The year 2022 drew to a close for the Hasso Plattner Institute (HPI) in Potsdam with extremely happy news, as in November it was officially selected by the German Federal Ministry of Education and Research (BMBF) as the location for the KI-Servicezentrum Berlin-Brandenburg (AI Service Center Berlin-Brandenburg). As one of four hubs across Germany, it is intended to facilitate access to the key technology of artificial intelligence for various players, as well as to further strengthen AI research and AI knowledge transfer through collaborations with various external partners. #ki_berlin talked to its head Prof. Dr. Holger Karl about the specific approach, goals and challenges as well as the importance of trying out and experimenting with AI for players in the capital region.
Hello Prof. Dr. Karl, you have been heading the new AI Service Center Berlin-Brandenburg at the Hasso Plattner Institute since November. How did you start the new year with this new task?
We started the new year the way we ended the previous year: with the search for employees. The start of the center came quite suddenly, so we are currently still busy building up the team. And it is generally known that IT experts and especially experts in machine learning are currently in high demand.
Specifically, we are facing two important tasks at the moment: First, we are planning the infrastructure that we want to provide to our users. Second, we are in the process of designing workshops and consulting services.
What tasks is the AI Service Center tackling? What is the main objective?
I always like to explain our offering as "help with the second step." There are already many tutorials available for free on the web for taking your first steps in artificial intelligence and machine learning: Tutorials, blogs, videos and introductions to tools. There is also a whole range of introductory courses and events, for example for companies from our colleagues at the Mittelstand Digital Center. You can achieve quite a lot with that; you are able to solve simple problems with ML.
But most of the time, the offer doesn't go beyond that. What do I do when I no longer just have a problem that fits on my laptop? But when I have to deal with larger data sets, when I wonder if I can use pre-trained models or if I have to develop everything from scratch? Or when I have to make a pragmatic investment decision: Do I build my own infrastructure for my ML applications or am I better off outsourcing to a hyperscaler? How do I make this decision? In general: How can I reliably use machine learning in my applications? All of these second-step questions are rarely addressed. With the AI Service Center, we want to contribute to bringing ML into practice.
To do this, we will have to solve research questions on the one hand - such as the efficient operation of an infrastructure that has to handle mixed applications, i.e., both ML and completely normal applications. On the other hand, we need formats to enable users to take precisely this second step. To this end, we are planning workshops or the joint development of small prototypes, so-called Proof of Concepts, or PoCs for short.
The offer is aimed, among others, at people interested in the field of machine learning. Where exactly can you find this target group and how can they interact with your center?
In fact, these are the primary addressees; AI is used almost synonymously with ML these days, even if this is, strictly speaking, wrong.
But the target group is also an exciting question for us! Because there is no one, homogeneous target group that we can address. AI and ML can be used in a wide variety of industries and by a wide variety of players. With our offerings, we want to address as many as possible. For example, small and medium-sized enterprises that have data from which they can presumably derive added value, but need help to do so. But public administration or research institutes that are a bit further away from information technology also face similar challenges. We are very open here, and I believe that the concerns of such different organizations are not necessarily very different at all. Our approach is therefore to understand the target group through prior knowledge and expectations. As I said, we are not primarily addressing beginners with the AI service center, but rather want to help advanced users take the next steps.
Interacting with us is easy: Just send an e-mail to email@example.com! During the course of the project, we will also expand our social media presence and announce the workshops there, for example. Our Website will also become more comprehensive.
All in all, access to AI is to be facilitated by the AI service center. How can one imagine your measures in this regard? What is planned in concrete terms?
First of all, we are planning a series of workshops to help professionalize and reliably deploy ML and operate the corresponding infrastructure. I hope that we will be able to start this at the beginning of the second quarter, perhaps even earlier. In addition, we will offer individual participants in these workshops the opportunity to jointly develop a proof of concept, i.e. a practical test. Together with our colleagues from the Mittelstand Digital project, we envision a kind of pyramid: Interested parties can first take the Mittelstand Center's beginner workshops. Some of them will then continue with our workshops, and a smaller group will then build prototypes together with us.
In addition, we are planning a second offering: trying out ML tools and processes in different environments. This is a key reason why we will make a significant investment in ML-enabled infrastructure. We want to enable experimentation. To do this, we will also design the infrastructure in an unusual way, namely quite heterogeneous, with different shares of graphics accelerators, different processor architectures, different main memory footprints, and different storage systems. This would be quite unusual in a production system, but it is appropriate for such an experimental environment.
And that is exactly what we want to and will offer: an experimental environment. We are not in a position, nor would it be covered by the project's intentions, to offer productive operation here. Routine commercial operation should not and cannot take place on a publicly funded infrastructure.
Why do you think the Hasso Plattner Institute and the Berlin-Brandenburg location are the ideal place for the AI Service Center? What advantages does the location have over others?
HPI - located in Potsdam - is very well suited as a specific location for an AI Service Center because, on the one hand, we have the necessary research expertise, both in research on ML methods and on the operation and efficiency of AI infrastructures. Our many years of experience with the SCORE Lab also mean that we are well positioned to collaborate with external partners in such experimental environments. In Berlin-Brandenburg, but also beyond. In general, cooperation with external partners and technology transfer into practice is an important and lived part of scientific work at HPI, but always based on our own scientific expertise. And in addition - as a fairly new employee at HPI, may I say - I am always impressed by the enthusiasm of our students and staff.
I am therefore certain that with the AI Service Center, HPI can make an important contribution to bringing AI and especially ML further into practice in Berlin-Brandenburg and beyond.
The funding from the German Federal Ministry of Education and Research (BMBF) will now run for three years. What goals do you want to have achieved at the end of this time?
We want to use these three years to further disseminate ML in practice and establish best-practice examples that will make it easier to move from initial game examples to productive application. There will certainly be some spin-offs in the process, and our scientific collaborators will have achieved interesting publications and doctorates. But first and foremost, our mission will be to help interested partners from a wide variety of organizations to deploy AI and ML.