Trust issues: Can we really believe our tech?

For complex optimization problems, such as those in industrial and logistics processes or in the analysis of large datasets, the use of artificial intelligence is practically inevitable. However, artificial intelligence often produces so-called black-box models, which are difficult to interpret or even understand. Evolutionary algorithms are programs that are inspired by natural evolution and can be used to optimize a wide variety of processes; they offer a viable approach to enabling explainable artificial intelligence (XAI) as they can be used to generate models that are readable and interpretable for humans!

Generative AI models have dramatically changed our lives, allowing newcomers to computer science to even write entire pieces of software using simple natural language instructions. This new paradigm, “aka” vibe coding, has introduced unforeseen possibilities in our daily lives. Researchers are not immune to the hype, nor to its uncontrolled usage. This talk will give an overview of the current landscape of practices, perils, and provocative ideas on how vibe coding is perceived, measured, and addressed from the scientific community. Disclaimer: No AI model has been threatened to write this abstract.

How much can we trust a computer program? Over the years a vast amount
of bugs in important programs have made their way to news headlines. With
out question, such failures can have severe consequences. My research is con
cerned with making complex programs more understandable and trustworthy.
In detail, we are backing up applications solving mathematical formulas with
certificates to verify their results. I will show you how powerful these pro
grams are in solving real life problems and why we should trust them more
than solutions taken from ChatGPT.