Intelligent technologies portend some risk to users as they behave in an unpredictable manner in complex and unstructured situations. Ighoyota Ben Ajenaghughrure developed trust assessment tools that implement a machine learning trust level classifier model based on users’ psychophysiological signals (i.e vital body organs like brain, cardiac, or skin conductance activities, reactions during episodes of cognitive or affective experiences).
Due to the importance of trust in sustaining the above-mentioned collaborative interaction, several efforts have been made towards the development of real time trust assessment tools that implement a machine learning trust level classifier model based on users’ psychophysiological signals. However, it is unclear to what extent users’ trust can be assessed in real time by using psychophysiological signals, considering that there are several approaches to building a trust classifier model, all of which are chosen at researchers’ will. In addition, prior attempts have not been fully successful. Ajenaghughrure used psychophysiological signals recorded during several controlled experiments designed to elicit users' trust dynamics to develop trust classifier models using machine learning techniques (dynamic, ensemble and unsupervised). The aim was to find solutions to the inherent challenges, such as the different feature selection methods, ensemble methods, psychophysiological signals, and performance in real-time context.
Ajenaghughrure discovered that instability vs accuracy trade-off was an inherent challenge with ensemble trust classifier models, hence a guideline was proposed. It suggests that hybrid feature selection method results in a more stable and fairly accurate model, stack ensemble method results in a more superior trust classifier model, EEG psychophysiological signal is more reliable than other psychophysiological signals for developing trust classifier model and the extent to which trust can be assessed in real-time is not yet at full capacity. The study yielded a results-grounded guide to researchers in the development of a trust classifier model that uses psychophysiological signals to estimate users' trust level in real time (during interaction) to foster symbiotic interaction in user-AI based systems collaborative interactions.
Ighoyota Ben Ajenaghughrure from Tallinn University School of Digital Technologies defended his doctoral thesis “Towards enabling human-computer symbiosis through trust: empirical guidelines for developing psychophysiological models to assess users' trust in real time” on 20 January. Supervisors were Tallinn University Associate Professor Sonia Claudia da Costa Sousa and Tallinn University Professor David Jose Ribeiro Lamas, opponents were Lissabon University Lecturer Hugo Humberto Plácido da Silva and Tartu University Professor Gholamreza Anbarjafari. The doctoral thesis is available in Tallinn University Digital Library ETERA.