Forget ChatGPT – AI of the future will bring major small changes

Today’s AI is artificial intelligence, but it just talks. It doesn’t make a single decision, and it doesn’t make things happen. We’re talking about ChatGPT, Claude, Deepseek, Gemini, Grok, Meta AI and the other AI variants that you hear about all the time. They are just language models.

Now a brand new centre is opening to develop artificial intelligence that can be used for something completely different. The centre has received a solid part of the AI ​​research NOK billion from the Norwegian government. SINTEF and NTNU are co-leading the centre, which will study how AI can be used to make decisions about everything from health to industry and energy.

“There’s been so much focus on language models,” says Signe Riemer-Sørensen. She is a research manager at SINTEF and one of two centre directors in the newly established Norwegian Centre on AI-decisions.

Calculating the right dosage of medicine

“In reality, completely different types of artificial intelligence actually offer the greatest potential. Linking AI to industry and automating large processes is when we’ll start to see a gain,” says Riemer-Sørensen.

She uses health and energy as examples. Imagine that you have diabetes and that you keep it in check with insulin. You need to know exactly how much insulin you need.

“You take insulin based on your bodily measurements. Actually, you should have insulin based on what you’re about to do. Some researchers are focusing on this future-based method, but it is still far from good enough. Artificial intelligence can create models of people and make these kinds of decisions about treatment,” Riemer-Sørensen says.

Managing electricity consumption

Another way that AI can tidy things up where humans can’t so far is in energy production. If you have solar cells on your roof, they tend to produce the most electricity exactly when you need it the least.

“It’s difficult to get rid of the extra electricity on the grid, because it can become overloaded. AI can provide better control all the way down to the solar cell level and out to the grid and manage how and when we use the electricity,” she says.

Many small models

“The whole point is that we shouldn’t just build one model that handles everything at once. We should build small models that handle individual tasks and that can communicate with each other. Even the language models are moving in the direction of many small, specialized models,” says Riemer-Sørensen.

The researchers at the SINTEF- and NTNU-led AI centre will focus on building these kinds of models that take care of specialized tasks. They will also build models that decide whether they need help from other AI models that specialize in other tasks.

“We have more than 50 partners from industry and the public sector on board. In addition to the research, we’re looking at what needs to be in place for people to use this artificial intelligence,” she says.

The similarities are greater than the differences in some places. For example, even though diabetes patients and the power grid are very different, the principle for how artificial intelligence can control what should happen is very similar. This means that researchers, technology experts and end users can learn a lot from each other.

Where should the snowploughs be sent first? Artificial intelligence can make the decision independently. Photo: Bård Asle Nordbø, Norwegian Public Roads Administration SHOW MORE

Making AI more independent

The key is autonomy, explains Sebastien Gros.

Gros is a professor at NTNU and co-leads the new centre with Riemer-Sørensen.

Whether deciding where the snowplough or the gritter should be sent out first, or which doctor the patient should talk to, AI that can make independent decisions is needed.

“Some partners at the centre want recommendations and alerts that a decision needs to be made. Others want full autonomy where the algorithms make the decisions,” says Gros.

The problems can still be quite similar: For example, the researchers have put energy professionals and the Norwegian Public Roads Administration around the same table, only to find out that they actually want to solve the same task.

Setting up autonomous AI systems like this is already possible with the technology that exists today.

“We can do it with sufficient data. Some of our partners have such systems in the energy market now,” he says.

Autonomous AI models are easier to create when the data involves clear numbers. If the basis for decisions involves text – like a handwritten note from a doctor about something a patient has said verbally during a consultation – then the task becomes more challenging.

“What we have to do in any case is to sort data, predict results and use them to make decisions,” says Gros.