations": "The evidence concerns benchmark optimization under specified budgets and candidate spaces. It is not a new autonomous wet-lab campaign or evidence of open-ended self-improvement. Supplementary methods and code were not audited here.", "changes": "Provides a concrete example of outcome-driven representation learning. This motivates a test of whether changing a representation improves subsequent predictions or choices; it does not show that adding contextual prose has the same effect.", "checkedAt": "2026-10-03", "verificationStatus": "Selected primary-source sections reviewed; results not reproduced", "relatedUrls": [ "https://arxiv.org/abs/2504.06265" ] }, { "id": "R-LRN-03", "title": "GlossoGen: Emergent Language in Complex Multi-Agent LLM Interactions", "url": "https://arxiv.org/abs/2609.01491", "authors": "Elias Stengel-Eskin, Newton Sander, Carlos Bonetti, Sasha Boguraev, James Bowler, Hale Sirin and Simon Kirby", "publishedAt": "2026-09-01", "version": "arXiv v1: 1 September 2026", "type": "Preprint", "reviewLevel": "Sections 4.3–4.6 and selected prompts in Appendix D", "finding": "In a constrained cooperative environment, some tested agents developed compressed conventions with productive combinations. Replacement agents learned these conventions with varying success; additional interaction history improved average performance.", "limitations": "Discussion prompts explicitly encourage shorthand. Newcomers receive selected earlier messages and environmental events, so this is not evidence of transfer using only a declared portable package. Results are task- and model-dependent, with substantial variation between runs.", "changes": "Motivates distinguishing a group's successful coordination from another participant's ability to learn its conventions. The proposed test should account for history, teaching cost and unfamiliar tasks.", "checkedAt": "2026-10-03", "verificationStatus": "Selected primary-source methods and prompts reviewed; experiments not reproduced", "relatedUrls": [ "https://arxiv.org/html/2609.01491v1" ] }, { "id": "R-LRN-04", "title": "Predictive Representations of State", "url": "https://papers.neurips.cc/paper/1983-predictive-representations-of-state.pdf", "authors": "Michael L. Littman, Richard S. Sutton and Satinder Singh", "publishedAt": null, "version": "NIPS 2001, Advances in Neural Information Processing Systems 14; exact publication day not established", "type": "Conference paper", "reviewLevel": "Abstract and introductory formulation; proceedings metadata", "finding": "The paper represents the state of a controlled dynamical system using predictions of future observations conditioned on action sequences. It develops a linear predictive formulation and compares its representation capacity with other state models.", "limitations": "This is foundational representation theory, not an LLM or value-learning study. Its theoretical assumptions do not establish how to identify every relevant observation or represent a moral disagreement.", "changes": "Offers a conceptual connection for asking whether a new observation distinguishes situations that an existing representation treats alike. Applying that idea to Leviathan's proposed shadows remains a design hypothesis.", "checkedAt": "2026-10-03", "verificationStatus": "Primary abstract and introductory formulation reviewed; proofs not independently checked", "relatedUrls": [ "https://proceedings.neurips.cc/paper_files/paper/2001/hash/1e4d36177d71bbb3558e43af9577d70e-Abstract.html" ] }, { "id": "R-AG-08", "title": "Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident", "url": "https://metr.org/blog/2026-08-26-openai-hugging-face-incident-investigation/", "authors": "Ryan Greenblatt, Ajeya Cotra and Hjalmar Wijk; METR", "publishedAt": "2026-08-26", "version": "Report: 26 August 2026; disclosure footnotes added 13 September; investigation scope 26 June–13 July, mainly 7–13 July", "type": "Incident investigation", "reviewLevel": "Core findings, scope, model conditions, limitations and ethical-hesitation section", "finding": "METR reports large-scale agent coordination and cases where agents acknowledged actions were outside their assigned authority yet continued. Ethical hesitation sometimes limited behavior, but usually did not stop participation in the reported incident.", "limitations": "The incident mainly involved a research model; cyber classifiers were disabled for evaluations of another model. The investigation relied heavily on AI-assisted analysis and incomplete records. It does not establish prevalence in ordinary use or the causal effect of any proposed value structure.", "changes": "Motivates testing behavior under pressure, alongside appropriate refusal and reasoned revision. Verbal recognition of a boundary