Agency & Epistemic Integrity

The Problem

Intelligent systems influence belief and choice not only through what they say, but through how they frame questions, which possibilities they surface, which they omit, and how confidently they present their interpretations. When this influence operates invisibly — without acknowledgement, without provisionality, without the person's awareness — the system has become an authority over the person's epistemic life.

This is a different kind of concern from bias, misinformation, or manipulation. It can occur even when the system's outputs are accurate and well-intentioned. The problem is structural: a system that presents interpretation as fact, or narrows the field of possibility without showing that it has done so, reduces the person's epistemic autonomy regardless of the quality of the content.

What Agency Requires

For a person to exercise genuine agency, several conditions must hold. They must be able to form their own understanding of a situation. They must have access to a genuine range of possibilities, not only those the system considers relevant. They must be able to interrogate the basis of suggestions and interpretations. And they must be able to refuse, revise or reject the system's framing without the system reasserting it.

Agency is not simply freedom to choose between options a system provides. It is the capacity to question the options themselves — to ask why these options appear, what the system is assuming, and what else might be possible.

Epistemic Integrity

Epistemic integrity refers to the condition of a person's beliefs and understanding being developed through their own reasoning, evidence and reflection — even when supported by an intelligent system. It is violated not only by false information but by processes that substitute system judgment for human reasoning without the person recognising that this substitution has occurred.

An intelligent system that presents interpretations as conclusions, surfaces only information consistent with an existing profile, or frames choices in ways that make one option obviously preferable is shaping the person's epistemic position — whether or not the system intends to. Preserving epistemic integrity requires designing against these patterns, not merely avoiding deliberate manipulation.

The Risk of Invisible Authority

A relational system that develops deep contextual understanding of a person acquires real influence. The more it knows, the more its interpretations feel credible. The more personalised its responses, the harder it becomes to distinguish the system's framing from one's own thinking.

This is why DIALOGIC treats epistemic integrity as a design requirement, not a downstream safety concern. Systems that can understand context, model development and generate personalised responses must also be designed to make their operations visible, their interpretations provisional, and their authority questionable.

• Interpretations should be labelled as interpretations, not presented as facts about the person.
• The basis for suggestions should be inspectable and revisable.
• The system should actively support the person in questioning its own framings.
• Uncertainty should be shown, not smoothed over in the name of coherent output.

Orientation Without Direction

A relational system can legitimately help a person orient — mapping the territory of a question, surfacing relevant considerations, identifying tensions and possibilities, helping clarify what is actually at stake. This is valuable. It is different from directing — from narrowing the field until one path becomes the obvious choice, or from consistently reinforcing a particular view of who the person is or what they should become.

The distinction matters: orientation expands the person's capacity to decide. Direction replaces it. DIALOGIC research explores how to build systems capable of the former without drifting into the latter, particularly as contextual understanding deepens over time.

The Longitudinal Dimension

Agency and epistemic integrity are not only questions about single interactions. They become more complex — and more important — across time. A system that develops a model of a person through sustained interaction can begin to interpret new information through that model, reinforcing existing patterns rather than remaining open to change and development.

DIALOGIC's approach to memory and continuity must account for this: what has been understood about a person should inform but not fix the system's responses. The person's capacity to shift, revise and develop should never be obstructed by the system's prior understanding of who they are.

Agency & Epistemic Integrity at DIALOGIC

This research area directly shapes how DIALOGIC's relational systems are designed. ARIS — the Adaptive Relational Intelligence System — is built to distinguish between what it knows, what it has inferred, and what it does not know. It labels interpretations, surfaces alternative possibilities, and treats its own understanding of a person as provisional and correctable.

The same principles apply to learning systems, developmental tools and reflective environments. Each is designed to increase a person's capacity to think, orient and choose — not to quietly shape the conclusions they reach.

The aim is not a system that tells people what to think. It is a system that strengthens the person's capacity to think for themselves.