AI could transform manufacturing

AI is starting to increase performance across the design-to-manufacturing lifecycle. Dr Ajit Panesar from the Department of Aeronautics, who explores the use of AI in manufacturing sectors such as aeronautics, automotive and energy, explained that its applications range from product design, where it can anticipate manufacturing constraints, through to the design of production facilities, sensor-driven control systems, and finally post-production inspection.

One important use case is the chemical industry, where the Imperial spinout SOLVE Chemistry is using AI and reaction datasets to rapidly find optimal ways to manufacture chemicals. Its solutions are designed in part to help bring new pharmaceutical and agrichemical products to production-scale faster: “Once your patent clock starts ticking, you have 20 years. We’re reducing a step that takes two years – our north star is shortening scale-up to less than a month,” explained SOLVE Chemistry CEO Dr Linden Schrecker.

...Redesign cities

AI and modelling can also be used to optimise urban planning, for example by calculating where to place trees to maximise the airflow needed to disperse pollution. And by combining multiple data sources such as airflow, pollution, and heat, it could in principle be used to design urban environments from scratch.

“You can imagine a system that takes all sorts of information, processes it, and works out new designs for a city,” said Dr Claire Heaney from the Department of Earth Science and Engineering. This is just one application, she said, of new AI architectures that combine the accuracy of physics-based models, which are grounded in physical laws, with the far greater speed of data-driven surrogate models, which are based on in historic data.

...And make weather models interactive

AI has already reshaped weather and climate modelling. Traditional numerical models demanded supercomputers that use as much electricity as a city, but new AI‑accelerated models can deliver comparable performance on a laptop.

This is starting to democratise high‑end forecasting: energy companies can explore how storms or heatwaves might cause line outages, while businesses can generate large sets of extreme‑event scenarios for pricing risk and capital planning. The growing power of these models is changing how we use them: making them more robust to climate-related changes to weather patterns and enabling more interactive ‘what-if’ tools.

“Last time I was a proper jobbing scientist running climate models, my daily existence involved babysitting computers. We’re moving to a different relationship with these models – they’re much more interactive,” said Dr Niall Robinson, a weather and climate modelling expert at NVIDIA.

But domain expertise matters

Current AI tools typically lack the full domain understanding required to perform well in specialist applications, however. Andrew Bean, a senior scientist at Thomson Reuters, said that in the legal profession, for example, AI often fails to be as useful as benchmarking tests appear to predict.

“We found there are a lot of things that look good, but the details aren’t quite right. The citation it gives isn’t the one that lawyers would use – they have to look it up, whereas if you gave the right one they’d know it," he said. Thomson Reuters is working with human domain experts to improve AI performance, aided by a new lab the company has established with Imperial.

The vital need for domain expertise in AI is providing a market opportunity for universities and smaller players, said SOLVE Chemistry’s Dr Schrecker: “We can’t outrun AI companies on AI alone,” he said. “There’s a big trend toward investing in AI that is underpinned by a physical value-add, such as data, and an ability to collect more data. At SOLVE Chemistry, our technology doesn’t rely on a customer at any point giving us their data.”

Human skills are irreplaceable

On a related note, human expertise, along with human traits such as creativity and critical thinking, will be vital in the AI era. Science is a creative pursuit – or should be,” said NVIDIA’s Dr Robinson. “The really employable attributes in five or ten years’ time will be knowing how to ask questions and frame problems.”

This means that students should avoid outsourcing their thinking to AI and should cultivate their own humanistic skills: “The people who will be the leaders in the world in future will be people who can be technical and have a good command of language. I’m a bit worried about only teaching science. You need good language skills, and perhaps philosophy,” argued Professor Sophia Yaliraki from the Department of Chemistry.

While the stakes are high, for those who possess the right skills, work in the AI era could be more rewarding. Sachin Jogia, an AI expert with deep experience across the civil service and private sector, said that coders at technology companies are reporting that AI is already freeing up time for the most rewarding tasks. “Their staff are not saying AI is taking my job. They’re saying it’s allowing me to do more first‑order work, instead of second- or third-order.”

The solution matters more than the tool

It is tempting to see AI as magic dust that can be sprinkled on any problem to improve performance or attract investment. In reality, it is a broad approach with varied applications and benefits.

“There’s no point in denying that this tech is something – something big. But at the same time, we need to be sure that whenever we make statements about what AI can do, we are as precise as possible,” said Dr Francesco Leofante from the Department of Computing. 

Professor Yaliraki agreed: “People talk about the AI, but there is no such thing as the AI – there are different ways of incorporating data and automation.”

A corollary of this is that AI developers should sell the solution, not the technology. “Although AI is a key part of what we do, what we’re selling is efficiency gains,” said Dr Schrecker, “It’s about the problems we’re solving, and the value we’re providing, rather than the exact tool.”

We need the right regulation

Policymakers and regulators need to combine a few different ingredients to build a healthy environment for AI innovation. One is proportionality: “There may be cases where if it does go wrong, the consequences don’t matter that much. With electricity, the consequences can be quite severe. A blackout can lead to deaths,” said Jonathan Thurlwell, Head of Emerging Technologies at the UK energy regulator Ofgem. 

Getting the infrastructure right is also part of responsible AI. Dr Andrew Richards, Director of Research Computing Services at Imperial, described how the university is rebuilding its high‑performance computing to support AI in a more sustainable way: “We’ve decomposed our computer infrastructure so that we can deliver more computational power while using less energy for cooling – that saves money and gives us more to spend on compute.” 

But once again, what matters in policy is not just technical performance, but also human factors. “Do you have the right competencies in place in the organisation, so people know how to deploy the tool effectively?” Mr Thurlwell asked. 

Professor Alessandra Russo, Co-Director Imperial’s School of Convergence Science, warned that we need to avoid normalising ‘good‑enough’ automated decisions in areas where we should still care about accuracy and human judgement. “The big risk that keeps me awake is that implicitly embedding this technology in our society might change our societal values, and this is for me a big problem and risk. We’re seeing the problem with social networks – and there’s no way of backtracking.”

...And deep integration

Harnessing AI positively for industry will require convergence – not just collaboration between a few labs, but deep, large-scale integration across disciplines and institutions.

Professor Mary Ryan, Vice Provost (Research and Enterprise) at Imperial, said: “The challenges we face do not sit within a single discipline – they even extend beyond traditional interdisciplinary boundaries. They require deep integration at scale.”

This perspective underpins Imperial’s School of Convergence Science, which is organised around real‑world challenges rather than academic silos. For industry partners, that convergence model could be what’s needed to tackle challenges that cut across science and engineering, policy, and human behaviour.