Can Robots Really Understand Our Emotions?

A recent study explores how robots could read human emotions using advanced vision language models. The findings suggest emotional capabilities are secondary to robots' functional reliability.
Robots are getting smarter and more dexterous, but are they becoming more emotionally intelligent? As machines increasingly share workspace with humans, their ability to understand emotions isn't just desirable, it might soon be essential. A recent study sheds light on this emerging frontier.
Emotional Robots: More Than Just Faces
Seung Chan Hong, an undergraduate researcher at the University of Melbourne, has made strides in teaching robots to read human emotions. His approach uses a vision language model (VLM), similar to large language models like ChatGPT, but geared toward visual inputs. The study, published in May 2023, involved training robots to assess human emotions by observing volunteers as they interacted with them.
The process was comprehensive. Volunteers watched videos of robots handing objects to humans. Their task? Describe the emotions of the people on screen. This wasn't just about facial expressions. Context mattered. A furrowed brow could mean anger, or just concentration.
VLMs vs. Conventional AI
The results? The VLM outperformed traditional AI systems, achieving a score of 0.86 compared to 0.77. Visualize this: a machine not just reading faces, but understanding the entire scene. It considers where a person is, what they're doing, and how they're interacting with the robot.
That's not all. In a second experiment, the VLM-equipped robots interacted with humans, sometimes intentionally making errors. The robots responded with either a pre-scripted apology or an emotionally adaptive one. The latter was preferred by a significant majority, 31 out of 40 participants.
The Limits of Emotional Intelligence
However, there's a catch. Despite their emotional adaptivity, robots still faltered trust. Errors led to decreased trust, regardless of how well they apologized. A robot's functionality trumps its emotional awareness. So, can we ever fully trust a robot that empathizes but fails to perform?
while the VLM matched third-party observations of emotions, its accuracy dropped when compared to participants' self-reported feelings. It turns out, robots aren't mind readers just yet. They see our expressions, but not our inner thoughts. The chart tells the story: Emotional intelligence in robots has its boundaries.
What's the takeaway? While robots learning to read emotions is promising, their primary role remains clear, consistent performance. In the end, what matters most isn't if a robot can tell we're upset, but if it can complete the task at hand. One chart, one takeaway.
Get AI news in your inbox
Daily digest of what matters in AI.