Trustworthy, friendly or freaky? How life-like digital humans in voting advice applications can stimulate young voters to inform themselves about politics

In the Dutch political context, the levels of political involvement among young voters (up to 25 years of age) appear to be particularly low, as indicated by diminished political interest, limited political knowledge, and relatively low voter turnout rates during elections. In an attempt to increase the political involvement of young voters, we have developed different versions of a digital human (or virtual human)—a life-like digital character with embodied behaviors—that young voters can interact with to seek information about political issues while using a Voting Advice Application (VAA).

The technology underlying this digital human is a conversational agent (chatbot), similar to the approach taken in previous studies by Kamoen and Liebrecht (2022). They demonstrated the beneficial effects of Conversational Agent Voting Advice Applications (CAVAAs) on users' perceptions of tool usefulness and political knowledge. However, their study focused solely on disembodied conversational agents (i.e., textual chatbots without a digital character), and it involved a heterogeneous group of users.

In our current study, we will compare a disembodied CAVAA to three different embodied versions: one featuring an unknown character, one in which the character is a known political expert (a television host regularly interviewing politicians), and one in which the character is known, but lacking political expertise (a television host of sports programs). These varying levels of embodiment were chosen because a source's expertise and likeability can be considered pivotal factors in individuals’ information processing.

The research question of our study was: What is the effect of different forms of embodiment of a Conversational Agent Voting Advice Application on a) tool evaluation measures (ease of use, usefulness, eeriness, anthropomorphism), and b) political measures (political knowledge, voting intention, proportion of non-substantive answers)?

Method
Our experiment employed a 4 (CAVAA: disembodied/ unknown character/ expert character/likable character) * 2 (young voters’ education level: high/low) between-subject design. First, we developed a disembodied CAVAA for the European Elections (to be) held in June 2024. Thereafter, three digital humans were created and connected to the CAVAA. The characters of the embodied CAVAAs were determined by a pilot study during the Dutch national elections of November 2023.

The experimental study that we will present at ECPR is still to be run, but the set up will be that people from the target group will interact with one of the CAVAA conditions and subsequently fill out a survey in which their evaluation of the tool (ease of use, usefulness, eeriness, anthropomorphism) and political outcomes (political knowledge, voting intention, proportion non-substantive answers) will be measured. Subsequently, they will be interviewed and their conversations with the CAVAA will be content analyzed. This way we can assess whether "embodiment" in general and different forms of embodiment in particular increase the evaluation of the tool and political outcomes.