Even smart systems need to be controlled
The hype surrounding ChatGPT earlier this year has shown that artificial intelligence will dramatically change our lives. The technology of the future has become part of our everyday lives. It's high time to take a more detailed look at the intelligent systems. New testing and certification procedures must be developed to evaluate learning systems according to objective criteria. Our expert Dr. Martin Saerbeck explains how this works in the video.
A similar learning behavior to that of humans
AI learns – in part comparable to humans – in small iterative steps to perform a task better and better. In the background of every artificial intelligence are algorithms that train a model, sometimes with considerable amounts of training data. This also means that the quality of the trained models is highly dependent on the quality of the training data. That is why data quality must be an aspect for certification by TÜV SÜD.
“TÜV SÜD is working hard to develop test methods for AI and is also contributing this knowledge to international standardization.“
Dr. Martin Saerbeck
CTO TÜV SÜD Digital Service
Putting data and processes to the test
The European Union is working on standardized rules on AI: From 2025, the EU Artificial Intelligence Act is intended to create the prerequisites for quality checks on and certifications of artificial intelligence. Checks would then be conducted, for example, on whether the training data used meets defined requirements and whether the entire development process has been set up and carried out in accordance with the applicable regulations. Ethical questions also play a role in this context, for example to exclude discriminatory or racist AI. Dr. Martin Saerbeck is certain of one thing: AI will change the world, just like the steam engine in the 19th century or the introduction of modern data processing in the 20th century. And this makes it a case for TÜV SÜD – to ensure that innovations are also safe.
Back