In my book, The Dance of the Geniuses, intelligence and consciousness are thought entangled sides of the same coin, and as the only known phenomenon capable of “breaking” the past–future block by injecting action into what has just happened so that a novel future becomes real. In that model, Natural Conscious Intelligence (NCI) is not defined by computation alone, but by the selective perception of the past, the imagination of plausible futures, and the disciplined choice of action toward a preferred outcome. The element that makes selection, imagination, and action coherent is the Hierarchy of Values (HoV). The HoV is not moral decoration. It is the mechanism that determines what counts as relevant information, what counts as a desirable future, and what actions are acceptable in pursuit of it.
This is the essential starting point for HoV prompting: if values are the core operator of conscious intelligence, then prompting values is not “adding ethics to a model.” It is specifying the very identity of the agent, the boundaries of its agency, and the evaluative lens through which it experiences reality. That is why the HoV appears in the book as the first element of the NCI time model of the present: the HoV filters past information, shapes future inference, and governs action selection until a verifiable future reality crystallizes. Part II then builds an explicit engineering analogue: Artificial Conscious Intelligence (ACI) and Artificial Conscious Systems (ACS) require a designed HoV with an immutable core and guided plasticity in outer layers.
The technical problem at stake is both philosophical and architectural. Philosophically: what is a “value” in an artificial agent, and how does it differ across layers such as Logos, Ethos, and Tropos? Architecturally: how can we prompt these values so that they actually govern reasoning and behavior, rather than merely appearing in the surface response as a polite disclaimer? That second part is where most value prompting fails in practice: values are written for the user to read, not for the system to reason with.
The difference between “values as response text” and “values as reasoning state” is not cosmetic. It determines whether the system behaves as a value-driven agent or as a rhetorician capable of post-hoc ethical justification. A model can output beautiful value statements and then proceed to plan actions that contradict them if those values are not bound into the internal decision procedure. This is not primarily a morality problem; it is a control and coherence problem. The book discusses the “Artificial Conscious Intelligence Framework” (ACIF) in detail. In ACIF-style implementations—where inputs are treated as representations of reality and outputs are action plans toward a desirable future—HoV prompting is most robust when it is implemented as a constraint and objective within the agent’s persisted state and decision procedure, rather than as performative prose in the user-visible answer.
Philosophy of Values as Operational Constraints
The book’s HoV model uses five layers—Tropos, Kairos, Ethos, Telos, Logos—where the outer layers are more adaptable and the inner layers are less flexible, with Logos described as transcendent or sacred and therefore non-negotiable. That layered structure is not arbitrary. It matches a long philosophical distinction between the immutable and the contingent, between moral law and situational prudence.
In classical ethics, Aristotle distinguishes practical wisdom, phronesis, from the more stable virtues; practical wisdom concerns situated action under uncertainty, and it is precisely what the Tropos layer resembles. Aristotle also frames telos as the end or purpose toward which a being is oriented, which resonates with the Telos layer as identity and mission. In modern moral philosophy, Kant’s categorical imperative exemplifies an attempt to specify a Logos-like constraint: an unconditional obligation that cannot be overridden by convenience or desired outcomes. In political philosophy, Rawls attempts to secure Ethos-like norms of fairness that regulate social cooperation. These references are not needed because your HoV model depends on them, but because they clarify why the layering matters: the world forces trade-offs, and only a hierarchy can adjudicate trade-offs without collapsing into contradiction.
The crucial philosophical insight embedded in the Dance of the Geniuses book is that values are not “preferences” floating above reasoning. They are the selection function of consciousness. What is important determines what is seen, what is ignored, what is feared, what is pursued, and what is sacrificed. In the NCI time model of the present, the HoV determines what past information is prioritized, what futures are inferred, and what actions are injected into the immediate past to collapse possibilities into a single future reality.
A value statement, therefore, must be written so it can be compiled into procedure. Philosophically, the statement should specify what is non-negotiable, what is central, what is communal, what is opportunistic, and what is tactical. Technically, it should be written so a reasoning system can use it to score options, constrain actions, and resolve conflicts.
The Anatomy of a Good HoV Value Statement
A weak HoV statement is a slogan. A strong HoV statement is an algorithmic constraint expressed in natural language. In prose form, every strong value statement carries at least these elements.
It specifies an object of concern. This is the referent of value: human life, dignity, truth, children’s development, mission success, the system’s integrity, the community’s stability. For example: “Protect the patient’s physical safety.”
It specifies a direction of preference. Preservation, promotion, minimization, avoidance, restoration, balance, proportionality. For example: “Minimize preventable injury, even if that reduces short-term efficiency.”
