My friend AI

The journey into the future leads to the Swiss suburbs. Here, tall rows of buildings shape the image of a grey cityscape where housing and jobs are scarce. People move from one temporary job to the next – those without work earn points through social commitments as part of a government-run credit scheme. And: AI with emotional capabilities is part of the fabric of everyday life, with gadgets such as smart rings that ask how their wearers are feeling.

Researchers from the Institute of Business Information Technology and the Institute of Applied Media Studies have brought this scenario to life – by way of a 20-minute virtual reality experience developed as part of the scientific communication project entitled “Experience the Consequences of Affective Computing – Immersive Science Fiction Prototyping for Responsible Innovation.” Using VR glasses, the experience not only shows what it might be like to live with emotional AI, but also raises fundamental questions about affective computing: will we be friends with AI in the future? And: is this something that we actually want?

To get answers to these questions, the project team had around 400 people play through the VR experience in 2023. The main target group were young people aged between 13 and 29. “To examine friendship as an extreme case of applying emotional AI to, this target group was perfect. Friendship is especially important at this stage of life,” explains Elke Brucker-Kley, one of the co-leaders of the project funded by the Swiss National Science Foundation. “Young people and young adults tend to use technology in a hedonistic way. If something isn't fun, they won't use it.” The researchers hoped that the target group would provide clues on what emotional AI should be like – or should absolutely not be like. However, “the goal was not to generate universally valid findings,” adds co-project leader Thomas Keller. With science fiction prototyping (see box), there is a risk of creating a scenario that is too dystopian or utopian. This is reflected in how participants respond – and therefore also in the conclusions that can be drawn, he adds. “Our primary aim was to engage in dialogue with young people and discuss their attitudes and expectations.”

In order to develop an experience that mirrors their current fears and wishes with regard to AI, interviews were conducted with young people in advance. These served as inspiration for the multi-linear story. Participants have to make decisions about the use of a digital assistance system that becomes increasingly “human”: from a talking smart ring and a tattoo that detects human emotions to “Youman,” a humanoid avatar that becomes a loyal companion in the virtual world.

The majority of the participants chose all of the technological leaps. 57 percent chose to get tattooed and more than 81 percent tested Youman, with 53 percent keeping the humanoid avatar at the end of the story. “Curiosity certainly plays a role in trying out emotional AI,” says Brucker-Kley. Whether the decisions made in real life would have taken a similar path cannot be determined. It also remains to be seen whether the participants kept the avatar Youman because they saw it as a friend – or because they took a liking to the fact that it performed everyday tasks for them in the scenario. “It is possible that they simply wanted a servant,” remarks Thomas Keller.

The constant availability and helpfulness of AI seemed a tempting feature for some participants. According to Brucker-Kley, this became evident in the discussion rounds, which were conducted after the VR experience and evaluated using discourse analysis. “At the same time, the constant presence of the AI gadget and its flawless, invariably helpful behaviour also felt somewhat off-putting.” In a survey conducted directly within the VR scenario, participants also expressed a certain sense of mistrust: they doubted, for example, whether an emotional AI gadget could really fulfil the key criteria for friendship, including trust and honesty. “There was a fear that the AI gadget would share and possibly manipulate the data.”

Despite this scepticism, the share of participants who could imagine being friends with an AI gadget rose to 35 percent after the VR experience, compared to 23 percent before. However, views remained almost unchanged on the question of whether an AI gadget could one day be a better friend than a human: even after the VR experience, 71 percent said no, compared to 76 percent beforehand.

The researchers now want to use these findings in other projects exploring social interactions with AI. One example, says Brucker-Kley, is a project with Aarau Cantonal Hospital in which an AI device advises people with chronic illnesses in a virtual environment. “The discussions with the young people revealed, for example, that they wanted to have a say in shaping the avatar’s appearance – that's something we have taken account of in the new project.”