The potential of physical AI as a force for good.
For years, artificial intelligence (AI) has been trained to answer questions, process information, generate insights and more recently support decision-making. The recent release of ChatGPT-5 has shown how fast AI is evolving, now operating at a “PhD level” according to Sam Altman, founder of OpenAI. Today, a new frontier is emerging quickly. We are seeing the rise of robotics pushing the technology into a new dimension, the physical world, a shift often referred to as physical AI. For instance, in health care, an area I am passionate about, AI is increasingly making the leap from virtual to physical to be by our side.
From the moment patients enter a hospital until they leave, new advancements are available at virtually every step. Today, disinfection robots and automated sterilization systems can be used to keep hospitals clean. Logistics robots deliver clean linens, supplies and medication to nurses while remote-controlled surgical assistants help with minimally invasive procedures.
Exoskeleton suits can help patients with mobility issues walk and navigate defined spaces with confidence. And in pharmacies, automated dispensing systems and medication management robots are used to fulfill prescriptions with greater accuracy.
Physical AI is transforming every industry, from manufacturing, where robots are already critical to factory lines that assemble devices, to agriculture, where autonomous tractors can till crops and use imaging data to harvest when ready. In retail settings, robots scan grocery aisles daily to give accurate updated inventory and restock shelves overnight. Meanwhile, the logistics industry has introduced robots that are able to retrieve, sort and package items for efficient and safe handling. The automotive industry continues to leverage Physical AI to move from driver-assisted cars to fully autonomous fleets that can be updated remotely.
Now we face a key question: Do people trust physical AI? Only with trust can these futuristic capabilities become the norm rather than interesting use cases.
Why we’re entering a new paradigm in robotics
This world is possible because the physical AI learning curve continues to bend toward faster, better and cheaper developments. Synthetic training environments have become a less costly and less risky alternative to expensive physical experimentation while accelerating improvements. For example, Boston Dynamics’ Spot robot has achieved 87% accuracy in detecting objects in simulation, thanks to the help of synthetic training data from NVIDIA’s Isaac Sim and Replicator. More companies are also turning to simulated environments for competitive advantage; for example, BMW invested €2 billion into a factory powered by a digital twin, aiming to accelerate development and improve planning efficiency by 30%.
Robotics developers, on even the smallest teams, increasingly have access to rich physical world data thanks to world foundation models (WFMs), which offer an advanced starting point for new capabilities. Early pioneers like Figure AI and Agility Robotics are already demonstrating how well physical AI can integrate into different human environments. Major advances are happening in robotics software too: Orchestration models such as DeepMind’s AutoRT have demonstrated how to control fleets of robots across different tasks with limited human intervention. In Europe, fleets of miniature robots are being trialed to help with search and rescue operations, navigating through collapsed buildings and piles of rubble to find people trapped underneath.
Current economic pressures are adding urgency to putting robots to work. If current trends continue, the US manufacturing industry could face a shortfall of 1.9 million skilled personnel by 2033, including industrial maintenance technicians and engineers. Facing such potential shortages, 80% of component manufacturers are planning on increasing automation by the end of the decade.
3 ingredients to get ready for the next wave of physical AI
As physical AI matures and industry adoption grows, we’re helping clients orchestrate these developments responsibly and take the leap from experimentation to enterprise-scale deployment. Just as generative AI fundamentally changed how we work, physical AI will transform how we operate and serve the world around us, helping us do good along the way. There are three key ingredients to enabling physical AI:
1. Build your foundation with AI-ready data
Physical AI runs on data and is first simulated on data before you ever even test a robot in a physical space. If the data isn’t accurate or reliable, the system won’t perform well — or, as we say, “garbage in, garbage out.” For instance, robots in biotech labs rely on accurate sample labels and up-to-date protocols to automate the screening of compounds. But if data is incomplete or mislabeled, errors can occur, compromising research integrity.
AI-ready data (AIRD) ensures that robots can clearly understand and reliably respond to their real-world environments. Physical AI must have AIRD to consistently perform tasks safely and effectively at a high quality. If the biotech lab in our example maintained a centralized system of data with standardized formats and routine audits, robots could operate reliably, producing results that researchers and regulators can trust. Lastly, AI-ready data must be maintained and governed in a way that ensures timely and secure access, but also the ability to navigate the complexity, breadth and speed at which the world around us changes.
2. Embed trust in the foundations of AI
Think about AI landing your plane? It is already being done. Through cameras and sensors, AI can see runways, understand the impact of weather conditions and make necessary adjustments. But in the cockpits of commercial airlines today, digital copilots are assisting humans, not taking over controls entirely.
Trust must be incorporated into the foundation of AI systems, and that is just as important for physical AI that shares space with human teams. Today, that involves keeping a human in the loop, like a pilot, for ongoing oversight of safety and compliance, so that machines augment rather than undermine human well-being.
Agile governance and responsible oversight are the cornerstones of trust. Through extensive experience in autonomous vehicles and robotics compliance, our teams have observed firsthand that clear governance fosters trust by assessing safety in a structured way and keeping stakeholders engaged. Effective governance also spotlights how AI systems make decisions through principles of explainability. In this way, risks can be mitigated, and employees can gain confidence about working with physical AI and the benefits it delivers.
3. Be clear on the values that matter
Businesses no longer have the luxury of multiyear transformations with unclear ROI. Organizations should prioritize use cases where physical AI addresses real pain points: Improving safety metrics, reducing downtime, increasing throughput or lowering costs.
Consider a hospital network that deploys AI-enabled robotic assistants to automate routine but critical tasks such as delivering medications, disinfecting patient rooms and restocking supplies. Leadership can monitor robots’ impact in near real time by establishing clear key performance indicators, including faster medication delivery times, fewer hospital-acquired infections and improved inventory accuracy. Success can be measured in how much more time clinicians can spend with patients or how many errors have been prevented, which can lead to the ultimate value metric — the ability to save lives.
Physical AI that transforms for good and with trust at the core
If AI is the brain, physical AI is the body. Just as large language models have transformed knowledge work, robots stand to transform how we live, from how food and medicine are delivered to how surgeries are performed.
With AI-ready data at its core, trusted governance mechanisms and a focus on values that make the world a better place, physical AI can be a force for good, driving sustainable growth, strengthening resilience and ultimately advancing humanity as a whole.
The views reflected in this article are the views of the author and do not necessarily reflect the views of the global EY organization or its member firms.
This article is published as part of the Foundry Expert Contributor Network.
Want to join?
