Fraunhofer Institute for Cognitive Systems IKS
Fraunhofer Institute for Cognitive Systems IKS
Architektur
© iStock.com/Gumpanat

© iStock.com/Gumpanat

AI regulation and AI standardization

What is standardization?

Individualization is a high priority – in future, therefore, products should be tailored more and more to users. But there is also a counterpart to every development, as has been the case for some time: standardization. It paved the way for assembly line work and thus for mass production; this was particularly important for complex and difficult to produce products. After all, standardization is nothing other than the unification of components, manufacturing processes, units of measurement, services and so on. The aim: Costs and production time are to be reduced, quality and safety increased. Standardization also enables easier use – for example, when it comes to television. While TVs were originally only able to play films of a certain television standard, in 1963, the German system PAL unified the television standard. For the health protection of consumers, standardization and regulation also play a major role, for example in the form of standards for food or food supplements and drinking water. For instance, the Drinking Water Ordinance prescribes standardized and regulated limit values and test processes.

Fraunhofer IKS: Standardization meets artificial intelligence

Fraunhofer IKS focuses above all on safety in the field of AI standardization: Thus, the pivotal point of the research in this field is to provide a certain functionality safely, in particular with respect to artificial intelligence. How can the use of artificial intelligence be optimized in such a way that the decisions made by the software are safe? And where is AI regulation necessary?

Standards in AI: Where is AI standardization and AI regulation relevant?

Nahaufnahme eines fahrenden Autos mit Blick auf die Straße
© iStock.com/mseidelch

On the Artificial Intelligence standardization roadmap, this is particularly relevant for artificial intelligence in the vehicle – as is the case for autonomous driving. Here, the software takes over the job of the human driver. But: If the AI systems in the vehicle do not work reliably, this can have fatal effects, which in the worst case can cost people their lives. Clear and uniform specifications for AI systems are therefore fundamental when it comes to use in vehicles. There are already proposals for AI standardization for the software in cars. However, an AI standard that explicitly focuses on the use and regulation of artificial intelligence in the context of safety has been missing so far.

AI standard and AI regulation: elementary in autonomous driving, today and in the future

Cars increasingly support drivers, for example by means of distance warning devices or autonomous parking. For this, they must recognize their environment, process signals and automatically match the information they receive with road traffic rules and possible hazards. Certain functions for assisted and especially autonomous driving can only be implemented with the aid of artificial intelligence. After all, unlike programmed software, AI should in future react flexibly to unforeseen events, as they are almost commonplace in road traffic: A child runs into the street, a driver disregards the right of way, a sudden traffic jam arises. If the vehicles are to be able to dispense with a human driver completely in the future, this becomes all the more important – and safety-relevant. The researchers at Fraunhofer IKS are working on developing AI systems as safely and reliably as possible.

Fraunhofer IKS AI standardization research: Reliable, safe AI deployment

Under the auspices of Fraunhofer IKS, an international working group is developing norms and standards that deal with artificial intelligence from a safety perspective: the ISO/PAS 8800 standard. It is intended to standardize the development, testing and, where appropriate, regulation of future AI systems which are used in the vehicle. In doing so, it will not only provide specifications for neural networks, but also for other, more easily comprehensible AI approaches, which are in many cases more suitable for safety functions. The new standards in AI bring advantages not only in autonomous driving, but also in other areas of artificial intelligence, as far as safety aspects are concerned – for example, in surface-dependent chassis tuning.

AI regulation ahead: what does the new ISO/PAS 8800 standard do?

The ISO/PAS 8800 standard currently being developed will provide a guideline for the development of AI software. To this end, it first addresses fundamental questions of the standardization and regulation of AI. How and with what accuracy must a function of artificial intelligence work in order to prevent an accident? From this system level, the research team of Fraunhofer IKS is developing measures in function development which specify the structure of the AI function or concern the selection of the correct training data.

© iStock.com/Just_Super

Compensating for residual artificial intelligence errors

Functions of artificial intelligence will never be perfect even with the best standard – after all, an AI system is not programmed, but trained on the basis of data, in contrast to classical software. An extremely difficult situation for safety engineers. In the new AI standard, the researchers at Fraunhofer IKS therefore define measures to compensate for these residual errors and to monitor the AI function. In addition to the development process, they also define measures to check the corresponding software. In this way, they lay the basis for the use of artificial intelligence in the vehicle, which will surely increase in the future.

Project ISO/PAS 8800

For Fraunhofer IKS, ISO/PAS 8800 represents a special project: In Prof. Simon Burton, a representative of the Institute is taking a leading role for the first time. Within the two-year project period, a quorate draft will be drawn up, which will be synchronized with other standards and bodies similar in content. The result will be a standard and standardization of AI which ensure high safety standards when using artificial intelligence in vehicles. In this way, the artificial intelligence of the future is to become safer for everyone.

Blog articles about artificial intelligence

Would you like to learn more about Fraunhofer IKS research on the topic of artificial intelligence? Then take a look at our blog. Here you will find all blog articles on artificial intelligence.

AI blog articles
 

© iStock.com/Widewingsstudio

Artificial Intelligence

The goal of Fraunhofer IKS is to take advantage of machine learning technologies in order to design a future that is safe. AI technologies must be 100 percent dependable, especially when it comes to safety-critical applications.

Artificial Intelligence
 

© iStock.com/AscentXmedia

Safety Engineering

The electronics inside vehicles and machinery are growing increasingly complex. It takes more advanced safety mechanisms to tame this complexity. This is why safety engineering is a focal point of research at Fraunhofer IKS.

Safety Engineering