Taking robots at face value

We found that the uncanny valley effect experienced by participants was similar in both brief and longer exposure conditions. This suggests that when making sense of what they see, people consider artificial faces as eerie early in the process even if the faces are shown briefly. This feeling of unease seems to develop almost instinctively and instantaneously.

These findings shed light on how robots, androids, virtual characters and 3D AI assistants can be designed to appear less unsettling. An artificial face that has either a natural human-like appearance or a more robotic appearance is more likely to be positively perceived, and does not fall into the pitfall of the uncanny valley.

THE FACE OF TRUST

Another factor to address in human-robot interactions is users’ trust in automation technologies, such as collision detection systems in vehicles. People need to trust machines and systems in order to effectively use them.

One way to gauge users’ trust in automation is by studying their facial expressions, which are made up of different movements in the muscles of the face. For example, to form a smile, we lift our cheek muscles – which creates small wrinkles around our eyes – and pull up the muscles at the corners of our lips. Together, these facial movements express happiness.

Our team sought to determine if specific facial expressions and movements could reliably indicate distrust. We examined various facial expressions of distrust and identified several combinations of ffacial muscle movements. These combinations were compared to the muscle movements involved in expressing the six basic emotions: anger, disgust, fear, happiness, sadness and surprise.

We found two main expressions of distrust indicated largely by the movements of the eyebrows and eyes. These two expressions are different from those of the six basic emotions based on their unique combinations of facial muscle movements.

This discovery suggests the potential of analysing facial muscle movements to detect distrust in real-time, which could be useful in developing automated systems that need high levels of users’ trust.

For instance, users of an autonomous driving system have varying levels of trust in it. If the system’s driving behaviour, such as for lane changing, is too aggressive, some drivers might become uncomfortable with the system and even distrust it. This can significantly impact the user experience and the driver’s ability to operate the vehicle safely.

However, if the system is able to detect the user’s trust level from facial expressions, this data could be used to improve and adjust the system’s driving style to boost the driver’s trust and acceptance.

BETTER HUMAN-ROBOT INTERACTIONS

These two studies highlight the complexity and nuances of how people perceive artificial faces and automated systems. They underscore the importance of considering people’s visual and emotional responses when designing artificial entities and systems.

As we continue exploring how people understand faces, we aim to contribute towards creating more effective and empathetic human-robot interactions to improve users’ experiences and how advanced technologies are perceived and accepted in society.

---

Assoc Prof Xu Hong from NTU’s School of Social Sciences is a psychology researcher who studies the neural mechanisms of how people perceive what they see, as well as their applications in real life and human-centric artificial intelligence (AI) systems.