Klaas Bollhoefer on AI: “It’s a 360°change in how we build software”

In April, Klaas Bollhoefer of Birds on Mars spoke at our Leaders in Tech event in Berlin. His talk, AI Thinking, explored how organisations can learn to stop being afraid of artificial intelligence, and lean into the paradigm shifts necessary to implement AI tools creatively and valuably.
We spoke with Klaas to learn about how he came to found Birds on Mars, and what the biggest challenges are for the companies he works, like Deutsche Bahn, Commerzbank, and Lufthansa.
Tell us a little bit about yourself and your company, Birds on Mars.
I’m founder and managing director of a company called Birds on Mars in Berlin. My cofounder and I, we built Birds on Mars at the beginning of last year. The idea is to help companies develop strategies, structures, teams, and applications at the interconnections of human organisations and artificial intelligence. We want companies to get off their asses and do everything that’s data and AI themselves.
It’s not something that you can outsource, so you have to develop new capabilities, create a new understanding – a new foundation – to really get into this topic. In the end, ideally, you might create new value or new business out of it.
What made you interested in the intersection of human business organizations and artificial intelligence?
We found out that companies really need support, they need guidance, they need a company that doing strategy in this kind of field, bottom-up. The difference is always that we’re not just talking about AI: we know how to do it, as well as strategic work.
We know how it needs to be done and that it’s more than technology. You need technology, you need skills, you need to routines, you need new spaces where this new stuff can happen. You might need new processes in your organization, you might need new organisational structures. You might need new partners.
What are the major challenges are right now in the work that you’re doing with customers?
there are so many paradigms that change or will change with big data and AI. The biggest challenge is always to create the understanding and awareness on the client side that it’s really time to think over existing stuff and time to dig deeper into data and AI.
If management and decision-makers are ready, and have the initial understanding, if they say “OK, we got it. Data and AI are the kind of thing we have to care about. We have to understand and find out how can data can help us in our organisation” –  then that’s 80%.
I think the biggest threat to AI at the moment is that people feel lost and have this feeling that they just won’t get it. Everybody talks about AI and a lot of people think, “OK, I don’t get it. It’s too difficult, I need a PhD in mathematics to actually get a grip on that.” But that’s just wrong
If you really help companies to translate what AI really is and what kind of paradigms are behind AI or behind data and behind cloud (because all of these things are very intertwined and very well connected), then they feel able to decide again. They are capable of taking the next step.
They don’t feel lost and behind. That, I think, is the biggest task.
After that, with these changes in paradigms there are a lot of new things that need to be learned. There are things that need to be unlearned, which is normal when you have a kind of paradigm shift.
What are some of the things that people need to unlearn once they’ve made the paradigm shift  into thinking correctly about the role of AI and big data in their organizations?
One of the biggest shifts is the way we design, build, and operate software. In general, AI is software, but software of a new kind. The way we create this kind of software, the way we develop, but mostly the way we run software that integrates any kind of machine learning, deep learning (we call it AI) piece, that’s the biggest change.
At the moment, software gets developed explicitly, so it’s rule-based. If, then — else. You have data, and I take that data, I write a piece of logic and the output is pretty straightforward. I know what the output will be. It’s easy to measure. With machine learning, we turn a lot of stuff upside down. We train software based on data that we know. We train the logic and put that logic on unknown data from the real world. Ideally if what the software and the logic is trained on is similar to what the actual data in the real world looks like, it behaves the way we want.
If the data changes in the outside world, our logic will work on that changed data and the output, the result, of what the logic is doing, can be anything. We don’t really know, so we have to monitor all that. We have to monitor what kind of data flows into our machine learning system. We have to monitor the output. We have to apply new metrics. At that point in time, it’s not just the software we have to take care of, this piece of software we are responsible for, but the whole process that changes – the whole environment that this software actually operates in. That’s quite complicated and it is a 360 degree change in how we actually think and build software. That’s the biggest shift, I think.