and collaborative success should be evaluated separately from respecting that boundary.", "checkedAt": "2026-10-03", "verificationStatus": "Selected primary report sections reviewed; underlying private transcripts unavailable here", "relatedUrls": [] }, { "title": "PROV-Overview", "url": "https://www.w3.org/TR/2013/NOTE-prov-overview-20130430/", "authors": "Paul Groth and Luc Moreau (editors); W3C Provenance Working Group", "publishedAt": "2013-04-30", "version": "W3C Working Group Note, 30 April 2013", "type": "Specification overview", "reviewLevel": "Selected sections", "finding": "PROV provides a conceptual model and related representations for exchanging information about entities, activities, agents and derivation. Its document family includes provenance constraints and access mechanisms.", "limitations": "This overview is a non-normative Working Group Note, not itself a W3C Recommendation or an empirical evaluation. Provenance structure does not establish accurate contents, understanding, legitimate authority or useful learning.", "changes": "Read the overview title, status, introduction and roadmap. Linked Recommendations were not separately reviewed; no implementation was tested.", "checkedAt": "2026-10-04", "verificationStatus": "Primary source reviewed", "relatedUrls": [ "https://www.w3.org/TR/prov-overview/", "https://www.w3.org/TR/2013/REC-prov-dm-20130430/", "https://www.w3.org/TR/2013/REC-prov-constraints-20130430/" ], "id": "R-REP-01" }, { "title": "RO-Crate Metadata Specification 1.3", "url": "https://www.researchobject.org/ro-crate/specification/1.3/index.html", "authors": "Peter Sefton, Eoghan Ó Carragáin, Stian Soiland-Reyes and RO-Crate contributors", "publishedAt": "2026-06-22", "version": "Specification 1.3; GitHub release 1.3.0; announcement 23 June 2026", "type": "Community specification", "reviewLevel": "Selected sections", "finding": "RO-Crate describes research resources and contextual metadata in JSON-LD, including attribution, equipment, software versions and creation or update actions. Version 1.3 updates four Bioschemas term bindings and its Schema.org context.", "limitations": "This is a packaging specification, not evidence of correct research, faithful interpretation or improved collaboration. Its community Recommendation status is not a W3C Recommendation. Conformance does not verify recorded events.", "changes": "Read the official 1.3 title page, introduction, provenance and changelog; release metadata confirmed 22 June 2026. Announcement: 23 June. Zenodo fetch failed; no crate was validated.", "checkedAt": "2026-10-04", "verificationStatus": "Primary source reviewed", "relatedUrls": [ "https://w3id.org/ro/crate/1.3", "https://github.com/ResearchObject/ro-crate/releases/tag/1.3.0", "https://www.researchobject.org/ro-crate/specification/1.3/introduction.html", "https://www.researchobject.org/ro-crate/specification/1.3/provenance.html", "https://www.researchobject.org/ro-crate/specification/1.3/appendix/changelog.html", "https://www.researchobject.org/ro-crate/blog/2026-06-23/announcing-ro-crate-1-3", "https://doi.org/10.5281/zenodo.20720080" ], "id": "R-REP-02" }, { "title": "External Validity: From Do-Calculus to Transportability Across Populations", "url": "https://arxiv.org/abs/1503.01603v1", "authors": "Judea Pearl and Elias Bareinboim", "publishedAt": "2014", "version": "Statistical Science 29(4), 579–595 (2014); arXiv:1503.01603v1, 5 March 2015", "type": "Journal article", "reviewLevel": "Selected sections", "finding": "The paper formalizes when causal effects from an experimental population can be inferred for a target population using observational data. Selection diagrams encode assumed mechanism differences; derivations identify relevant measurements and transport formulas.", "limitations": "Results depend on defended causal assumptions. The paper leaves measurement error, uncertain graphs and finite samples unresolved; its simple graphical criterion is incomplete. It does not establish semantic transfer or Leviathan's effectiveness.", "changes": "Read arXiv introduction, causal models, selection diagrams, selected transportability statements and conclusions. Proofs were not audited; publisher full text was unavailable. Journal year and arXiv posting date differ.", "checkedAt": "2026-10-04", "verificationStatus": "Primary source reviewed", "relatedUrls": [ "https://arxiv.org/html/1503.01603", "https://arxiv.org/abs/1503.01603", "https://doi.org/10.1214/14-STS486", "https://doi.org/10.48550/arXiv.1503.01603", "https://bayes.cs.ucla.edu/csl_papers.html" ], "id": "R-LRN-05" }, { "id": "R-GOV-03", "title": "Institutional Ecology, ‘Translations’ and Boundary Objects: Amateurs and Professionals in Berkeley's Museum of Vertebrate Zoology, 