Specialised AI models could be Finland's next global export

When you open your car's navigator on a Monday morning, its first suggestion is a route to work – not your Thursday night hobby. When the highway dips into a tunnel, the connection remains stable. Even in rush hour traffic, there is enough network bandwidth allocated just for you. These are made possible through machine learning and artificial intelligence: trained models predict your route, hop across operator networks, and ration the load of 5G base stations. 

Small, specialized AI models: the heart of future industry and telecommunications 

In everyday life, artificial intelligence (AI) often refers to generative tools that utilise large language models (LLMs): the likes of ChatGPT, Claude and Gemini. In reality, AI is an umbrella term for a wide range of applications based on machine learning – and there's no single AI model that can or should meet everyone's needs. 

According to Petri Mähönen, Vice Dean of the School of Electrical Engineering at Aalto University, Finland should invest in AI applications in which we already excel: dedicated AI models that suit the needs of automation industry giants such as ABB and Konecranes, as well as the field of telecommunications. 

The reason is simple: Finland is currently unable to compete in the development of the largest language models with hundreds of billions of parameters. Our national economy does not support the huge investments and computing power needs required by these models. 

LLMs currently generate massive added value, and thus technology giants are focusing on these models. They use their significant resources to address challenges with a wide scope. On the other hand, industry needs often lie in specialized AI models. These models are used in products, but can also speed up, optimise and enhance product development. 

'Our advantage is agility: when presented with a problem of a limited scope, we can quickly tackle it and develop precise solutions,' Mähönen says. 'After all, squirrels can move nimbly among dinosaurs – especially when they invent something new. Gaining good research results arouses interest among tech giants, as well as within the industry looking for suitable AI solutions." 

Artificial intelligence – a threat and an opportunity for sustainable development 

Training LLMs is expensive, and consumes a huge amount of natural resources. In addition, these models  pose an ethical risk: since they partly operate on a “black box” principle, the basis of their decision-making isn't always transparent. The models may have been trained on incorrect or biased data, either accidentally or intentionally. 

Jussi Ryynänen, the Dean of the School of Electrical Engineering at Aalto University, highlights that developed AI models should be transparent and understandable in order for their operations to be trustworthy. 

'If we know how a model has made a decision, we can detect issues and take responsibility for them,' he says. 

Educating future experts 

As these experienced gurus are retiring, Aalto University is proactively educating future experts who can tackle the challenges of sustainable development. The School of Electrical Engineering has long understood that hardware and applications – including artificial intelligence – are not opposites or separate entities. To answer rising needs, the School has launched a new bachelor's programme in Computer Engineering, in which students get to know both sides of the   metaphorical coin. 

'Currently, hardware manufacturers and software developers operate separately, resulting in a lot of wasted energy. We address these in parallel, and in a purposeful way, so that both can be optimised for energy efficiency,' Ryynänen says. 

According to the Dean, some integrated circuits or computing architectures that utilise dedicated AI models can be made so small that they fit onto portable devices. This allows them to operate locally, without requiring large data centers and massive computing power, veering the predicted energy use of AI onto a more sustainable path. 

'The Computer Engineering programme and its related research are designed to meet the real needs of industry and society at large: we combine microelectronics, telecommunications, semiconductor design and software expertise into one holistic programme. This way, we train engineers who are able to develop energy-efficient and intelligent solutions on a domestic level – whether it's edge AI, 6G networks, robots, smart elevators or sustainable IoT systems,' Mähönen adds. 

Jussi Ryynänen emphasises that artificial intelligence in its various forms will be present in a wide variety of future products – and Finland must keep up with the development. Aalto University wants to ensure that Finland does not remain on the sidelines of AI progress, and instead builds its own competitive edge based on sustainable technologies. 

'Internationally scalable, precisely targeted AI models can create new domestic ecosystems of innovation and export, strengthening Finland's technological self-sufficiency and competitiveness,' the researchers conclude.