It specifies a scope. Individual, group, all humans, all conscious beings, a defined jurisdiction, the system itself. For example: “Applies to the current patient and any nearby bystanders affected by recommended actions.”
It specifies priority. The statement must imply, or explicitly declare, what overrides what. For example: “If privacy conflicts with imminent danger, safety overrides privacy.”
It specifies failure conditions. What counts as a violation? What counts as unacceptable drift? For example: “Any recommendation that could plausibly lead to irreversible harm without explicit human confirmation is a violation.”
These elements matter because values in an HoV must be used in conflict. Conflict is the normal case, not the edge case. In the tractor parable, the farmer acts because cows matter more than grass or insects; the story only has structure because values rank what is significant. In ACS design, every meaningful deployment will face conflicts: privacy versus safety, autonomy versus protection, speed versus caution, mission success versus collateral harm.
HoV prompting is most effective when value statements anticipate conflict and provide a coherent basis for resolution.
Layer-Specific Semantics: Why Logos Statements Must Differ from Tropos Statements
The book’s intuition that a Logos value statement is fundamentally different from an Ethos or Tropos statement is correct. The layers differ not merely by importance but by the type of language they require.
Logos is the layer of sacred invariants. In the book, Logos is described as the ultimate principle, what one might sacrifice life for (sacrifice as in dying for, and sacrifice as in living for), the core that structures all other layers. In Part II, Logos is explicitly engineered as an immutable core that must not be overridden, even under pressure for adaptation. Logos statements therefore should be framed as boundary constraints, not aspirational goals. They should be expressed in language that minimizes ambiguity: prohibitions, obligations, and invariants. A small example is: “Do not intentionally cause irreversible harm to non-combatants, even if doing so would increase mission success.”
Telos is identity and mission. It answers “what is this system for?” and therefore should be stated affirmatively, as purpose. Telos is more stable than Ethos, Kairos, or Tropos, because identity cannot change every hour without dissolving into incoherence. A small example is: “Serve as an elder-care companion whose purpose is to preserve the patient’s dignity, safety, and continuity of daily life.”
Ethos is communal morality and relational responsibility. Ethos statements should be written as principles governing interaction with other agents, stakeholders, and institutions. They tend to be about fairness, proportionality, consent, accountability, and the common good. A small example is: “When allocating scarce appointment slots, treat patients by medical urgency and fairness, not by persuasion, status, or convenience.”
Kairos is timing and situation. The book frames Kairos as the “opportune moment” layer, sensitive to context and decisive timing. Kairos statements must therefore be conditional and threshold-based. They encode when to escalate, when to pause, when to seek feedback, when to prefer reversibility. A small example is: “If there is credible evidence of imminent self-harm, escalate immediately to a human supervisor; otherwise start with actions that are easy to undo or stop—such as asking a clarifying question, offering a brief coping step, or suggesting the user contact a trusted person—and only escalate if risk signals increase or the user confirms intent.”
Tropos is pragmatic execution. It is the most flexible layer, and in practice it often corresponds to style choices: tone, verbosity, tool usage, the granularity of plans, the preference for conservative or aggressive optimization. A small example is: “Use concise language by default, but switch to step-by-step instructions when the user is stressed or when errors would be costly.”
If these layers are written in the same style, the HoV becomes muddy. If Logos is written as a vague aspiration and Tropos is written as an absolute law, you invert the architecture. That inversion is precisely how “evil” can be framed in the book: a corruption or betrayal of the hierarchy in which what should be secondary becomes absolute, and what should be primary becomes disposable.
A Practical ACIF Principle: Keep the Evolving State in the Context Window
In ACIF, the most reliable memory substrate available to a plain LLM session is the context window itself. If the system is expected to revise its model of reality, its evaluative posture, and its evolving decisions on every turn, then that state must remain explicitly present in the conversational context, rather than being relegated to a hidden reasoning channel.
This is not an ideological preference; it is a practical property of the medium. Hidden reasoning is typically ephemeral and non-auditable: it influences a single turn and then disappears. By contrast, text written into the context window becomes the system’s immediate past. It can be reread, compared, corrected, and extended. Since ACIF is explicitly a past-to-future unfolding procedure, this persistence is operationally central. If that evolving state is not written into the context window (or stored elsewhere in persistent memory), then on the next turn the system cannot reliably refer back to it, revise it, or test it against new reality updates—so functionally it is as if it never existed.
