Critical principles to keep in mind as you start your artificial intelligence journey.

The pace at which AI is advancing is dizzying. Nearly four years ago, when OpenAI launched ChatGPT, we were in awe that a chatbot could do everything from distilling complex scientific concepts into quick summaries to writing college essays to penning haikus.

Now, AI is writing code that’s phenomenally good and is capable of detecting security vulnerabilities in nearly every operating system and browser. Keeping up with digital AI, which is most of what has occupied our cultural zeitgeist and enterprise adoption to date, is challenging.

Physical AI, which refers to AI that operates and interacts with the physical world, has the potential to be even more transformative and it’s evolving just as quickly. This is the AI we see in industrial settings across manufacturing, mobility and energy, in self-driving cars, in surgical robotics and in factory and manufacturing sites.

Physical AI has been an emerging concept for several years, but it’s starting to move from concept to reality. And big companies are taking note, with Nvidia, Google and Tesla all investing heavily in physical AI.

Industrial settings are perhaps the sector where physical AI has the greatest opportunity for disruption. For example, physical AI-powered robots are being deployed to perform real-time quality control and inspections on an assembly line, reducing product defects. Cobots are handling the dangerous or repetitive tasks that we don’t want humans to do. Physical AI is turning equipment management into a proactive rather than reactive exercise by predicting failures before they occur, reducing time-consuming and expensive downtime. And this comes at a pivotal time, with more manufacturing shifting to local markets, leveraging technology for efficiencies will be critical, especially given the anticipated labour shortages for new factory jobs.

But just because physical AI promises to carry all of these benefits doesn’t mean it is easy or analogous to how we use AI chatbots. For industrial leaders who want to turn this promise into tangible, real-world impact at scale, here are three critical principles to keep in mind as you start your physical AI journey.

Optimise for brownfield environments
While a greenfield facility offers the luxury of building an AI-native operation from the ground up, the reality for most manufacturers is a landscape of existing brownfield facilities. As a result, any physical AI solution must seamlessly integrate with (and enhance) existing, often decades-old industrial infrastructure. This requires robust architectural flexibility and backward compatibility, focused on unlocking new value from existing assets without costly “rip and replace” scenarios.

For manufacturers, the path forward is strategic and incremental. Start by mapping all of your current manufacturing processes. From there, identify one or two high-ROI but relatively low-stakes processes where AI could step in – think quality inspection or handling hazardous materials. Focus on noninvasive integrations by prioritising mobile or modular solutions (like advanced cameras or wheeled robots) that can be overlaid or bolted on to existing machinery and equipment, boosting capabilities without disruptive infrastructure overhauls. Start small; don’t try to insert too much AI at once. Be highly selective, run limited pilots, and allow them to learn without interrupting core operations.

Build an environment that’s interoperable by design
A factory floor is a dynamic, complex ecosystem, built over years with machinery from diverse OEMs. Many of these machines operate in their own proprietary protocols (“language”). For physical AI to truly deliver on its promise, it must be able to seamlessly speak to and draw data from all equipment, regardless of the equipment manufacturer or technology.

As a manufacturing leader, when evaluating physical AI solutions, it’s imperative to prioritise OEM-agnostic interoperability. Look for solutions that can handle this complexity. This approach will allow manufacturers to maximise existing investments and avoid “rip and replace,” prevent vendor lock-in while maintaining control of data, leverage a diverse set of best-of-breed tools, and achieve a unified view of all operations.

Prioritise edge AI for real-time safety and control
In a manufacturing plant, milliseconds can often mean the difference between smooth operations and a costly incident. For physical AI to meaningfully impact a manufacturing facility – safely controlling machinery, optimising processes, or ensuring quality – it demands real-time/near real-time decision-making. This means bringing the intelligence directly to the factory floor, processing data at the edge where it’s generated.

As a manufacturing leader, prioritising edge-native processing is important because it directly impacts the bottom line, safety, and operational resilience. This will be augmented by intelligence in the cloud. Leverage solutions that offer ultra-low latency for critical actions like collision avoidance so that physical AI can react with near-instantaneous reflexes. Additionally, ensure that your systems can operate even in harsh, intermittently connected environments to prevent downtime. Finally, look for options with efficient local data processing capabilities. This approach minimises bandwidth consumption and cloud costs, while also strengthening data security and privacy by keeping sensitive intellectual property within your facility’s control.

With as fast-moving as the tech industry is, it’s easy to see the latest tech trend as nothing more than a fad. But physical AI is a genuine paradigm shift, and one worth paying attention to. And with intensifying calls for onshoring and reshoring, physical AI may be less of a “nice-to-have” and more of a “need-to-have.” Not only does this have the capability to make manufacturing faster and more efficient, perhaps more importantly, but it has the potential to make the daily lives of millions of manufacturing workers safer and better.

This is the direction manufacturing is heading – are you going to be a leader or a follower?

This is the viewpoint of Frank Antonysamy, Chief Growth Officer, Hitachi Digital and it first appeared in Advanced Manufacturing.