aau/MüllerHow can systems solve complex problems such as creating a duty roster in a hospital or a time and room schedule at a university – while remaining clear and easy to follow? In the project ‘EX3: EXplain and EXploit Knowledge EXtracted to Improve ASP’, funded by the Austrian Science Fund FWF, Alice Tarzariol will be working on new methods for solving these kinds of issues more efficiently and accurately with the help of Artificial Intelligence.

At the heart of the project is ‘answer set programming’, a type of logical programming that describes problems using rules and logic. This means that the machine is told what rules apply, rather than how to calculate the solution. The programme then searches for solutions that comply with all the rules. Systems like this are needed, for example, when drawing up duty rosters in hospitals, where numerous rules have to be taken into account: What are the maximum daily working hours for each type of staff employed in the hospital? When do shift handovers need to take place, and how can you manage overlaps between different types of staff? Usually, problems like these are addressed by the experts who draw up the duty rosters.
“One of the major advantages of Answer Set Programming is that these experts do not need to be programming specialists,” as project lead Alice Tarzariol, postdoctoral researcher at the Department of Artificial Intelligence and Cybersecurity at the University of Klagenfurt, explains. “They can describe problems at an abstract level, and the ASP systems available can handle the complex search for solutions. However, if the problem descriptions influence the automated search, it takes in-depth technical knowledge and a lot of time to find a solution.”
This is where the project ‘EX3: EXplain and EXploit Knowledge EXtracted to Improve ASP’ comes in. The aim is to develop a system that does not only find solutions, but also supports experts in understanding and formulating problem descriptions. “We want to develop methods for automatically identifying hidden connections and symmetries in problem descriptions. This will speed up the process of finding solutions and help to explain them in a more comprehensible way,” Alice Tarzariol suggests. It is also essential that users – for example, those who enter the conditions for a duty roster into the system – gain deeper insight into how it works, as Tarzariol goes on to explain: “We aim to develop a system that can provide comprehensible explanations and graphical illustrations to show exactly how a particular result has come about. This white box approach is not only intended to promote trust and control, but also gives users the advantage of being able to check whether they have described their problem properly. They are expected to gain additional insights into their area of use and also interact with the system in order to deploy it in a targeted manner and improve their own inputs.”
As well as making a significant contribution to fundamental research, Alice Tarzariol hopes that her project will lead to the development of a functional system that demonstrates how explainable Artificial Intelligence can be used in real-life settings.
Der Beitrag New research project: Artificial Intelligence that thinks logically erschien zuerst auf University of Klagenfurt.