Autonomous systems are going to control more and more in our society, but how do we ensure that they make the right decisions? Two new research projects aim to address this.

Artificial Intelligences must be certified before they are given control

In our everyday life, we are deeply reliant on control systems. They maintain our supply networks and ensure that there is water in the taps and electricity in the sockets, even when production and demand vary.

However, existing control systems do not fully exploit the vast amounts of available data. This results in underutilization of the potential for optimal control. Implementing artificial intelligence will ensure optimal control, but we hesitate due to concerns about reliability and risk.

The challenge lies in the fact that the neural networks used are not created by humans, but by data. This makes the decision-making processes incomprehensible. And while neural networks are impressive, they can make mistakes. Errors that, in some cases, can have serious consequences, but which we cannot identify. On the other hand, there are clear rules for how control systems must act.

Assistant Professor Christian Schilling will address this concern by using existing rules to develop algorithms capable of analyzing neural networks. This way, he can establish certification procedures that build trust in the new generation of control systems, allowing us to safely implement them into the supply network.

Robots should make decisions in the same way as humans do

With the implementation of automation and artificial intelligence in industries, there's a need for conducting larger and more complex calculations on vast amounts of data. Robots in production must make independent decisions and choose the best and safest solutions. Currently, these decisions are made using traditional techniques involving extensive data analysis based on evaluating all alternatives, which makes them slow. Additionally, the numerous calculations result in increasing resource consumption, contradicting sustainable development efforts.

There is a need for automated and informed methods to manage the planning process

Through his project, Associate Professor Alvaro Torralba aims to introduce a new approach to digital decision-making processes, where robots and artificial intelligences make choices by assessing pros and cons instead of going through all the data. With this approach, he can develop algorithms and tools ensuring that the decisions are both safe and optimal.

Congratulations on the grant.