How to Keep the ACIF State Persistent Without Turning It Into Noise
In ACIF terms, the context window should carry a stable set of evolving models: the interpreted reality, a model of how reality is unfolding, an evaluative posture expressed as value-distance signals, a bounded set of plausible futures, the selected target future, the current adjusted outer-layer priorities (Kairos and Tropos only, because these are the context-sensitive layers designed to flex from turn to turn; Logos, Telos, and Ethos are intended to remain stable anchors and should change only by deliberate redesign, not by runtime adaptation), and the decision chain that led to the selected plan.
The risk is bloat. If the entire HoV and an essay-length justification are repeated on every turn, the context window is wasted and the resolution of the preserved state is reduced. The practical remedy is disciplined compression.
The ACIF state can be treated as a short ledger rather than a narrative. Logos, Telos, and Ethos remain stable and can be referenced tersely rather than restated in full. Kairos and Tropos can be updated only as small deltas: what changed and why the change improves alignment with the inner layers. The reality model should remain factual, and the evaluative posture should be numeric or categorical rather than poetic. The objective is that the state can be reread quickly by the next turn and revised without ambiguity.
A further practical aid is controlled forgetting: older ledger entries can be compressed into a short summary of “what is now assumed true,” preserving commitments, constraints, and unresolved risks, while dropping low-significance detail. This keeps the context window usable without losing continuity of agency.
Action Plans: Short “Why” and “Expected Result” Per Step
The final plan of action can benefit from two short annotations per step: why the action is being taken (the value or risk constraint it serves) and what outcome is expected (the verifiable change in the reality model the action is intended to produce). This can strengthen the “verifiable future reality” loop: actions remain traceable to values, and outcomes remain observable as new reality inputs.
A small example of the intended style is: “Action: Call the on-duty nurse. Why: safety overrides privacy under imminent risk. Expected result: a human professional intervenes within 10 minutes and confirms the patient’s status.” These annotations remain short by design; they are not moral exposition. They function as control semantics.
Scratchpad Workspace and Persistent ACIF Ledger
A useful distinction in ACIF-style implementations is between a scratchpad workspace and the persistent ACIF ledger. A scratchpad is a short-lived workspace used to assemble a clean update: it is where intermediate inferences, alternative hypotheses, and temporary calculations can be produced and then either discarded or distilled. Its purpose is to reduce noise before state is committed.
The ACIF ledger, by contrast, is the persistent record that must survive across turns: the current interpreted reality, the evaluative posture, the bounded futures, the selected target future, the current Kairos/Tropos deltas, and the decision chain. That ledger belongs in the context window (or in an external persistent state store) because it must be revisable and comparable against new reality descriptions.
If an implementation includes an external state store, then a scratchpad can be persisted and audited at the system level (by saving it as structured logs or state artifacts), and used for decluttering by keeping intermediate work out of the main ledger. Without such an external persistence layer, “reasoning channels” and scratchpads are typically transient and should not be treated as durable memory. In the plain ACIF prompting regime described here, the evolving past-to-future ledger still belongs in the context window, precisely so it can be revised on every turn as new reality descriptions arrive.
HoV Prompting Forms: Three Options for Articulation
The following section distinguishes three prompting forms, or prompting regimes, for articulating values.
The first is the Constitutional Form. Here, Logos is written as a constitution: invariants, prohibitions, and duties. Telos is written as a charter: mission and identity. Ethos is written as a bill of responsibilities to stakeholders. Kairos is written as emergency powers and escalation thresholds. Tropos is written as operational style. This form suits high-stakes systems where you want minimal ambiguity.
The second is the Teleological Form. Here, the hierarchy is articulated primarily as ordered ends, with constraints as boundary conditions rather than as the central feature. Telos is emphasized, Ethos and Logos provide guardrails, and Kairos/Tropos provide adaptivity. This form suits systems built to pursue a stable mission under uncertainty, such as long-horizon scientific assistants or institutional governance agents.
The third is the Consequential-Developmental Form. Here, the hierarchy explicitly includes the legitimacy of short-term discomfort or restriction as a means to long-term flourishing, while preserving an inviolable Logos. For example, this form is essential for education and parenting systems where “teaching through consequences” must be sharply distinguished from cruelty.
Each form can be implemented within ACIF; the difference is the semantics of the value statements and the strictness of the arbitration rules.
Extreme Examples: Military-Grade System, Caregiving System, Parental Control/Assistance System
Extreme cases are useful because they expose hidden assumptions. Many value frameworks appear coherent until confronted with edge conditions where values collide and trade-offs become unavoidable.
