Intro
Some companies, they say we want to save 20 million a year using AI. That's absolutely the wrong approach, because you're coming with a tool and you're not solving problems, you're creating problems. What you should rather say is: we want to get faster, cheaper, better in a certain process, and then do this and pick the right tools. And sometimes that's AI, sometimes not.
Speaker introduction
Hi, I'm Patrick. I've been in and around AI for the last 15 years. For the last 6 years, I have the role of professor of AI here in Bavaria, at the university. So I teach AI, I publish, but I also do consulting, help companies to leverage AI. And I also do political advisory.
How can teams know they're on the right track when using AI?
Metrics are important, and you need to define these in the very beginning, when you identify use cases. So you need to define how good or bad a company is now in a certain process. And oftentimes companies struggle a lot with that. But you need to do that. Otherwise you're investing in AI and ultimately you don't know if you have improved—maybe you even got worse. So you need to spend enough time on the metrics, and then you can measure more reliably later on.
What are companies getting wrong with their AI investments?
There's a lot of, you know, narratives that AI would be insanely expensive, and that scares companies. But it's not. So obviously, building your own LLM from scratch is super expensive, but you can use pre-trained ones, or use a cloud service, or there are lots of other approaches too — because I think for 95% of the use cases I work on, LLMs do not play a role. And there are so many other AI models, and some of them are super cheap. You can even train them on a watch, for example.
Which teams should be involved with AI programs?
The AI team will have lots of ideas about AI, but they don't know the main business of the company, so they need to work closely together with the domain experts. And the domain experts, on the other hand, don't know much about AI, so you need to work together very closely. And, you know, there's always something that correlates in the data. But does it make sense from a business perspective? Well, the data scientists don't know. So I think working together with the domain experts is crucial.
And yes, you need to think about compliance too. I know here in Europe—especially in Germany—we tend to first think about all the regulatory stuff, and then we never get to solve those actual problems. And in some domains you have to think that way, for example in healthcare. But there are lots of other sectors where I would really encourage: get stuff done first, build a prototype, and bring in compliance. But maybe compliance should not be the ultimate, you know, priority during an R&D phase, or when you build a prototype.
How should companies approach AI governance?
From my experience, you should include lots of stakeholders in AI governance—not just the AI team or the cybersecurity team, but also the people who get the actual work done in the company, because they may have plenty of ideas that you never thought about, but also new challenges that you never thought about. And ultimately, proper compliance, proper governance—it must foster innovation and not stifle innovation. I think that's so important.
There's a lot of AI happening and we have lots of young talent. Universities are training an enormous amount of AI experts, and we have plenty of graduates already, and they are building more and more AI applications. So I'm just super excited to see all the new applications that will be built in the next 5 to 10 years. And there's plenty of potential that Europe can become a bigger AI leader. Obviously, there's the US and China, but there's plenty of opportunities for Europe too.
_HighPerformer_HighPerformer.svg)
