Contestable LLM Agents for Sustainable Behaviour Change : A Quantitative Argumentation Model based on the Theory of Planned Behavior

Sammanfattning

The fashion industry is a well-known accelerator of climate change. A promising solution to decrease its environmental impact is through prolonging the lifetime of our clothes and engaging in second-hand clothing consumption. However, even though consumers generally want to act more sustainably, there is a clear discrepancy between their attitudes and actions, where consumers tend to engage in unsustainable consumption regardless of their environmental beliefs. Thus, a change in consumer behaviour is needed. A well researched behaviour change intervention is the use of AI-supported chatbots, a technology with increasing potential for behaviour change support through recent years advances in Large Language Models (LLM’s). However, using AI to promote certain behaviours – even for a good cause – raises both ethical concerns and risks, such as manipulation, bias, and spread of misleading information. To mitigate these risks, the current study proposes to combine the Theory of Planned Behavior (TPB), Quantitative Bipolar Argumentation Frameworks (QBAFs) and LLM agents, introducing a psychologically grounded hybrid reasoning model, called the TPB-QBAF Model for Positive Behaviour Change Support; thereby ensuring a responsible AI-supported behaviour change intervention, in the area of second-hand clothing consumption. In closer detail, the study demonstrates how TPB-factors, relevant for second-hand clothing consumption, may be empirically derived and synthesized to inform a QBAF-based argumentation model. It further proposes that a TPB-QBAF model should satisfy both classical QBAF argumentation properties and TPB-specific properties, and suggests the integration of a set of Responsible Design Requirements, to support contestable and positive behaviour change promotion – thereby demonstrating a concrete and implementable approach to responsibility and contestability, in LLM-facilitated behaviour change interventions.