While Artificial Intelligence is undoubtedly one of the fastest growing fields, machine learning is what drives innovation behind the scenes of this technological revolution. Machine learning models are found in a countless number of everyday applications, whether it be image recognition, recommendation systems, chatbots, diagnostics, or forecasts, machine learning models are given their “intelligence” by using large amounts of data to train them.

Can I Trust You?
This kind of intelligence can, however, appear unsettling, thereby raising questions: Can I trust this kind of artificial intelligence and is it safe? Does AI work the way it is supposed to? Is the training data used accurate? Is the AI handling the large amount of sensitive information carefully? The Institute for Machine Learning at the Johannes Kepler University and TÜV AUSTRIA are working together to create corresponding certification methods designed to support manufacturers and ensure they can develop safe, reliable and high-quality machine learning models. This also includes offering users a ‘quality seal’ confirming trustworthy AI systems.

Prof. Sepp Hochreiter, head of the institute, believes AI is paving the way to the future: "Machine Learning is currently one of the most important enabling technologies and in the long-term, it will massively impact both our technical environment and our lives. This makes it all the more important to strengthen consumer confidence in this type of technology by certifying machine learning applications. This is why we are more than happy to work on playing a key role in shaping the required quality criteria."  Regarded worldwide as a pioneer of modern artificial intelligence and initiator of Austria's first AI degree program at the JKU Linz two years ago, Prof. Hochreiter is an ELLIS board member, a network of excellence featuring high-profile European scientists in the field of machine learning and their applications. Prof. Hochreiter heads the LIT AI Lab and the ELLIS Unit Linz.

Thorough Testing
The first step has been successful, certifying so-called ‘supervised learning applications’ in a low to medium range of risk. Dr. Stefan Haas, CEO of TÜV AUSTRIA Group, remarked: "We are proceeding to conduct initial certification projects primarily in the area of industrial applications but also in the consumer sector as well. Upcoming developmental phases include expanding current approaches in order to be able to certify more safety-critical applications based on a broader spectrum of machine learning methods. It is a challenging undertaking and we are particularly pleased to have the JKU Institute of Machine Learning at our side as a highly competent, internationally recognized partner."

The detailed certification process includes examining the machine learning models and related development procedures in several levels. This means looking at the trained models’ actual function and dependability as well as software security and whether or not it has been developed correctly to meet the needs of the corresponding area of application. In addition, researchers will study the procedure to ensure personal information is being handled confidentially and whether or not potential ethical issues have been taken sufficiently into account.

Recently published: White Paper (published by TÜV AUSTRIA and the JKU):

www.tuv.at/loesungen/digital-services/trusted-ai/, opens an external URL in a new window