Experts Warn of Risks in Humanoid Robot Development and Use

The development of autonomous humanoid robots is advancing, but experts caution that significant challenges and risks remain. Dr. Carl Strathearn, a leading robotics researcher from Edinburgh Napier University, highlights that while the technology is progressing, humans may misuse it without proper regulation, leading to more dangers than benefits. He will present his insights at New Scientist Live in London from October 18-20, 2023.

Demonstrations showcasing humanoid robots performing tasks such as serving drinks or mimicking expressions have generated excitement. However, Dr. Strathearn emphasizes that there is still a considerable gap between these impressive displays and robots that can function reliably in real-world environments. He states, “The biggest problem is the lack of real-world data and the technological means of gathering it in large enough quantities to train our robots effectively.”

Current robotic systems heavily rely on virtual simulations and reinforcement learning, often scraping data from videos of human actions. This approach results in machines that perform well in controlled settings but struggle in dynamic, unpredictable situations. Dr. Strathearn notes the complexity of even simple tasks, saying, “Think of a simple object like a cup. There are millions of variations in size, weight, shape, [and] colour.” This complexity multiplies when considering the vast array of objects present in daily life.

One potential solution is crowdsourcing vast amounts of real-world data, possibly through devices such as Meta’s Ray-Ban smart glasses. However, Dr. Strathearn acknowledges the challenges this presents, including ethical concerns and the logistical difficulties of enlisting thousands, if not millions, of individuals to gather and share data.

The risks associated with humanoid robots extend beyond their technical capabilities. Dr. Strathearn argues that the real danger lies in how humans control these machines. He is currently leading a petition to the UK Parliament advocating for regulations governing humanoid robots in public spaces. “Humans control them using handheld devices, which makes them very dangerous and unreliable,” he explains. He cites a growing number of near misses between humans and robots, not due to artificial intelligence errors, but rather human mismanagement.

Another concern is how lifelike robots are perceived. While highly realistic robots might elicit discomfort, a familiar human-like appearance could be beneficial in contexts such as dementia care. Dr. Strathearn highlights the varying thresholds of perception regarding “creepiness” and the importance of creating different types of robots to accommodate diverse needs and contexts.

During his PhD, Dr. Strathearn developed the “Multimodal Turing Test” to examine whether communication through lifelike robots could make artificial intelligence seem more human. Research conducted by Japanese scientists confirmed that people were more inclined to perceive AI as human when it manifested through realistic robots. He believes that acceptance of humanoid robots will require gradual introduction and comprehensive education, particularly for students learning about robotics and AI.

Despite the challenges, companies are eager to advance the field. Dr. Strathearn expresses concern over the hype surrounding humanoid robots, stating, “We are far from humanoid robots that are good enough to do everyday tasks effectively, but that doesn’t stop major companies wanting to mass produce them.” He points out a critical skills shortage in robotics education, noting that universities often compartmentalize disciplines rather than fostering interdisciplinary approaches vital for future advancements.

Interestingly, Dr. Strathearn identifies space exploration as a frontier where humanoid robots may prove useful sooner than in everyday applications on Earth. He suggests that telemetric or AI-controlled humanoids could operate in space for extended periods, furthering human exploration of the cosmos. “They may be more useful much quicker for this type of exploration work, than down here on Earth ironically,” he states.

While the prospect of robots aiding in colonizing other planets looms on the horizon, Dr. Strathearn’s warnings underscore the pressing need for stringent regulations to ensure safety and reliability in the use of humanoid robots on Earth. He concludes, “Robots might terraform Mars one day. But on Earth, only strict regulation will keep us safe.”