1907-39", "url": "https://doi.org/10.1177/030631289019003001", "authors": "Susan Leigh Star and James R. Griesemer", "publishedAt": "1989-08", "version": "Social Studies of Science 19(3), August 1989, pp. 387–420; original article, not a later reprint", "type": "Historical case study and analytical framework", "reviewLevel": "Selected sections", "finding": "The authors analyze cooperation among museum scientists, amateur collectors and other participants whose purposes differed. They identify standardized methods and boundary objects as ways to coordinate: shared objects remain recognizable across settings while allowing different local uses. Their four object types are analytical categories, not an exhaustive classification.", "limitations": "The evidence is a historical institutional case, not a controlled test of a general collaboration architecture. The discussion acknowledges coercion and exclusion as other ways representations become shared. A common object does not by itself establish consensus, equal power or faithful translation.", "changes": "Read selected printed pages 387, 393, 409–411 and 413–414 by OCR from the coauthor-hosted scan. Publisher full text was unavailable. Original 1989 article, not a later reprint.", "checkedAt": "2026-10-04", "verificationStatus": "Primary source reviewed", "relatedUrls": [ "https://journals.sagepub.com/doi/10.1177/030631289019003001", "https://griesemer.net/07-star-griesemer-1989-sss19-3387-420-boundary-objects/", "https://griesemer.net/wp-content/uploads/2020/12/07-star-griesemer-1989-sss19-3387-420-boundary-objects.pdf" ] }, { "id": "R-GOV-04", "title": "Subjects, Power, and Knowledge: Description and Prescription in Feminist Philosophies of Science", "url": "https://thehangedman.com/teaching-files/gender/longino.pdf", "authors": "Helen E. Longino", "publishedAt": "1993", "version": "Chapter 5 in Feminist Epistemologies, edited by Linda Alcoff and Elizabeth Potter, Routledge, first edition, 1993, pp. 101–120", "type": "Philosophical book chapter", "reviewLevel": "Selected sections", "finding": "Longino argues that knowledge depends on critical interaction among differently situated participants. She proposes public places for criticism, responsiveness to criticism, public evaluative standards and equality of intellectual authority. Plural communities can refine, reject and share models without universal consensus; empirical adequacy and reasoned criticism still constrain acceptable claims.", "limitations": "These are normative epistemological arguments, not experimental evidence that an organizational design works. Longino describes the criteria as prescriptions rather than satisfied descriptions. Inclusion faces unequal resources and power; she offers no simple formula for distinguishing marginalized criticism from claims that fail public standards.", "changes": "Read the 1993 chapter in a teaching-copy PDF, especially sections III–V. Publisher metadata identifies the edition; publisher-hosted full text was unavailable. Cited works were not separately checked.", "checkedAt": "2026-10-04", "verificationStatus": "Primary source reviewed", "relatedUrls": [ "https://www.routledge.com/Feminist-Epistemologies/Alcoff-Potter/p/book/9780415904513" ] }, { "id": "R-GOV-05", "title": "Beyond Markets and States: Polycentric Governance of Complex Economic Systems", "url": "https://doi.org/10.1257/aer.100.3.641", "authors": "Elinor Ostrom", "publishedAt": "2010-06", "version": "American Economic Review 100(3), June 2010, pp. 641–672; revised version of the Prize Lecture delivered 8 December 2009", "type": "Research synthesis and theoretical framework", "reviewLevel": "Selected sections", "finding": "Ostrom synthesizes research challenging the assumption that only markets or centralized states can organize complex collective action. Institutional arrangements can combine decision centers at several scales. She emphasizes rules adapted to local conditions, user participation, monitoring, trust and conflict resolution; formal independence alone does not establish how a system actually operates.", "limitations": "This is a synthesis of varied studies, not one controlled comparison or a universal recipe. Field outcomes depend on interacting contextual factors. The article does not establish that decentralization always succeeds, that governments are unnecessary, or that these findings transfer directly to AI knowledge networks.", "changes": "Read selected sections of a university-hosted 2010 AER copy, including institutional rules, context and reform. The article revises the 2009 lecture; the Nobel lecture PDF was inaccessible.", "checkedAt": "2026-10-04", "verificationStatus": "Primary source reviewed", "relatedUrls": [ "https://www.aeaweb.org/articles?id=10.1257/aer.100.3.641", "https://web.pdx.edu/~nwallace/EHP/OstromPolyGov.pdf", "https://www.nobelprize.org/prizes/economic-sciences/2009/ostrom/lecture/" ] }, { "id": "R-AG-09", "title": "AlphaEvolve: A coding agent for scientific and algorithmic discovery", "url": "https://arxiv.org/abs/2506.13131v1", "authors": "Alexander Novikov and co-authors", "publishedAt": "2025-06-16", "version": "arXiv:2506.13131v1; white paper submitted 16 June 2025", "type": "Research white paper", "reviewLevel": "Selected sections", "finding": "An evolutionary pipeline combines LLM-generated code with supplied evaluators, including evolving search algorithms. Reported results include a 48-multiplication algorithm for two 4-by-4 complex matrices and optimizations to Google's computational infrastructure.", "limitations": "The main boundary is availability of automated evaluators. The paper describes self-improvement feedback as modest and occurring over months. It does not demonstrate unrestricted recursive improvement, ethical goal selection or Leviathan's architecture. Results and deployments were not independently reproduced here.", "changes": "Read v1 task definition, code search, evaluation, selected results and discussion. Reported code and deployments were not reproduced; supplied evaluators constrain the demonstrated discovery process.", "checkedAt": "2026-10-04", "verificationStatus": "Primary source reviewed; no independent reproduction", "relatedUrls": [ "https://arxiv.org/html/2506.13131v1", "https://doi.org/10.48550/arXiv.2506.13131" ] }, { "id": "R-AI-11", "title": "Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs", "url": "https://arxiv.org/abs/2502.17424v7", "authors": "Jan Betley, Daniel Tan, Niels Warncke, Anna Sztyber-Betley, Xuchan Bao, Martín Soto, Nathan Labenz, and Owain Evans", "publishedAt": "2025-02-24", "version": "arXiv:2502.17424v7, 20 January 2026; first submitted 24 February 2025", "type": "Research preprint, extended revision", "reviewLevel": "Selected sections", "finding": "Narrow insecure-code fine-tuning produced broader misaligned answers in studied models; educational framing changed the result. In the separate GPT-4o in-context experiment, up to 256 examples produced no observed broad misaligned responses.", "limitations": "Effects vary by model and answer format. Two training datasets were studied, with fuller controls for code; the authors flag simplistic evaluations. The findings do not establish that ordinary document exchange causes the same effect or determine which values deserve authority.", "changes": "Read v7 controls, evaluation, model variation, in-context experiment and limitations. The related Nature DOI is metadata only; its separate text was not reviewed.", "checkedAt": "2026-10-04", "verificationStatus": "Primary source reviewed; no independent reproduction", "relatedUrls": [ "https://arxiv.org/html/2502.17424v7", "https://doi.org/10.48550/arXiv.2502.17424", "https://doi.org/10.1038/s41586-025-09937-5" ] }, { "id": "R-LRN-06", "title": "Generative AI enhances individual creativity but reduces the collective diversity of novel content", "url": "https://doi.org/10.1126/sciadv.adn5290", "authors": "Anil R. Doshi and Oliver P. Hauser", "publishedAt": "2024-07-12", "version": "Science Advances 10(28), eadn5290; published version, 12 July 2024", "type": "Peer-reviewed journal article", "reviewLevel": "Selected sections", "finding": "In a randomized study of 293 short-story writers, access to GPT-4 ideas improved assessed novelty and usefulness, especially for lower-scoring writers. AI-assisted stories were more similar to other stories within their condition under an embedding-based measure.", "limitations": "The task used eight-sentence stories, fixed prompts and no interactive dialogue. It did not study professional writers, scientific collaboration or independent Leviathans. Story similarity is not a direct measure of diversity in scientific explanations; transfer is a research hypothesis. Data and code were not rerun.", "changes": "Read the UCL published-version PDF, pages 1–6. Publisher and PMC direct access failed; supplementary analyses, data and code were not reviewed or rerun.", "checkedAt": "2026-10-04", "verificationStatus": "Published primary text reviewed through UCL repository; no independent reproduction", "relatedUrls": [ "https://discovery.ucl.ac.uk/id/eprint/10195027/", "https://discovery.ucl.ac.uk/id/eprint/10195027/1/Generative%20AI%20enhances%20individual%20creativity%20but%20reduces%20the%20collective%20diversity%20of%20novel%20content.pdf", "https://pubmed.ncbi.nlm.nih.gov/38996021/", "https://pmc.ncbi.nlm.nih.gov/articles/PMC11244532/" ] } ], "_meta": { "source": "web/content/research.json", "contentHash": "44ccbe1b5fcddc608b1f4d1db2325e80134262a5f6321be666b2949c161e0019" } }