In recent years, workflow automation platforms have become a large subsector within AI, with startups offering tools that can plan and execute tasks.
The focus of these tools is rapidly shifting towards the physical and science sectors, experts say. Spurred on by advancements in computing, scientists are using AI as a discovery tool in fields from clean energy materials to medical and therapeutics development.
According to Sifted data, European advanced materials startups have raised €3bn this year already — a significant increase from the €1.6bn raised in the whole of last year. Meanwhile, drug discovery startups have raised €4bn this year, on course to overtake the €4.7bn raised last year.
Many of these startups are already using AI and investor interest will likely increase as the technology develops.
Sifted sat down with experts to unpack how AI is being used in real-world environments — from science to computing — and what the future of these sectors looks like.
Moving into the physical world
There’s only so much value one can derive from purely digital tools.
To tackle global issues, such as climate change and disease, AI must be directed toward physical problems, says Chad Edwards, cofounder and CEO of CuspAI, which uses AI to generate and analyse material properties to tackle climate change and energy transitions.
“A lot of economic activity is driven by the physical economy and the opportunity space is much larger than anything we've seen in the language world,” he says.
We're helping to empower scientists with new tools.
Historically, AI in physical environments has lagged behind digital software due to slow and expensive experimental processes. “With software you can produce code, run it and see the results almost instantly,” Edwards adds.
“In the physical world, the feedback loop takes longer, whether it's running experiments to make materials or testing them in the real world.”
To survive these feedback loops, AI can’t just look for statistical patterns in data, but needs to actually “understand” fundamental laws of physics.
In AI modelling for materials science and biology, this means building physical constraints directly into neural networks.
“You need to train models with an understanding of what is physically feasible and with a deep understanding of that process,” says Anthony Bradley, cofounder and chief scientific officer at DaltonTx which uses an AI platform to power drug discovery for small molecules and biologics.
“For example, if you're building generative capabilities for antibody design, the AI needs to understand what antibodies look like and what can and cannot be expressed, and then be able to generate within that space.”
Domain expertise and humans-in-the-loop
Edwards and Bradley both say that breakthroughs require strong scientific domain knowledge.
“Domain expertise, strong AI and then I'd actually add a third as well: you need amazing engineers too,” says Bradley. “The systems we're building are based on complex science problems.”
We massively overestimate AI's ability to do genuinely new things without data.
But in order to move AI out of the lab and into the physical world, strong engineering and infrastructure is also needed, says Iraia Ibarzabal, chief growth officer at quantum AI company Multiverse Computing.
“The blocker we’re seeing is focused on the infrastructure. Something that can work in a lab is not the same in the real world,” Ibarzabal says.
To combat this, organisations will need to build new workflows around machine intelligence from day one, says Bradley.
“If we do everything as we’re currently doing them and then just sprinkle a bit of AI on top, it won't transform how we do things. We need the bravery to change the process,” he adds.
This also involves designing the technology around the reality of the end-user, says Ibarzabal.
“From the beginning of our projects, we define what the end solution is we want to deploy. That’s how you have a product that can run in the real world, rather than a proof of concept or a small project for the lab."
When integrated correctly into a system, AI doesn’t replace human experts but frees up their time to focus on other tasks.
Scientific research can involve labour-intensive tasks such as manually scanning academic literature, curating data or running trial-and-error tests.
"We're helping to empower scientists with new tools that remove that burden,” says Edwards. “Scientists are able to search scientific literature almost instantly as opposed to looking through every individual paper.”
Future of scientific discovery
Across materials science, drug discovery and computing, the value of AI is generally measured by its ability to deliver real-world outcomes.
“We massively overestimate AI's ability to do genuinely new things without data. AI models are strong at combining what's already been measured, but are much weaker the moment you step outside that space,” says Bradley.
“We need to dampen some of the hype and focus on problems that are impactful and the solutions that actually improve outcomes."
When looking at the next decade, Edwards and Ibarzabal point towards a future where AI becomes an invisible layer.
From the beginning of our projects, we define what the end solution is we want to deploy.
In computing, Ibarzabal predicts the most significant breakthroughs will come from pushing AI out of cloud data centres and directly onto physical edge devices in real-life environments.
“What we’re seeing now is once a model fits on a device without needing constant cloud access, it can run in disconnected, regulated or resource-limited places that it couldn’t do before. For example, in the defence sector or in industrial equipment,” she says.
Edwards predicts over the next five years, the main metric of success won't be found on a server dashboard but in our physical surroundings.
“My dream is that we'll be able to stand in the world and point at certain devices or equipment and say ‘that is powered by our materials’ and they’re having a positive impact on the world,” he says.
“Whether it's making solar more scalable, making chips more efficient or undoing some of the damage we've done to our planet.”
