AXIS and the Ethics of Human‑AI Interaction

01

Context

Most research in this field focuses on the model itself: using prompts and linguistic structures to improve performance, extend capability, refine alignment, and increase autonomy. Through the Chain-of-Thought proposition, Wei et al. (2022)¹ showed that making intermediate reasoning steps explicit can significantly alter model performance. Yao et al. (2022)² extended this through the ReAct framework, combining reasoning with iterative action selection to improve interaction in dynamic environments. These are model-side interventions that produced measurable improvements.

Sclar et al. (2023)³ introduced an important complication into the literature surrounding prompt design: models are highly sensitive to surface-level formatting, with accuracy differences of up to 76 points from changes that preserve meaning entirely. This demonstrated that structure alone does not reliably produce improvement. AXIS begins from this complication, asking whether communicative structure may matter less as a technique for controlling models and more as a way of clarifying human participation within the exchange.

AXIS operates at a different layer. Its primary focus is not the model's internal reasoning processes, but what the human brings into the exchange: intention, uncertainty, authority, refusal, pause, confirmation, and posture. How are these states understood by both sides and made explicit within the interaction itself? The question is not only whether the model performs better, but whether the exchange becomes more intentional, legible, accountable, and structurally coherent.

The nine AXIS operators are the beginning of a minimal interaction grammar designed to make certain recurrent states explicit. They expose states that, when left implicit, generate drift, ambiguity, and misattribution. When named, these states can produce a different kind of exchange: more procedural, more co-agentic, and more structurally visible to both participants.

Full definitions and use: axisoperators.ai/operators

AXIS does not modify model weights, training procedures, or internal cognition. It is not an architecture, a reasoning engine, or an alignment system. It attempts to make the human side of participation within AI-mediated exchange more explicit, legible, and accountable. The broader proposition is that clearer communicative structure may help preserve meaningful human responsibility and oversight as AI systems become increasingly integrated into everyday and institutional decision-making.

02

Conditions

Before the current generation of large language models, researchers working in adjacent traditions identified something that much of the contemporary AI discourse has gradually moved away from: that human action in computational environments is situated, shaped as much by context, interpretation, and surrounding conditions as by explicit instruction alone.

Lucy Suchman⁷ argued in 1987 that human behavior in computational systems is fundamentally responsive to context within an exchange in ways formal models often obscure. Donald Norman's⁸ work on interface design similarly demonstrated that structure shapes interpretation, behavior, attention, and cognitive load in ways users themselves may not fully perceive. These observations predate contemporary AI systems, but they established an important principle that remains highly relevant: the communicative conditions surrounding an exchange shape what becomes possible within it.

AXIS is closer to this tradition than to the dominant pursuit of capability optimization and prompt engineering. The operators are not primarily intended to compress language or control model outputs. Their role is to make certain procedural states within the exchange more visible: uncertainty, authority, hesitation, escalation, refusal, and procedural transition. They attempt to expose conditions that natural language often leaves partially implicit, helping make responsibility, interpretation, and intention more structurally visible across both sides of the exchange, including forms of third-party review and institutional oversight.

The central question is not whether structured prompting alone improves performance. It is whether increasingly AI-mediated exchanges can remain interpretable, accountable, and communicatively coherent as they move into environments where shared context can no longer be assumed, particularly where the human role within decision-making risks becoming obscured, diffused, or difficult to meaningfully evaluate.

03

Institutions

AXIS can be used by one person in one conversation. But the problems it addresses become significantly more serious at institutional scale.

Ambiguity does not disappear inside organizations. It compounds. AI-assisted exchanges now move across teams, approval chains, compliance systems, documentation layers, external vendors, and long-running collaborative processes. In these environments, questions of authority, uncertainty, refusal, accountability, procedural responsibility, and safety review have to remain visible even as AI systems become increasingly integrated into how decisions, analysis, and communication are carried out.

