A new whitepaper by TÜV AUSTRIA and the Institute for Machine Learning at Johannes Kepler University (JKU) has established a certification methodology for Artificial Intelligence (AI) algorithms.
The paper first acknowledges how AI is one of the fastest growing subject areas around the globe, being behind many everyday applications involving image recognition, recommendation systems, chatbots, diagnostics or forecasting.
It then goes on to raise these questions:
- Is this artificial intelligence trustworthy and safe?
- Does the AI perform the function expected of it?
- Does the use of AI bring the expected value added?
- Is the training data clean and used correct?
- Is the large amount of sensitive data handled carefully?
The new certification methods developed by TÜV AUSTRIA and JKU aim to support manufacturers in the development of safe, reliable and high-quality machine learning models. They will also offer users a seal of quality for trustworthy AI systems.
Professor Sepp Hochreiter is Director of the Machine Learning Institute.
“Machine learning is currently the most important enabling technology and will have a massive influence on our technical environment and our entire lives, on society, in the long term,” Professor Hochreiter says. “Therefore, it is even more important to strengthen consumer confidence in this technology through the certification of machine learning applications. Which is why we are delighted to be involved in helping to define the necessary quality criteria.”
Professor Hochreiter has a worldwide reputation when it comes to AI and played a key role in implementing Austria’s first machine learning chair and AI first degree at the JKU Linz. He is also a board member of ELLIS, which is a network of excellence of the best European scientists in the machine learning field.
The first level of success, according to the whitepaper, has already been achieved. Supervised learning applications in the low to medium risk can already be certified.
DI.Dr. Stefan Haas is CEO of the TÜV AUSTRIA Group.
“We have started to carry out the first certification projects, whereby the applications are mainly found in the industrial environment but also in the consumer sector. In the next phases of the development cooperation, the current approaches will be expanded in order to be able to certify safety-critical applications, based on a broader spectrum of machine learning methods,” says DI.Dr. Haas. “We are particularly pleased to have the JKU Machine Learning Institute at our side as a highly competent and internationally recognized partner for this challenging undertaking.”
The certification involves the machine learning models and their development processes being checked in detail in several dimensions. This means not only are the actual functions and reliability of the trained models examined, but also the software security and its development process. There is also an examination of how personal data is handled confidentially and other ethical issues.