Conference Agenda

Contemporary Artificial Intelligence (AI) engineering, while rooted in technical optimisation, often relies implicitly on psychological premises that have been studied for decades by Dialogical Self Theory (DST). This research suggests that many industry-standard multi-agent architectures already mirror a "society of mind" without explicitly naming it. By recognising these underlying structures, we can reframe technical modularity as polyphony and system governance as reflexive metapositioning, moving AI beyond mere obedience toward genuine ethical deliberation.

The methodology of this study departs from an existing industry framework (Question Crafter and the Graph-of-Thoughts (GoT) protocol) to demonstrate how these "agentic" designs inherently contain dialogical elements that can be significantly refined through DST. The model evolves through four phases: eliciting eight expert "I-positions," implementing polyphonic loops, incorporating "shadow positions" to handle human friction, and a final Grounded Truth validation phase. In this final stage, expert evaluation of 50 cases per iteration measures how the quality and "human truth" of the responses improve as DST concepts are explicitly integrated.

Ultimately, this work serves as a reciprocal reflection: while DST enriches AI's responses and architectures, the field of AI agents provides DST with a unique empirical testing ground. By providing formal mechanisms (such as memory structures and decision rules) AI allows for the operationalisation and measurement of historically interpretive dialogical processes within complex, contemporary sociotechnical environments.