Over the past year, agentic AI has gained a lot of attention. But many projects have fallen short or become outdated. Companies are now more cautious about adopting physical AI, recognizing the serious risks involved. Experts emphasize the need for clear rules and boundaries to prevent costly mistakes.

Why cautious approaches are essential for physical AI

Physical AI involves machines like robots, drones, and industrial devices that interact with the real world. Unlike software AI, these systems can cause physical harm or damage if they malfunction. Tianlan Shao, CEO of Mech-Mind Robotics, stressed the importance of keeping these machines in controlled environments. He pointed out that robots wielding chainsaws or other dangerous tools should only operate under strict human supervision to avoid accidents.

Today, more than half of companies worldwide are already using some form of physical AI, with expectations that this number will grow to 80% soon. These applications include industrial robots, security cameras, forklifts, and inspection devices. However, progress is faster in controlled settings like factories and warehouses. When these systems are deployed in open environments, the risks and challenges increase significantly.

Real-world deployments focus on safety and efficiency

Discussions at recent events, like the World Economic Forum, show that most physical AI efforts are centered on practical pilots rather than futuristic visions. Francisco Martin-Rayo, CEO of Helios AI, explained that the focus is on deploying AI in areas like logistics, agriculture, energy, and manufacturing. These sectors face urgent labor shortages and need efficiency gains, making physical AI a valuable tool today.

Though slower to develop than software AI, physical AI could have lasting impacts on society once mature. Its influence may reshape industries and daily life, but the journey there will be cautious and deliberate. Experts warn that many hardware challenges—such as power use, mobility, and costs—still hinder progress. We are far from having “ChatGPT for robots,” and hardware limitations remain a significant obstacle.

Despite these challenges, emerging use cases in eldercare, logistics, and automation show promise. The focus at recent discussions was more on building a solid foundation rather than flashy innovations. A major issue is the lack of a standardized development layer. Many companies are creating their own ecosystems, which slows widespread adoption. Without common standards, progress will continue to be slow and fragmented.

Bridging the gap between virtual and physical worlds

Another concern is how well software designed for virtual environments translates to physical systems. Jinsook Han from Genpact pointed out that defining this interface will take time. Questions remain about what physical AI should be allowed to do and how much autonomy it should have. This ongoing debate reflects the cautious approach many experts advocate.

There is no clear timeline for when physical AI will become more mainstream or fully integrated into society. Many believe that significant breakthroughs are still years away. For now, the industry focuses on incremental steps, emphasizing safety, standardization, and practical applications. This cautious path aims to prevent the mistakes that could hinder adoption and cause harm.

Inspired by

• https://www.computerworld.com/article/4133378/with-physical-ai-gunslingers-and-risk-takers-need-not-apply.html

Sources

• mech-mind.com
• deloitte.com
• helios.sc
• bairesdev.com
• deloitte.com