Tại sao một số robot lại gây cảm giác rợn tóc gáy? Chúng tôi cố tình chế tạo một con để tìm hiểu điều đó

Robots no longer exist only in popular culture or science fiction. While we might be familiar with the likes of WALL-E, Baymax, the Terminator and Ultron from the big screen, versions of them are increasingly entering everyday life.

If you visit the University of Waikato campus you may even encounter a robot in the wild. Cruz and Marshmallow are robotic student ambassadors who, as well as giving out snacks and treats, generally delight visitors at community and outreach events.

Some people stop to take photos, some wave, often kids will “herd” the robot to see if it can escape. But others can have quite different responses – they might be wary, suspicious or simply find Cruz and Marshmallow creepy.

Why is it the same robot can inspire a sense of fun in some people but discomfort in others? Why are some robots perceived as creepy and others cute? We wanted to find out, so designed a survey to measure how people reacted to different robots.

This isn’t just about robots, either. With the rise of artificial intelligence and large language models, chatbots, avatars and human–like technologies are everywhere. They are routinely being used in schools, hospitals, airports and restaurants.

We need to understand if these contexts are appropriate, fit for purpose and whether people accept, trust or reject these technologies. More broadly, it’s about how our everyday world will be shaped by these future robot companions (or overlords, depending on who you ask) .

This matters because robots increasingly occupy spaces where people cannot simply ignore them. A robot that seems harmless or engaging to its designers may feel intrusive, unpredictable or threatening to someone expected to interact with it.

Measuring perceived creepiness

To understand how appearance can shape perception, we need look no further than how Disney has developed the look of its fictional heroes and villains.

Characters such as the Disney princesses adhere to what has been called the “baby schema” – large eyes and head, rounded cheeks and small noses that remind us of babies. Because of that, we perceive them to be cute.

In contrast, Disney villains are angular, usually tall and imposing, and are often drawn to look more realistic – Claude Frollo from the animated Hunchback of Notre Dame, for example. But is this the same thing as being creepy? This is the question we wanted to answer.

In 2021, expert in human–computer interaction Paweł Woźniak and his team asked this same question about technology in general. Their “perceived creepiness of technology scale” measures how creepy someone finds a technology on their first encounter across three aspects:

undesirability – “I don’t want this technology around me”

implied malice – “does this technology have bad intentions?”

unpredictability – “I’m not sure what this thing is going to do next”.

Problem solved, right? We could apply the scale to social robots such as Cruz and Marshmallow and work out why some people find them creepy while others do not. Well, not exactly.

We asked five experts in human–robot interaction to evaluate six commercially available social robots to see whether the creepiness scale helped accurately assess them.

We found a robot can score as “non-creepy” despite having characteristics people might find concerning, such as giving away personal data or actively trying to influence a decision.

More interestingly, when the respondents were asked to rank the six robots from least to most creepy there was no consensus. Notably, a robotic dog ended up on both the most creepy and least creepy lists, highlighting how subjective this can be.

Spike, the deliberately creepy robot

It seemed perceived creepiness was not as simple as identifying particular features of a robot and measuring their effect. The person encountering the robot mattered too.

So we turned the question around. Rather than looking at existing robots and asking whether they were creepy, we went back to the research literature, identified the characteristics that previous work suggested might make a robot creepy, and deliberately built them into one robot.

We named him Spike. He is neither clearly human nor animal nor machine. This hard–to–classify appearance – as well as spikes, unusual “facial” features, dark colours and unpredictable movements – should make a robot “creepy”.

Using the “human–robot interaction evaluation scale”, we asked 104 participants from a variety of backgrounds to describe the level of discomfort Spike caused. Along with their responses to other characteristics – Spike’s perceived sociability, animacy and agency – this helped us understand which feelings most influenced perceptions of creepiness.

However, the results of our study weren’t as clear cut as we expected. While some described Spike as angry or unsettling, others thought he was cool or even cute.

Rather than uncovering a formula for creating a creepy robot, the study showed that perceptions of creepiness are highly personal and subjective, and vary from one person to the next. If our research tells us anything, it’s that there probably isn’t a universally creepy robot.

As huge investment in AI and robotics continues, we need to ask what roles we want these robots to play in our communities and everyday lives. We argue the people living alongside them need to be part of deciding what that future looks like.

The authors do not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and have disclosed no relevant affiliations beyond their academic appointment.