According to technology experts and executives, the latest advancements in AI require meticulous scrutiny before being deployed in real-world applications.
In recent times, agentic AI has emerged rapidly, yet its path has been marked by numerous unsuccessful initiatives and sophisticated, albeit unrefined, technology that businesses are still struggling to integrate. Consequently, it is understandable that tech industry leaders are advocating for a prudent approach to physical AI, given that errors could lead to profound commercial and social repercussions.
Tianlan Shao, CEO of Mech-Mind Robotics, a manufacturer of industrial robots, emphasized, “Clear boundaries…, definitions, and rules are essential.”
Shao stated during a panel discussion at the recent World Economic Forum (WEF) that businesses cannot afford to experiment with tools such as a robot equipped with a chainsaw, unless such trials occur within a controlled setting and under human supervision to prevent errors.
According to Deloitte’s State of AI in the Enterprise study, released last month, over 50% of global companies currently utilize some form of physical AI, a number projected to reach 80% in upcoming years. This includes applications like robots, drones, inspection tools, smart security cameras, forklifts, and various other industrial uses.
Deloitte’s study noted that “Physical AI applications in regulated environments, like factories and warehouses, generally advance significantly quicker than those in uncontrolled, real-world settings, which present substantially higher complexities and dangers.”
Francisco Martin-Rayo, CEO of Helios AI, who monitored technology discussions at WEF in Davos, observed that conversations around physical AI have centered more on practical trials and tangible results than on fantastical “Jetsons-style” robots. Martin-Rayo explained, “The focus has been on implementation within restricted settings such as logistics, agriculture, energy, and manufacturing, where current labor deficits and demands for increased efficiency are pressing issues.”
Martin-Rayo indicated that while physical AI is anticipated to develop at a slower pace than software AI, its eventual progress will have more profound and enduring effects on the operational structure of societies.
Nacho De Marco, CEO of BairesDev, another observer of key discussions at WEF, noted that physical AI faces hurdles beyond experimental phases that could impede its advancement. Currently, there isn’t a “ChatGPT equivalent for robots,” and technological development is hampered by hardware issues like power usage, movement capabilities, and expense.
“A widespread sentiment is that we are still in the nascent ‘floppy disk’ era of [physical] AI,” De Marco commented.
Despite promising applications surfacing in areas like eldercare, logistics, and industrial automation, De Marco described the discussions at WEF as “more fundamental than extravagant, which is, candidly, a positive indicator.” He added, “A unified development framework for physical AI is absent. Each entity is constructing its own ecosystem, which consequently impedes widespread integration.”
Jinsook Han, Genpact’s chief agentic AI officer, raised concerns about the integration effectiveness between virtual software and the physical environment. “I believe defining that interface will require significant time,” Han stated. “The core issue revolves around the extent and scope of tasks we are prepared to entrust to physical AI.”
Han remarked, “I don’t claim to be a futurist predicting outcomes five or seven years from now, but I do believe we are steadily nearing that point.”
According to Beena Ammanath, global head of the Deloitte AI Institute, the foundational elements for physical AI were established over a decade ago, beginning with IoT and sensors, progressing through robotic process automation and data science, and culminating recently in autonomous agent-driven execution.
Ammanath explained, “The core groundwork was put in place approximately 12 to 13 years ago, and this now allows us to integrate more advanced intelligence into that existing framework.”
Physical AI is gaining significant traction, particularly within surveillance and security systems, where smart cameras are equipped to manage alarms by incorporating AI capabilities directly into the devices. Furthermore, numerous collaborative robots are now capable of inter-communication and autonomous decision-making. The application of physical AI in retail environments is also on the rise.
Ammanath observed, “Within retail outlets…, there’s a growing drive to automate processes like returns and point-of-sale interactions, enabling conversational capabilities. These functionalities are invariably supported by some form of large language model (LLM) operating behind the scenes.”
For many years, AI has been a catalyst for industrial processes, demonstrating noticeable productivity improvements, stated Deepak Seth, a director analyst at Gartner. For example, modern automobile manufacturing facilities can operate in complete darkness, as robots handle operations around the clock without requiring illumination for human interaction, thereby reducing electricity costs.
Seth suggested, “The subsequent progression involves making AI more anthropomorphic and integrated into our daily routines, such as a robot powered by AI that can discern your meal preferences and prepare them for you.”
Jinsook Han mentioned that Genpact, a company with a strong foundation in agent technologies, is also exploring advancements in physical AI. Han elaborated, “Our roots are with GE. Given our history from supply chain to manufacturing, we are deeply invested in physical AI, and this has been a long-standing area of focus for us.”
