Feasibility study LEAM - Prerequisites for large AI models in Germany

from | 27 January 2023 | [at] News

We are currently experiencing the beginning of the second wave of the AI revolution, which started with the release of GPT-3 by OpenAI in the summer of 2020. On the basis of huge amounts of data and with an enormous investment of developer resources, money and computing capacity, OpenAI has created a language model that has been able to show a previously unattainable performance.

Feasibility Study "Large AI Models for Germany

Companies in Germany and Europe will also have to rely more on the use of AI applications in the future. What prerequisites are needed for this was examined as part of the feasibility study "Large AI Models for Germany", which was carried out in cooperation with Alexander Thamm GmbH, among others. 

Need for an AI ecosystem

The study, presented by the Large European AI Models (LEAM) initiative, shows that over 80% of the experts surveyed advise building an AI ecosystem and developing AI foundation models based on European values. Currently, the basic AI models that companies need to develop AI applications come mostly from the US and China. This poses major challenges for European companies because, for example, access to proprietary services is difficult and data is not processed in a GDPR-compliant manner.

Building an AI supercomputing infrastructure

"An AI service centre can be successfully established and operated by a joint initiative of business, science and politics in Germany," Jörg Bienert, partner at Alexander Thamm GmbH and president of the AI Bundesverband, is convinced. To this end, the study of the LEAM initiative now presents a concept for the establishment of a dedicated AI supercomputing infrastructure.

 A team of specialists operates a dedicated hardware infrastructure specialised in large AI models. It develops these AI models further and makes them available to others. In addition, the team collects and refines the data necessary for operation and applications and implements software and services around these AI models that simplify the training and tuning of large models and make them easily usable for different target groups.

Author:inside the feasibility study

The LEAM feasibility study was conducted in cooperation with Alexander Thamm GmbH, German Research Centre for Artificial Intelligence (DFKI), eco - Verband der Internetwirtschaft e. V., Fieldfisher LLP, Fraunhofer Institute for Intelligent Analysis and Information Systems (IAIS), Merantix Momentum GmbH, Simmons & Simmons and Ubermetrics Technologies GmbH. The KI Bundesverband e.V. was responsible for the overall project management.

The entire study can be read free of charge here: https://leam.ai/wp-content/uploads/2023/01/LEAM-MBS_KIBV_webversion_mitAnhang_V2_2023.pdf

As part of the study, an interview about foundation models with our founder and CEO Alexander Thamm was also published. 

Alexander, where do you use Foundation models? What is your use case?

The use of artificial intelligence is an important topic for our clients. We develop AI strategies, concepts and implement projects based on the latest scientific findings. In this context, language processing and foundation models are playing an increasingly important role and we are investing in the use of the technology in Germany, including active participation in the OpenGPT-X project. Our teams specialise in diverse areas of AI development and implement projects in the fields of image processing, natural language processing, forecasting, anomaly detection, among others. Examples include an AI-controlled system to support train dispatching at DB, robotic systems to support elderly care and novel procedures for autonomous driving.

What influence do AI foundation models have on your business model or your projects?

Foundation models are becoming central to AI applications and infrastructure in many areas. Currently, we often develop individual AI applications from scratch using specific customer data. In the future, there will be a shift towards transfer learning or tuning of existing, powerful foundation models. At the same time, the use of foundation models will open up new areas of application and we will develop applications for our customers that are currently difficult to implement - especially in the area of NLP. The market will grow and we see a great opportunity here for us, but above all for the competitiveness of the German economy. This has an intensive impact on our business model, especially if we were to depend on the use and licensing of foundation models that we can only access via APIs and over which we have no direct influence. If we can then only obtain these models from non-European providers, we also have to deal intensively with data protection and data security aspects.

To ensure that our customers and our company do not end up in a one-sided dependency, it is enormously important that we can also access foundation models that were developed in Germany or Europe and that we cannot only use them via APIs. At the same time, the consideration of European values, e.g. on the topic of bias, has enormous significance for us and our customers.

What difficulties and problems do you see that only the US and China are currently providing AI foundation models comprehensively?

US internet companies are currently investing heavily in the development and dissemination of foundation models. By making them available via APIs, the first services on the market can collect a lot of data at the same time, e.g. about the focus of use. There is thus a danger that monopolies will form here again - as with search engines - and an increasing technological dependency will arise. If the central AI applications only come from overseas, in the long term our activities will be limited to the design of frontends and workflows. We will have little or no influence on the models, which is problematic especially in terms of quality and bias. Thus, this development could also become a potential threat to our current business model - and to our corporate mission to ensure the competitiveness of the European economy in this area.

How would European models - open source, covering all European languages, with high data protection standards and minimal bias - help you?

European foundation models, which we could use as a basis for our AI developments, would enable us to continue developing innovative applications in the future. Since we would not only have access via APIs, but would have the models available as open source, we could also intensify our own research activities in many areas and build state-of-the-art AI systems for our customers. This would ensure that we can continue to optimise processes for our customers and enable new products and business models in the future.

Author

Michaela Tiedemann

Michaela Tiedemann has been part of the Alexander Thamm GmbH team since the early start-up days. She has actively shaped the development from a fast-moving, spontaneous start-up to a successful company. With the founding of her own family, a whole new chapter began for Michaela Tiedemann at the same time. Hanging up her job, however, was out of the question for the new mother. Instead, she developed a strategy to reconcile her job as Chief Marketing Officer with her role as a mother.

0 Kommentare