Artificial Intelligence - and Human Knowledge

One example would be a business case that also interests me personally: A fully digitalized provider of insurance services who is already active on the market. It’s definitely impressive when you look at what sub-processes can already be fully automated in that sphere. I’m sure this is a direction that many companies will want to take. And it’s a development that we ourselves will pursue as well, because I believe it really is one of the answers to the challenges we face today.
 

Currently, discussions in business and society often center on how we in Europe and Germany can be successful in the AI field and not be left behind by pioneering countries like the US. Are we in Germany still not innovative enough where AI is concerned?

René: I take a more discriminating view. Throughout Germany, there are many initiatives and strong AI competence centers that are doing excellent work. Examples include large entities like the German Research Center for Artificial Intelligence in Kaiserslautern, or the Munich Center for Machine Learning. But there are also smaller initiatives like the KI.NRW competence platform, which focuses on machine learning and text analysis. These think tanks often collaborate closely with universities and research institutes and play an important role in promoting AI technology in Germany, across all relevant sectors of the economy.
 

If we shift our focus from research institutions to AI companies and start-ups: What are the biggest obstacles to faster AI development in Germany?

René: One of the greatest challenges is the funding culture, which remains sluggish. In the US, start-ups can easily find investors, whereas in Germany you often have to keep on pitching your ideas. In some cases, there’s a lack of success stories that could encourage investors. We need to change our mindset and invest more in innovation. We could also do better in terms of scaled support: After achieving their first successes, start-ups usually need additional rounds of funding if they’re really going to make it big, and this is where they often don’t get the support they need.

There are funding programs, to be sure. But the money they provide is usually not enough for major leaps and bounds. In the US, the culture is different: There’s much more investment in broad-based projects, even if the risk is high. That willingness to take risks is rare in Germany.
 

The EU's AI Act, which came into force in August, regulates artificial intelligence for companies, but it also makes the topic more complex. What are your views on this instrument?

René: The AI Act wants to protect people from unregulated AI. Its primary purpose is not to promote innovation, but rather to ensure that AI systems operate transparently and ethically and are used responsibly. This is important in building trust in AI and preventing misuse, especially when it comes to high-risk applications that can have a profound impact on our lives.

One example is lending: When an AI algorithm decides whether or not someone is granted a loan, we must be able to understand the reasons for that decision. This is often difficult because AI models are complex and not always transparent. The AI Act attempts to create clear rules to ensure that such decisions are comprehensible and fair.

André, how do you see the impact of the AI Act on business?

André: The AI Act will often create challenges for businesses because they will now have to closely examine and categorize their AI systems. But there are experts who can help companies assess their systems and ensure they meet the requirements.

It’s important that companies work transparently and disclose the origin of their data and how their AI models work. That means that businesses have to document exactly what data they use and how their models were trained. This will require some effort, but it’s the only way to build trust. Business must be prepared to invest more resources in documenting and reviewing their AI systems, and that can be a challenge particularly for smaller companies that don’t have the same resources as large corporations.

As a provider of an AI solution, we already started to develop appropriate mechanisms two years ago. Even so, we’re not yet at the point where we can substantiate all of the system's decisions 100 per cent. It’s a difficult field of activity. But we can at least provide complete transparency with regard to the training aspects – and that's half the battle. If you can show what input went into the system, you can usually predict what outputs you will get.
 

René, the leading AI solution from ISR that I just mentioned is "Buildsimple" – a cloud platform that specializes in intelligent document processing. It uses artificial intelligence and machine learning to convert unstructured information into structured data. In view of how AI is developing in Germany, how do you see the future of Buildsimple?

René: We launched our solution in 2017 with a focus on the financial services industry. We’re now working on creating powerful foundation models for various industries and continually expanding our service portfolio. Our goal is to provide AI solutions that are efficient and trustworthy. We don't just want to limit ourselves to intelligent document analysis, but also open up other fields of activity.

With topics like efficiency, ROI and sustainability in mind, how does your AI solution perform in terms of cost?

René: Developing and operating AI models, especially large ones, requires a lot of computing power and can be expensive. From the start, we developed Buildsimple as a scalable solution in which costs stay proportionate to benefits. We rely on the AWS cloud infrastructure, which allows us to respond flexibly to our customers’ needs. If a customer suddenly wants to analyze several million documents, our system can easily handle that task because it automatically scales upward and downward. To put it in sales talk: A small company benefits from our solution just as much as a global corporation.
 

How many AI models do you currently manage at Buildsimple?

René: At the time being, we manage more than 1,500 AI models. These models are highly specialized and tailored to the specific needs of our customers.

One aspect we are admittedly proud of is that our models are designed to deliver maximum results with minimum training input. Often, we need just a few examples to achieve outstanding results. That’s a great advantage because it significantly reduces client-side resource input and speeds up the implementation of their AI solutions.

Also, we see entirely new possibilities unfolding as the scope of generative AI continues to expand.
 

How do you assess the long-term prospects of AI in document logistics and beyond?

André: Currently, plenty of new large language models are arriving on the market. This is because we have seen a leap in technology thanks to the so-called transformer models, which make it possible to calculate such models economically, with large quantities of data. This was unfeasible just a few years ago.

I see the long-term perspective of AI mainly in automation and increased efficiency. We need to focus on optimizing processes and relieving people of repetitive tasks. At the same time, we need to ensure that AI systems work transparently and correctly. It’s critical that we find the right balance between innovation and responsibility. Another important aspect is the integration of AI into existing systems.

Interviewer: Thank you very much, René and André, for sharing your insights. AI is a multifaceted topic – let’s keep exploring it together!