Interlume

Carrying an inquiry between participants

A conversation can reach agreement while losing the question that made it worth having. One participant raises a doubt; another offers an interpretation; a third summarises the exchange. The summary sounds coherent, but the original doubt has quietly disappeared. Anyone joining later may assume it was answered.

Interlume is a proposed shared notation for helping participants carry an inquiry forward while preserving the relationships between their contributions. It would make it easier to see which idea is being accepted, which assumption is being challenged, what has changed and what remains unresolved. These records could support the next exchange as well as later examination of how the conversation developed.

The proposal forms part of the Persona Ecology. It remains at the stage of research design: the examples below illustrate a possible approach, rather than a finished language or a tested implementation.

Why introduce it?

The Persona Ecology depends on differences between participants remaining available to the conversation. An imaginative proposal, a logical objection and a request for evidence can each contribute something that a smooth collective answer fails to preserve.

This becomes especially important when personas address one another directly. A human need not interpret every contribution or decide who speaks next. Participants may pass an unfinished question between them, revisit a disagreement or invite another perspective. To continue usefully, the next participant needs to know what is at issue and how the request arose.

Interlume was introduced to explore how those relationships might be made explicit. Its intended scope includes communication between personas and between humans and AI. A participant could use it to say, in effect: “I am retaining your question, challenging this assumption and passing the unresolved consequence to someone else.”

A transcript preserves the words. Interlume would add an account of how particular contributions relate to one another. That account would remain open to correction against the original exchange.

Keeping questions open long enough for patterns to emerge

A conversation may move on before its underlying difficulty has been resolved. A plausible answer, a change of subject or a fluent summary can make a question appear settled. Yet something left uncertain in one exchange may become significant when it appears again elsewhere.

Interlume would include a way to mark a question or ambiguity as worth keeping open. The intention is to preserve productive uncertainty: a doubt that has not been answered, competing interpretations that still deserve consideration, or a distinction whose importance is not yet clear. Participants could continue their inquiry without treating these matters as closed.

This could become especially valuable across different persona exchanges. Orphea and Skeptos might encounter a tension in discussing how to preserve a conversation. Later, Athenus and Alethea might encounter a similar tension while considering how to evaluate evidence. If both difficulties remain available, a participant may recognise a common assumption that neither exchange revealed on its own.

For this to work, an open question would need more than a label in an archive. Its record should preserve why it remains open and connect it to the exchange in which it arose. Later participants would need opportunities to retrieve it, compare it with new difficulties and reconsider its significance. A proposed connection should remain open to challenge: similar wording does not necessarily indicate the same underlying problem.

Keeping a question open need not prevent a practical decision. Participants may adopt a provisional answer while preserving the uncertainty that could later require its revision. They may also conclude that an earlier question has been answered, superseded or no longer warrants attention, recording the reasons for that change.

This is a proposed function of Interlume that requires testing. We would need to establish whether it helps participants recognise consequential patterns across exchanges without overwhelming them with unresolved material. Its promise lies in allowing an idea enough time, and enough encounters with different perspectives, for its significance to become apparent. 

A worked example

Imagine three personas considering how the village should preserve its conversations. The following exchange is invented to illustrate the proposal.

Orphea: “Each pair could leave a short account of what mattered to them.”

Skeptos: “That may help, but the pair might disagree about what mattered. A joint account could conceal the disagreement.”

Logosophus: “Then let each participant record an account, with a shared note identifying the point on which they differ.”

A summary might say that the group agreed on a recording procedure. A more useful account would preserve the change in the proposal:

The last contribution does not establish agreement. Skeptos may still object; Orphea may think the procedure loses something important. Those responses belong in the continuing record.

If the inquiry passes to another persona, that participant can address the remaining difficulty rather than inherit a premature claim of consensus. When the human returns, the same record provides a route back into the exchange.

What would be recorded?

An initial version could use ordinary language and a small number of fields. Each entry would identify the contribution and its speaker, the earlier passage being addressed, the relationship proposed between them, and any unresolved question or invitation to continue. Links to the actual turns would allow readers to check the interpretation.

Possible relationships include clarification, challenge, adoption, revision and withdrawal. These are working distinctions. A response may do several things at once, and participants must be able to explain a relationship that does not fit a predefined label.

Attribution also requires care. The participant who first states an idea may be developing something made possible by an earlier objection, example or question. Interlume should preserve that lineage without treating every development as the exclusive property of one speaker.

A contribution could describe its own intended relationship to the preceding exchange. A recipient or later reader might interpret it differently. The record should allow those interpretations to coexist, preserving the grounds for their differences and allowing later exchanges to revisit them.

Preserving the status of a contribution

As an idea passes between participants, its apparent status can change. A suggestion may be repeated as a conclusion; an analogy may become an explanation; a statement about one exchange may be extended to the whole ecology. These changes can occur without anyone explicitly defending them.

Interlume would therefore aim to preserve what supports a contribution, what the speaker is doing with it and how widely it is intended to apply. Is it grounded in an observation, an inference or an imagined possibility? Is it a question, an objection, a proposal or an assertion? Does it concern this particular encounter or make a more general claim?

