Juhku - stock.adobe.com
Finland bets industrial recovery on health AI, maths and quantum
The ‘world’s first’ national AI model leads Finnish plan to concentrate national resources on industrial revival
Finland has hunkered down on its strengths in health, quantum and maths to nurture an industrial recovery that has been wanting since the decline of national champion Nokia in the smartphone era.
A research project intended to create the world’s first national health artificial intelligence (AI) system and to use it to predict the risk of each and every person in the country developing any one of 200 diseases was among six that the national trade and investment agency funded in June to stimulate industrial growth. Four of the six concerned biotech, an area where Finland’s national data and legal systems are claimed by those involved to be uniquely well formed.
Two aspire directly to the long-term development of the quantum technology industry that emerged from Finnish scientific leadership, centred at the Aalto University Low Temperature Laboratory, in ultra-low-temperature physics, cryogenic engineering, superconducting quantum circuits and nano-electronics.
Finland’s world-leading position in the field of inverse mathematics and computation, which tackles the fiendishly difficult problem of calculating causes from observed effects, produced a project to do advanced medical imaging on cheap, inferior scanners.
Business Finland picked from among nearly 200 proposals those that were most likely to achieve strategic aims it set last year to accord with a government policy to stimulate an industrial recovery that diversified its economy away from ICT where it had long been heavily concentrated, and therefore vulnerable.
Finland’s R&D spending, at its height in the late 2000s, was greater as a share of GDP than any other country in the world but Israel, according to OECD data. That came largely from private investment in the ICT sector, and mostly from Nokia, which accounted for over 40% of all Finnish R&D. Its spending fell dramatically after Nokia sold its mobile phone business to Microsoft in 2013, as the country’s innovation ecosystem weakened and a “long economic stagnation” set in. Nokia still dominates.
“We have academic strengths. We want to build them into industrial strengths to create more variety in Finnish industry,” said Karin Wikman, chief innovation adviser of Business Finland, who worked on the Rise to Challenge fund. It awarded shares of €30m to six projects that had the boldest of visions and that promised to stimulate the broadest industrial growth, she said.
The agency granted funding to the health AI project FINe-Health Foundry because it built on Finnish strengths in health data and the urgency of its mission to cut costs and increase capacity in a health system becoming overburdened by an ageing population, said Wikman. It maps to priority sectors the government set for its industrial recovery strategy in November.
Model of health
The FINe-Health team intends to create what its claims will be the world’s first national health AI model, made possible by Finland’s centralised health records system. It emerged from FinRegistry, a project that collated 19 national health data sources for all 5.6 million people in Finland. This included everything from prescriptions, health conditions, lab reports, health clinic visits, socio-economic data, and family connections, so that medical computing researchers can use it to train machine learning models. It will incorporate also a genomic database that has records of around a tenth of Finns.
The result will be a foundation AI model more advanced than the machine learning models that have thus far been trained on isolated cohorts – subsets of patient records – for specific clinical aims, such as predicting mortality, said professor Arto Klami, who is working on FINe-Health at the University of Helsinki.
This would not have been possible in most countries of the world, he said, because health records systems tend to be fragmented, protected by local hospitals and clinics and not available country-wide.
FINe-Health will spend three years building and training the model on national health data, and working with clinicians and planners to map their workflows and identify ways an AI might help them, said Klami. It will build prototypes to test how well the AI meets those needs. Then it aims to encourage health tech firms to develop full applications.
“The foundation model in isolation, sitting in some secure computing environment, does nothing,” said Klami. “We see societal benefit only at the point it is available, where there is a real need, a clinical workflow or practice we want to improve. It’s going to be a complicated puzzle, getting all the pieces together, and much of it can’t be done by a university research project. It involves lots of public and private actors.”
The system’s first likely use would be in population-level analysis that planners and policymakers could use to do such things as predict disease, identify vulnerable groups and allocate budgets. The FINe-Health team intend it to help doctors make decisions. That would be difficult, said Klami, because it raises ethical questions that go beyond FINe-Health’s technological remit to develop and prove the technology.
Current health AI systems are limited by the fundamental technology by which foundation models mostly learn to derive correlations from data, said professor Samuel Kaski, founding director of the ELLIS Institute Finland, which is running the project. FINe-Health, attempting to create a model that can be used in clinical decisions, must be able to do cause-and-effect reasoning as well.
How to train models to capture causality and make valid inferences from it was one of the fundamental research challenges in machine learning, he said, adding: “While partial solutions have been proposed, this is far from being a solved problem.
“We are not the only people working on this, but our project intends to make a contribution which is especially needed in healthcare, where it is absolutely crucial that decisions on treatments of patients, for instance, are made based on the disease mechanisms instead of spurious correlations.”
FINe-Health will attempt to establish a model of human-AI teamwork and decision making as well. Contributions from health practitioners will be vital, said Kaski.
Read more about healthcare tech
- Estonia is a thriving hub for advancements in healthcare technology, harnessing artificial intelligence and genomics.
- IT services firm opens a window to the future of healthcare and physical training as tech advancements converge.
- Healthtech company Cera's artificial intelligence app can also predict the onset of winter viruses, and is being keenly adopted across the NHS.
