AI Accountability & Governance

Trust in the Age of Collaborative Conjecture

Reasoning with AI instead of delegating to it

Artificial intelligence is often described as a system that produces answers. In practice, its most valuable role may lie somewhere else entirely: assisting the human process of reasoning itself. Rather than delivering finished conclusions, AI systems can participate in exploratory dialogue — where ideas get proposed, challenged, and refined rather than simply handed over. This is what this research programme calls Collaborative Conjecture: a human–AI dialogue where hypotheses evolve through iterative exploration rather than arriving pre-finished.

Reasoning as Exploration

Most knowledge actually emerges through tentative proposals rather than final answers — scientists formulate hypotheses, engineers test possible designs, policy analysts explore competing interpretations before settling on one. AI systems can genuinely accelerate this exploratory phase by generating perspectives, counterexamples, and alternative explanations. Their real value lies less in delivering conclusions than in expanding the actual space of possible conjectures worth considering in the first place.

Dialogue Rather Than Delegation

Collaborative conjecture is fundamentally different from simple delegation. Delegation treats AI as a tool that produces a finished output. Collaborative conjecture treats the interaction as an evolving reasoning dialogue, where human and AI alternate roles:

• The human proposes an idea
• The AI expands the space of possibilities
• The human critiques the interpretation
• The AI surfaces implications the human hadn't considered yet

Each exchange clarifies the reasoning structure itself, rather than just producing another draft to react to.

The Problem of Ephemeral Insight

Dialogue alone doesn't preserve reasoning across time — insight produced during a conversation can simply disappear once the session ends, and without structured records, reasoning stays ephemeral no matter how good the exchange felt in the moment. That gap only gets more significant as AI accelerates how fast intellectual exploration can happen in the first place.

From Dialogue to Record

Preserving reasoning continuity means exploratory dialogue has to produce durable reasoning artifacts, not just a transcript. This research programme captures that as structured Pay-It-Forward Records, documenting the actual trajectory of reasoning behind a conclusion as it emerges — not just the conclusion itself once it's finished.

Trust Through Visibility

Trust in human–AI reasoning doesn't come from believing the machine. It comes from the reasoning process itself staying visible and inspectable — where assumptions can actually be examined, interpretations challenged, and errors corrected before they compound into something harder to fix — the same discipline this research programme calls Audit & Repair, applied to a conversation instead of a formal decision. Seen this way, collaborative conjecture doesn't undermine institutions. Done properly, it strengthens them, because it makes the reasoning behind a conclusion something other people can actually stand behind rather than just accept on faith.

The most valuable role of artificial intelligence may not be answering questions for us. It may be helping us ask better ones.

— Arne Mayoh

Part of the Audited State Research essays.