Marisa Tschopp

Artificial intelligence has found its way into our everyday lives in a more or less conspicuous way, whether through smart speakers, facial recognition, or the TV program and music selection, all promising the maximum potential for us humans. But in addition to the complex technical and legal challenges associated with developing and deploying AI systems, there is another challenge: humans. Humans have a remarkable tendency to humanize non-human entities from deities to chatbots, that is why we cannot help but treat these digital entities as social actors. This comes with a plethora of opportunities, like more efficient user interaction but also challenges, like manipulating users by humanized design. This is one reason, why "trustworthy AI" is on a rise and debated across the globe. Despite the plethora of expert guidance on the development and implementation of "trustworthy AI," there is still a surprising amount of disagreement about what constitutes user trust in AI - is it the same as trust in a human? Can we, do we or should we trust AI, just like we place our trust in humans? Moreover, the topic of overtrust is also widely neglected. Since the level of trust influences how users interact with technology, overtrust and over-reliance as subsequent behavior - leads to a faulty human-automation relationship: like the Tesla Driver Walter, who unfortunately died as he trusted the “auto-pilot” so much, eventually hitting a barrier, crashing with two other vehicles. It is long overdue that we talk about how to develop an "appropriate" level of trust for better and safer interactions with non-human agents, which are after all - just man-made machines.