Robots, not chatbots, will realise AI’s potential

Unlock the Editor’s Digest for free

Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.

Since the launch of OpenAI’s ChatGPT three-and-a-half years ago, the world has grown familiar with large language models. The technology is speeding up and augmenting activities ranging from life admin and research to medical diagnoses. AI’s true potential, however, extends well beyond clever chatbots.

Beyond generating digital content, machines fitted with AI can independently carry out physical actions in the real world using cameras and sensors. Examples of applications include intelligent equipment, autonomous vehicles and humanoids. They use “world models” — systems that understand how objects move, interact and respond, as opposed to LLMs, which are more akin to predictive text engines — and are trained using real and simulated data.

In the service-driven economies of the rich world, it is perhaps unsurprising that much attention and investment has been absorbed by the application of AI to cognitive tasks. But leveraging the technology in the physical world — particularly on the factory floor — is likely to be the greater economic prize.

First, the potential productivity improvements could be significant. “LLMs may deliver faster near-term gains, but in theory physical AI can have the larger long-run productivity upside because it targets physical bottlenecks,” says Daniela Rus, director of the Massachusetts Institute of Technology’s Computer Science and Artificial Intelligence Laboratory. “After all, most working hours are still spent moving atoms, not bits.”

Developed nations in Europe and Asia have already made strides integrating robots into assembly lines to drive higher growth. But AI enables robots to also learn and adapt in complex production, construction and logistics environments, which can boost efficiency and quality, and slash changeover times between different products.

Early deployments illustrate the potential. Foxconn, which produces intricate electronics including iPhones, found that vision-guided, self-adjusting robotic arms improved its assembly cycle times by up to 30 per cent, while reducing error rates by 25 per cent.

Adoption by Amazon in a US warehouse found that package-carrying robot fleets fitted with AI have learned to navigate around moving humans and obstacles, cutting their travel time by 10 per cent. (This video demonstrates how physical AI works in industrial settings.)

Next, AI-powered robotics could alleviate labour shortages in rich nations. University-educated talent is in ample supply for knowledge-intensive services jobs, which now appear to be on the frontline of LLM disruption. But manufacturers frequently cite a lack of workers as a constraint in production.

In a 2026 survey by consultancy Capgemini, more than 50 per cent of industrial executives said labour shortages, costs and regulation would be among their top five reasons for adopting physical AI. (Autonomous machines can also conduct bespoke repairs in environments and conditions that are hazardous to humans.)

Political tensions over immigration policy, ageing demographics and shifting attitudes towards physical work raise the long-term case for investing in AI-enhanced robotics.

Though the technology will displace some jobs, it will also create higher value-added roles on the factory floor, where knowledge of manufacturing processes is still needed to supervise, reconfigure and train robots. (In rich nations, where anxiety about AI’s impact on jobs is rising in dominant white-collar work, a shift towards embedding the tech into manufacturing might be relatively less politically disruptive.)

Finally, physical AI complements efforts by governments to boost domestic supply-chain resilience in critical industries. Indeed, multitasking machines could reduce the need for large assembly lines, which means production can take place at home at a lower marginal cost.

There is, then, a clear economic and political case for rich-world policymakers to place greater emphasis in supporting the adoption of physical AI in industry. China has made significant advances in applying the technology to the real world and Beijing’s latest five-year plan includes a strategic pivot towards “embodied AI”.

The transition will take time in the west. Capital, further developments in world models and vast quantities of training data are required. The necessary supply chains to build robotics hardware are essential, too. And factories and warehouses need redesigning. But venture capitalists are increasingly seeking out investments in robotics.

There are, of course, broader applications of embodied AI. Use of autonomous vehicles, such as robotaxis, is growing. Humanoids are often touted as the future of social care and household chores. And use cases in surgical robotics look promising.

Yet manufacturing is likely to offer the fastest route to scale. Relative to roads, homes or hospitals, factories are more contained areas that can be designed around machines. Rather than building technology that mimics humans, firms can create specialised robots to achieve levels of speed, precision and endurance well beyond human limits.

LLMs have demonstrated AI’s ability to process information. But for rich nations seeking greater growth and economic resilience, the bigger opportunity lies in applying that intelligence to production. After all, the greatest technological transformations tend not to replicate what humans can already do, but enable what they cannot.

Send your thoughts and rebuttals to [email protected] or on X @tejparikh90.

Food for thought

Did smartphones contribute to the fall in fertility rates? This paper looks for causal evidence.

Free Lunch on Sunday is edited by Harvey Nriapia