Look closely at almost anything built by nature and you find a material that does more with less. The strongest spider silks are stronger than steel and tougher than almost any other fibre, natural or man-made.[1] Bone repairs its own cracks.[2] A lotus leaf stays clean without detergent.[3]
Our built world works differently. We mine, heat, refine and eventually throw away. Our economy is linear. Cement alone accounts for around 8% of global carbon dioxide emissions,[4] and the bitumen that holds our roads together is a by-product of crude oil.
As negotiators gather in Antalya for COP31, the Turkish presidency has billed it as the “COP of the Future”, focused on moving from consensus to concrete action, no pun intended.[5] Its implementation targets include raising the share of circular materials in manufacturing to at least 15% by 2035, halving the growth in waste, and cutting the energy intensity of buildings by at least a quarter.[5] All of them depend on the materials we make, and this is where artificial intelligence can help.
Nature, the original materials engineer
Silk is protein, wood is cellulose and lignin, shells are mostly chalk. What makes them remarkable is how those building blocks are arranged, from the scale of atoms up to the whole organism. They are made at ambient temperature from local resources, and they break down again when they are no longer needed.
Copying that is hard, almost impossible, but it is still a source of inspiration. The link between a molecule and how a material behaves is buried under layers of structure, and testing every candidate in the lab is slow and expensive. Bio-based ingredients add a further twist: an oil made from waste wood or algae is not one molecule but thousands, and its make-up changes from batch to batch.
Where AI comes in
COP31 names technological innovation among its key themes.[6] In my group at King's, we combine multi-scale modelling and simulations with machine learning to design materials on a computer before anyone goes near a lab bench.
Asphalt cracks as the bitumen in it ages and stiffens, and small cracks grow into potholes. Working with Swansea University and Google Cloud, we used AI and atomistic simulations to understand that ageing at the molecular level, and to develop a bio-based asphalt that heals its own cracks, borrowing an idea from how living tissue repairs itself. Tiny natural spores filled with recycled oils release their contents as a crack form, softening the bitumen so the crack closes.
The same approach runs through our other work. We build data-driven molecular models of oils made from waste biomass, so they can replace fossil ingredients in roads. We model cellulose, the main structural component of plants and an almost inexhaustible raw material,[7] to make water-repellent coatings inspired by the lotus leaf.[3] And in the EU-funded SATISPHACTION project, we simulate more sustainable polymers, such as plastics that bacteria make naturally, to help design packaging that can be recycled rather than dumped.
Why it matters for the climate
This work speaks directly to three themes of the COP31 Action Agenda: climate-resilient cities, zero waste and green industrialisation.[8] Longer-lasting roads and buildings need less maintenance, saving energy and raw materials. Ingredients made from waste reduce our dependence on fossil feedstocks, and some, like biochar, can lock carbon away. And materials designed to break down or be taken apart keep resources in use: globally, only 9% of plastic waste is successfully recycled.[9]
AI speeds all of this up. Moving a new material from discovery to market can take 20 years or more.[10] Climate targets for 2035 and 2050 do not leave time for decades of trial and error.
What policy can do
Nature-inspired materials will not solve climate change on their own, and AI is no magic wand. Computing carries its own energy cost, and access to large computers is unequal. We are tackling that too: compact “embedded” molecular representations developed in my group cut the memory and disc space some AI models need, with comparable accuracy.[11] That means less energy, and lets researchers with modest computers take part.
But if COP31 is to be about implementation, three steps would help. Public procurement, which pays for most of our roads and buildings, could reward durability and low embodied carbon rather than the lowest upfront price. Open, shared data on bio-based materials would let AI models learn faster. And funding that brings chemists, engineers, computer scientists and industry together would turn promising lab results into real products.