We must create safety rules when AI develops

”Human beings will continue to develop machinery to serve us, but robots need rules to follow,” says Magnus Hjelmblom, researcher at the University of Gävle.

AI requires rules

Developers of robot systems or intelligent software have many things to take into consideration. Amongst other things required are certain rules and an internal system for supervision of parts, so that the so called agents only do that which they have permission to do.

”My research involves regulating the behaviour of the agents with the help of what I call norms, which include both unwritten rules and formal regulatory systems and rules,” says Magnus Hjelmblom.

Development of a framework

The new approach is that Magnus starts from algebra and not from the more common approach which is logic, and used it as an instrument with which to regulate the actions of executable systems.

”It is a rather new approach and I have developed it from a theoretical concept to something that can be implemented practically; shown that one can create norms that are meaningful and that give a valid result.

Algebraic methods are basic mathematics and often very appropriate for processing on the computer.

I have developed a framework so that someone else could then later create a specific application. As an example I have also developed a number of concrete programs/applications that illustrate how the framework functions and can be applied.”

Simulation of evolution

In one of the examples, a so called genetic algorithm is applied, that simulates evolution to develop something that solves a problem more successfully every time.

This can be done partly by using a simple function to direct the behaviour of the agents expressing what the agents want to do and partly a normative system that says what they are allowed to do.

- A combination of these can lead to quite a forceful problem-solving ability.

Machines are taking over more and more

Machines have more and more influence in society and they are becoming more and more advanced. They are taking over larger areas of employment such as those within care. It is easy to understand the importance of making sure that these machines are provided with clear rules for what they may or may not do.

So it will be interesting to consider how one can regulate the behavior of these artificial agents

Human beings must remain in control

• Machines are good in many ways and we must use them to help us. But it is we humans that must make the crucial decisions. We must develop useful tools for decision makers so that we can ensure that the decisions will be as good and rational as possible.

The future

Magnus changes the subject to Decision, risk and policy analysis.

Earlier I always thought of myself as a teacher and when I have publicly defendedmy doctoral thesis I will return to teaching.

• But there are many interesting tracks to pursue as far as my research is concerned. I have a few other interesting projects on the go that I have worked with during my doctoral studentship. They concern development of computerised decision support systems.

  Magnus Hjelmblom defended his licentiate thesis "Norm-Regulation of Agent Systems: Instrumentalizing an algebraic approach to agent system norms" at Stockholms University on the 3rd ofNovember.

  Professor Magnus Bohman, KTH, and Professor Jan Odelstad, HiG have been his supervisors.

  For further information, please contact:

  
  Magnus Hjelmblom, lecturer in Decision, Risk and Policy Analysis (DRP) at the University of Gävle
  Tel: 026-64 86 48
  Email: magnus.hjelmblom@hig.se

  Text: Douglas ÖhrbomPhoto: Ove Wall