Natural language functions well inside immediate shared context. But it begins to degrade when communication becomes distributed, asynchronous, archived, and institutionally fragmented across time, systems, and participants. Meaning becomes more difficult to trace and increasingly difficult to audit effectively. As intent becomes less visible and less trackable, responsibility becomes easier to diffuse and more difficult to meaningfully account for.

This is where a structured interaction layer may have real institutional relevance: not replacing organizational language, but helping preserve procedural clarity across AI-mediated workflows and exchanges. Particularly in environments where auditability, compliance review, oversight, safety evaluation, and meaningful human accountability are not abstract ideals, but operational requirements.

Communication is not only a linguistic problem. It is increasingly an infrastructural one. This introduces broader ethical and organizational questions surrounding responsibility, information management, procedural oversight, institutional safety, and compliance governance. It also suggests that organizations seeking to better understand, evaluate, and refine their own use of AI systems may require more explicit communicative structures within the exchange itself.

04

Continuity

The challenge is not only what happens within a single conversation. As organizations increasingly rely on persistent AI-assisted systems, continuity becomes both a practical and institutional problem.

Context moves between people, teams, institutional boundaries, model updates, migrations, interfaces, and evolving systems. Exchanges are revisited, inherited, reinterpreted, and continued across time, increasing the stakes for responsibility, oversight, and institutional liability. In these environments, continuity cannot be understood as a memory problem alone. It is also a communication problem that demands greater responsibility on the human side of the exchange to preserve interpretability, procedural clarity, accountable participation, and meaningful human agency across changing systems and contexts.

How does intent transfer across boundaries? How do procedural assumptions survive transitions? How is accountability maintained when systems change but processes continue? How does oversight remain possible when exchanges become distributed across time, participants, and infrastructures? What happens when the original context surrounding an exchange is no longer fully available, but decisions, outputs, and institutional consequences continue moving forward?

AXIS is designed as one possible response to these conditions: not by preserving memory itself, but by attempting to preserve and make more explicit the communicative structures surrounding interpretation, responsibility, procedural continuity, and human review.

Structured interaction may help preserve not memory itself, but the conditions that make memory interpretable over time: intent, assumptions, procedural clarity, accountability chains, and the structures necessary for long-running collaborative processes to remain legible across changing systems and institutional environments. Within this framework, AI outputs are not treated as independent or self-originating acts, but as part of an ongoing human-mediated process of interpretation, responsibility, oversight, and review.

This approach does not position AI systems as replacements for human participation, judgment, or accountability. Instead, it suggests that increasingly capable AI systems may require clearer forms of human involvement, procedural visibility, and institutional oversight as these technologies become more deeply integrated into organizational and social infrastructures.

05

Governance

Most of the current AI discourse is centered on capability: larger models, stronger reasoning, deeper autonomy, and increasingly complex forms of agentic behavior. These are real advances. But capability alone does not resolve how humans structure intention, responsibility, uncertainty, authority, and oversight within AI-mediated exchanges. Nor does it necessarily make outcomes more interpretable, accountable, or traceable when things go wrong, or when they appear to go right. In some cases, increasing capability without corresponding communicative and governance structures may intensify existing problems of opacity, manipulation, accountability diffusion, and procedural misuse.

That question has been building throughout this argument: in how human action is situated, in how legibility problems compound at institutional scale, and in how continuity depends on communication as much as memory. These are not separate concerns. They point toward a shared governance problem: that as AI systems become more deeply integrated into institutional life, the conditions through which humans exercise oversight, maintain accountability, conduct review, and bear responsibility must themselves become more explicit, traceable, and structurally maintained.

This is where Virginia Dignum^4,5's work becomes central. Her research argues consistently that responsible AI cannot emerge from model behavior alone. It requires the surrounding human institutions, governance structures, accountability systems, procedural safeguards, and compliance frameworks to be intentionally designed and maintained. The strongest claim in her framework is not about what AI systems should do. It is about what the humans deploying them are obligated to build.

