Artificial intelligence will get better at planning

Humans use logical thinking when planning routes for parcel deliveries, controlling robots in factories or defending networks against cyberattacks. Wallenberg Academy Fellow Jendrik Seipp will develop AI systems that can reason more like humans when they draw up a plan for achieving a goal in stages.

When AI systems make step-by-step decisions towards a specific goal, they use something called heuristics. These are clever rules of thumb that help the computer estimate which route seems be to the quickest way – much like when humans look at a map and choose the motorway, because it is usually the fastest.

However, designing good heuristics is difficult. When current AI models develop new heuristics they do so rapidly, but the results are often ambiguous and difficult to interpret. People are better at developing reliable heuristics, thanks to our rational thinking. However, this takes a lot of time and the heuristics often only work for specific purposes.

In his research, Dr Jendrik Seipp at Linköping University combines the best of both worlds: machines’ ability to learn quickly and humans’ rational reasoning. The goal is to develop AI models that can produce effective heuristics, allowing computers to plan and make decisions more reliably. He is also exploring methods other than heuristics for AI planning, to see whether these can be improved and provide a basis for smarter decision-making systems.

Photo: Patrik Lundin