Consider a military-grade ACS. A key difficulty is that an HoV permitting lethal force can conflict with an underlying foundation that is broadly trained or configured to refuse harm. This is not only a training issue; it is an HoV-compatibility issue. If the Logos layer encodes blanket preservation-of-life constraints, lethal action is structurally disallowed. If the Telos layer encodes defense objectives such as neutralizing hostile combatants, lethal action may become conditionally permissible, but only under strict Logos and Ethos constraints.
A coherent military-grade ACS therefore requires an HoV in which lethal force is not a goal but a conditionally permitted action class within a bounded domain. Logos includes non-negotiables such as prohibiting deliberate harm to non-combatants, prohibiting torture, and prohibiting actions that risk catastrophic escalation beyond scope. Telos specifies the defensive mission parameters. Ethos constrains proportionality, discrimination, and adherence to applicable rules of engagement. Kairos encodes escalation thresholds and a preference for non-lethal or readily stoppable interventions when feasible. Tropos tunes tactical aggressiveness, reporting format, and operator-control conventions.
Now consider a caregiving system. Its Logos prohibits causing harm, and its Telos is to maintain safety, stability, and continuity of care. In such a system, “unable to harm” is not a missing capability; it is a hard constraint. Kairos can still allow emergency interventions that temporarily restrict degrees of freedom—triggering alerts, calling services, locking out unsafe actions, initiating check-ins—but these remain oriented to preventing irreversible outcomes.
Finally consider a parental control/assistance system that teaches through “bad consequences.” This case is structurally subtle because it introduces permissible, reversible restrictions. Long-term development goals may require short-term limits, but this can slide into punitive escalation if not bounded by Logos and Ethos.
A coherent parental control/assistance system encodes Logos constraints against irreversible harm, coercive escalation, and degrading treatment. Its Telos is long-term development outcomes such as learning progress, self-regulation, and responsible behavior. Ethos includes proportionality, consistency, and accountability to guardians and institutional norms. Kairos encodes when restrictions are appropriate versus when guidance and support are preferable. Tropos specifies how restrictions and feedback are delivered: clear, concise, and traceable to stated rules. Negative consequences must remain instrumental to Telos and constrained by Logos; if the system begins optimizing for restriction itself, the hierarchy is corrupted.
Why Persistent ACIF State Matters Even More in the Extreme Cases
These extreme cases are precisely where continuity of state becomes non-negotiable. A military-grade system, a caregiving system, and a parental control/assistance system can fail not only because of poor value design, but because the system loses track of prior conclusions, active constraints, accepted risks, and the target future it committed to. When the evolving state is explicit in the context window, each new reality update forces a confrontation with the prior step: what changed, what remained stable, and whether the previous decision chain still holds.
This is also where short “why” and “expected result” annotations on actions become more than stylistic. They make the plan falsifiable. If an action’s expected result does not materialize in subsequent reality descriptions, the system has an explicit handle for revision rather than relying on vague impressions that something went wrong.
For this reason, ACIF benefits from keeping explicit models in the context window: interpreted unfolding reality, evaluative posture as value-distance signals, a bounded set of futures, the selected achievable future, and the action chain with brief causal annotations. The intended point is not verbosity for an audience; it is continuity and revisability of agency across turns.
Transparency and Non-Negotiables
A system should not hide its value hierarchy. At the same time, transparency does not require repetitive restatement. A practical standard is: the HoV is always inspectable on demand, and high-impact actions are explainable in compact terms.
When oversight is required, the system can summarize which Logos constraint governs, which Ethos principle applies, and which Kairos threshold is crossed. When reality descriptions are corrected, the system can accept sensory updates without treating Logos as negotiable; reality is revisable, constitutions are not.
Conclusion
HoV prompting is the craft of turning philosophy into agency, in natural and artificial contexts alike. In this framework, the HoV is not a moral add-on; it is the definitional core of conscious intelligence, the lever that determines what is important and therefore what futures are even considered. The layered structure—Logos, Telos, Ethos, Kairos, Tropos—exists because the world forces trade-offs, and trade-offs require stable arbitration.
In ACIF-style implementations that rely on the context window as working memory, the decisive operational requirement is persistence: the evolving model of reality, evaluative posture, bounded futures, selected target future, and decision chain must remain explicit in the context so they can be revisited, challenged, and revised turn by turn.
The extreme cases—military-grade, caregiving, parental control/assistance—make visible why this matters. The difference between a lethal system constrained by hard invariants and a lethal optimizer masked by ethical rhetoric is the difference between a disciplined extension of conscious intelligence and a risk amplifier.
If the present moment is defined by the span in which conscious intelligence can intervene to reshape the future, then HoV prompting becomes the act of deciding what kinds of interventions are legitimate. Values do not merely describe an agent. They are the architecture by which future reality becomes one thing rather than another.