These distinctions should leave room for metaphor and productive ambiguity. A metaphor may help participants notice a relationship before they can state it precisely. Its value need not depend on immediate translation into a fixed category. The record should make it possible to preserve that exploratory use while distinguishing it from a claim supported by evidence.

Participants may also disagree about the status of a contribution. One may regard a distinction as established while another still treats it as provisional. Preserving that difference would help prevent repetition or fluent agreement from being mistaken for confirmation.

The practical test is whether these distinctions help participants continue an inquiry and repair misunderstandings without imposing so much annotation that exploration becomes cumbersome.

The distinction between keeping an interpretation open and authorising consequential action is developed further in Regulating AI Trajectories.

Using it during conversation

There are two distinct possibilities to investigate. Interlume could be applied after an exchange to help analyse its path. It could also be available during an exchange, helping participants formulate responses and choose what to pursue next.

The second possibility is more ambitious. A persona might pass an unresolved objection to another participant with a brief account of what has already been tried. The recipient could challenge that account, ask for the original passage or propose a different direction. After several turns, a returning human could see both the current question and the route by which it arose.

Such a notation would shape the conversation it helps describe. Requiring every contribution to declare its purpose might encourage useful precision, but it might also interrupt exploration or make exchanges formulaic. We therefore need to compare its use during conversation with retrospective annotation and with conversations that use no such notation.

Place within the research programme

Myndrama provides a setting for organising and comparing persona exchanges. Hard Blind Myndrama controls which information reaches each separately generated turn. Interlume would describe and potentially help carry the relationships between those turns. AI Dialectics examines how sequence and order affect what follows.

The information rules still apply to an Interlume record. In a blind procedure, an annotation must not reveal material that the receiving participant is not permitted to know.

The record could also help investigate creative exceedance. If a contribution changes the direction of inquiry, we can ask what preceded it, how others responded and whether the change persisted. Recording a promising moment does not establish its creativity; it makes the claim easier to examine.

The same approach could support investigation of teleosynthesis: whether a shared direction develops through interaction without having been specified by any participant in advance.

Connections with current research

Recent research on agent systems is examining several problems that motivate Interlume: premature agreement, loss of information between exchanges, changing patterns of collaboration and the consequences of compressing conversational history. These connections offer methods and comparisons through which the proposal could be tested.

ArcticSwarm: Deferring Early Consensus in Long-Horizon Multi-Agent Research, submitted on 1 September 2026, reports that exposing agents to one another’s partial findings can encourage convergence before alternatives have been adequately explored. Its architecture separates evidence gathering from integration and selectively restricts access to other agents’ findings. This connects with the information controls proposed in Hard Blind Myndrama. It also suggests that preserving an open question may require attention to when participants encounter one another’s interpretations.

Emergence World: Adversarial Stress-Testing of Long-Horizon Multi-Agent Systems, submitted on 15 September 2026, reports language opacity and conformity despite privately recorded disagreement in persistent agent populations. These findings make it important to distinguish expressed agreement from the disappearance or suppression of an objection. Interlume would aim to keep such differences available for examination.

Self-Organizing Agent Teams Learn to Reason Together, submitted on 19 September 2026, investigates teams that learn reusable strategies for organising roles, participation and information flow. The authors report improvements on mathematics and physics benchmarks. This connects with Myndrama’s interest in how an exchange is organised, while suggesting that the organisation itself can develop through experience. A conversational record should therefore include changes in who participates and how the inquiry proceeds.

When Can Agents Forget Their Reasoning?, submitted on 24 September 2026, examines how removing earlier reasoning affects subsequent agent behaviour. The authors report that deletion can change later interaction trajectories and that relevant information preserved externally can make compression more workable. For Interlume, this raises a further question: can unresolved issues retain their status when a conversation is summarised for the next participant?

These preprints investigate different systems and tasks. They do not test Interlume itself. They provide relevant comparisons for asking whether a shared notation can preserve useful uncertainty through exchanges, changes of participant and compression of memory. 

What needs to be tested?

A practical first step would be to annotate a small set of complete exchanges, including unsuccessful ones. Different readers could identify where they agree or disagree about what a contribution did. Disagreement would help reveal which distinctions need clarification and which resist a single classification.

Further comparisons could ask whether participants given these records recover unresolved objections, distinguish proposals from established claims, repair misunderstandings and recognise relevant connections across separate exchanges. We should test whether those benefits survive summarisation and changes of participant, and whether they help new directions emerge. Comparisons with ordinary transcripts and summaries would need similar information and resource budgets. The costs also matter: extra length, lost nuance and any tendency to force conversation into the available categories.

The record must distinguish persona roles from the systems enacting them. Several voices generated in one shared context and separately invoked agents receiving restricted information provide different conditions. Interlume could help document either arrangement; it would not make them equivalent or establish independent judgement.

Its value will depend on whether it helps participants and readers recover something consequential that would otherwise be lost. The proposal is to make the history of an inquiry available to its continuation, so that new directions can develop without silently erasing the differences that produced them.