Exploring how users trust when interacting with a human-machine collaborative system

The trust a user places in a financial service is the primary condition for this user's membership in this service. This relationship's foundation is where the user entrusts part of their money. With the rise of digitalisation, most of these financial services reveal, behind their digital window, the complex meanderings of collaborations between humans and machines working to deliver this service and the resulting advice. In the specific context of fintech, but more broadly in the context of digital services offering to provide advice based on AI or algorithms, we propose to explore how the mental models and social perceptions that a user can construct on how this service is delivered relate to how trust elaborates.

Through a double study on French fintech Mon Petit Placement's users, we investigate the possible dominant perception of machines or humans (AI score) and its relationship with trust. A first cluster analysis study on 41,542 users attempts to describe four clusters identifiable by their behaviour with the application and their way of progressing with the subscription process. A second quantitative study, in the form of an online survey based on 1,536 participants of the first study, feeds this behavioural analysis with an enrichment around the perception of the predominance of humans or machines and the relationship this may have with users' trust in the service. Finally, the open questions of the survey questionnaire reveal interesting learnings, allowing a better interpretation of the quantitative data.

Both studies helped define four user profiles (Curious, Hesitant, Decided, and Expert), which helped understand behaviours with AI score and Trust. The results demonstrate a correlation between human predominance and trust in service and advice provided. Mon Petit Placement provides advice mainly using video media. In this context, we also observed a significant correlation between the usefulness of utilising videos to deliver advice when this service is perceived as predominantly human. These studies also reveal that, for this kind of service, users expect that a human should at least supervise such advice in a human-machine collaboration setting.

Furthermore, users' mental models reveal that, in such a setting, considering machines as the fruit of humans' work, the latter are those whom one must ultimately trust. This would be a key to developing confidence in the advice or outputs provided by a hybrid intelligence.

Finally, this research paves the way for future research on the perception of Hybrid Intelligence, reflected by a Human-Machine Collaboration or a Joint Cognitive system and its impacts on Trust.