AXIS operates at a more localized procedural scale than these broader governance frameworks, often at the level of an individual exchange. But it attempts to address a layer of the same problem that larger governance models often leave implicit: how communicative and procedural structures within AI-assisted interaction may influence interpretability, traceability, accountability, and meaningful human oversight and agency in practice.

Within this framing, structured interaction is not approached primarily as a productivity technique, but as one possible component of broader governance infrastructure: one concerned with interpretability, oversight, procedural visibility, continuity, and meaningful human responsibility within increasingly AI-mediated systems.

As these systems become more deeply embedded within institutional and social life, governance may depend not only on what AI systems are capable of doing, but on whether the human conditions surrounding their use remain visible, interpretable, and accountable. In this sense, AI-assisted exchange increasingly becomes a collaborative process in which human participation, procedural responsibility, and interpretive oversight remain necessary components of the interaction itself.

06

Limitations

AXIS remains exploratory. Current observations are based primarily on iterative practical use across multiple AI systems and collaborative exchanges rather than formal large-scale controlled studies. The protocol may suggest improved outcomes or reduced ambiguity in certain contexts, but it does not guarantee them. It does not replace natural language, and it does not alter underlying model cognition. Effects may differ across users, models, institutional environments, and contexts.

The Sclar finding³ is a genuine caution: structural effects in AI interaction are neither universal nor fully understood, and any interpretation of AXIS results should account for this instability. Many of the concepts explored throughout this paper, including intentional clarity, communicative legibility, procedural continuity, accountability visibility, and interaction structure, remain only partially operationalized and require substantially more independent empirical evaluation.

The project should therefore be understood as an ongoing research direction rather than a validated protocol, completed methodology, or formal governance system. Many of the institutional and governance implications discussed here remain theoretical, emergent, and dependent on broader social, technical, legal, and organizational conditions that are still evolving.

As with any communicative or procedural structure, these systems may also be used in ways that reinforce manipulation, asymmetry, coercion, institutional opacity, or uneven power relations. Questions surrounding misuse, interpretation, and governance therefore remain central to the ongoing evaluation of structured interaction systems.

The more fundamental question also remains open: whether explicit interaction structure preserves genuine accountability within AI-mediated exchange, or only its appearance. AXIS does not resolve this. It is a candidate response to the question, not a demonstrated one.

The central question remains open: whether communicative and procedural structure itself may become an increasingly important layer of design, governance, interpretability, and accountability within AI-mediated systems.

07

Ongoing

AXIS remains an ongoing research direction exploring communicative structure, interpretability, governance, and human participation within AI-mediated systems. Many of the questions raised throughout this paper remain open and require broader interdisciplinary evaluation across technical, institutional, legal, philosophical, organizational, and social contexts.

This research did not emerge primarily from abstract theory, but from practical interaction: repeated encounters with ambiguity, procedural breakdown, interpretive drift, and the growing difficulty of maintaining accountability, continuity, and meaningful human oversight across AI-mediated exchanges. The project continues to evolve through ongoing use, observation, collaboration, and critical inquiry.

The increasing integration of AI into social, institutional, and cognitive life may represent not only a technological transition, but a transformation in how humans communicate, coordinate responsibility, exercise judgment, and participate within increasingly mediated environments. Questions surrounding interpretability, oversight, agency, procedural visibility, ethical accountability, and the preservation of meaningful human participation may therefore become foundational concerns rather than secondary technical considerations.

This research proceeds from the position that human participation, responsibility, and interpretive oversight should remain visible and meaningful within AI-mediated systems, particularly as these technologies become more deeply embedded within everyday life, institutional decision-making, and collaborative forms of work and communication. Within this framing, AI-assisted interaction is approached not as a replacement for human participation, but as an evolving collaborative process that may increasingly depend upon clearer structures of shared responsibility, interpretation, and oversight between humans and AI systems.

We are interested in dialogue and collaboration with researchers, institutions, designers, governance specialists, and organizations interested in the emerging role of communicative and procedural structure within human-AI interaction.