By allowing artificial intelligence (AI) to mimic the reasoning employed by human experts, AI methods can be improved and their acceptance broadened. This is seen in new research from the University of Skövde.
AI has a unique position among the digitalisation techniques currently transforming our society. Working with AI is also expected to become a key factor and a competitive advantage within the industry when it comes to, say, process refinement and efficient resource use.
Niclas Ståhl is a doctoral student at the University of Skövde and has studied how the actual AI methods could be improved. In his thesis, he has focused on AI methods within the steel industry and the pharmaceutical industry.
“In my research, I’ve studied how AI methods can be improved by getting them to mimic the reasoning employed by human experts to solve certain problems,” says Ståhl.
Human reasoning helps AI to perform better
His thesis includes several case studies in collaboration with stakeholders within steel manufacturing and the pharmaceutical industry.
The case studies within steel manufacturing are focused on rolling and smelting while the focus within the pharmaceutical industry is on the development of new drug candidates and predicting the properties of different molecules.
Niclas Ståhl has compared AI models in which expert knowledge is included to AI models in which expert knowledge is excluded. The results showed that the inclusion of expert knowledge enables the models to perform better. The human experts have worked with the same problems for many years and therefore already have a great deal of knowledge about the problems that the AI models are to solve, he explains.
“By mimicking the experts’ way of thinking, we can include knowledge about how to solve the problem right from the start. This way, the AI model doesn’t need to start from scratch. The fact that the AI models mimic human reasoning also makes it easier for others to trust the models,” says Niclas Ståhl.
Greater gains from AI use
Improved AI methods also mean greater gains from using them. The results of Niclas Ståhl’s research can be used, for example, to optimise the development of new drug candidates, to improve the control of industrial processes and to automate repetitive tasks.
This means financial gains for the industry, but also gains for society. The faster new drugs are developed, the sooner they can benefit patients. Improved processes mean reduced resource use and reduced waste, which benefits the climate. In the case of repetitive tasks within the industry, routine mistakes can be avoided as AI never tires of its task but instead continues to perform just as well.
Niclas Ståhl will be defending his thesis Integrating Domain Knowledge into Deep Learning on 10 May.
Read more and watch the broadcast from the defence:
Public defence: Integrating Domain Knowledge into Deep Learning