AI in Europe: How do we position ourselves in global competition?

While 2024 saw strong growth in the use of generative AI in marketing and customer communications, I hope to see more developments in production, science and software development in 2025. Current research in the field of large action models (LAMs) could contribute to making it possible to control robots and complex systems using natural language. Scientific LLMs can become helpful tools in research, for example in drug development.

In a current project at the Bavarian Institute for Digital Transformation (bidt), we are investigating LLMs for generating code from natural language specifications. On the one hand, the challenge here is how to evaluate the quality of generated code; on the other hand, we are addressing the problem of the possible loss of competence among computer science students. We must ensure that generative AI and human expertise interact meaningfully in a co-creation process. The fact that content generated by one system is used or evaluated by another generative AI system without human review can lead to a dangerous and absurd detachment from reality.

When it comes to highly specialised products such as machine control systems, medicines or programme code, critical review of generated content can only take place if the reviewers themselves are highly competent in the relevant field. Accordingly, new didactic concepts are becoming relevant in all degree programmes that meaningfully support the acquisition of skills in the context of using generative tools. Human expertise and the ability to critically assess and correct generated content are indispensable. This is a huge challenge for our education system.

Open source is important for development progress. But not when it comes to data. Europe, and Germany in particular, should keep a close eye on its data. Thanks to its multilingualism and strong industrial production, Europe – and Germany – has a decisive advantage that we should not relinquish.

The AI Act can be a unique selling point that speaks in favour of EU products. Germany and Europe are very well positioned in terms of the explainability and trustworthiness of AI. It is now important to implement certifications quickly and in a practical manner. There are already very well-developed process models for companies, for example at Fraunhofer IAIS. There are libraries for robustness or energy scoring for the individual steps in this process. For the automatic generation of tests for trained models and the comprehensible presentation of test results, there is, for example, the Care Label Framework from the Lamarr Institute. Further scaling is important here, and with it the development of many best practice examples.

The promotion of machine learning requires better structures.

The idea of establishing a European network of talent is not new. It began with the networks of excellence, in which machine learning laboratories could meet for targeted discussions on research topics, projects, teaching and training – for example, at the ECML and then the ECMLPKDD. The networks of excellence were a catalyst for community building.

The next step was the idea of centres for machine learning. Each centre combines excellent research, international networking, best practice, graduate schools and computer resources for experiments. Close cooperation with companies is intended to inspire new industrial applications and start-ups. Talents are also trained here and promoted in diverse careers as top performers – not least through good permanent employment contracts. Networking these hubs will facilitate the exchange of students and lectures, promote collaboration in research and enable the sharing of algorithms, codes and data. This idea was launched in 2018 with the AI centres in Germany and France.

However, various framework conditions are hindering further progress even before the actual goals have been achieved: On the one hand, sufficient computer resources must be available for further expansion of the hubs, and these must be available in a variety of computer architectures that enable experimental environments. Integration into networked hubs would be the much-vaunted CERN of AI.

On the other hand, the integration of companies for best practice studies and the promotion of start-ups is not sufficiently supported and too fragmented in the application process. Long-term perspectives and corresponding positions are needed here. This also applies to training courses at companies, schools and training workshops, which are part of a hub's remit. AI.NRW is an example of a contact point that already exists in many places – but often only as temporary individual projects.

Finally, the international and regional networks that each hub has established must be adequately and agilely promoted. It's about more than just nice meetings and photos for the press!

If Europe now understands the urgency of significantly improving and expanding the network of AI centres in Europe and invests in human and computer resources, then the dream long cherished by most AI researchers could come true and Europe could position itself well between the US and China.

Reasonable AI enables the intelligent combination of different AI methods to create multi-paradigm systems. Reasonable AI systems use an appropriate amount of resources and are based on high-quality data. They are not only powerful, but also capable of adapting to new situations. Just as a delicious cake is not only created by many ingredients (data) and a great oven (model and infrastructure), but also by a well-thought-out recipe (the interaction of intelligent algorithms), AI systems must also be orchestrated sensibly.

Germany and Europe have great opportunities in this development. While companies in the US and China are currently focusing on scaling, Europe could go its own way with ‘sensible’ AI. Such a strategy would not only make economic sense, but would also address the growing criticism of the environmental and social costs of the AI industry. The increasing energy consumption of data centres can no longer be ignored. At the same time, it is becoming apparent that not only huge language models, but also specialised AI algorithms are bringing enormous benefits to science, medicine and industry. This is where European companies and research institutions could score points with innovative, highly specialised technologies.

Investments should be directed specifically at key areas such as neurosymbolic AI, adaptive learning methods and hybrid models. The combination of symbolic systems and neural networks improves the explainability and traceability of AI decisions – essential for regulatory requirements and social acceptance. Multimodal systems that link different data sources also offer promising opportunities. The future lies in the intelligent combination of existing technologies to achieve better results in a resource-efficient manner.

The AI competition is not yet decided. It is up to Germany and Europe to pursue a strategy that focuses not only on short-term effects through scaling, but also on sustainable, well-thought-out and long-term viable AI technologies. The future of AI lies not in gigantomania, but in reason.

Current dynamics of AI research and development

Advances in scaling AI, particularly machine learning based on neural networks such as transformers, have led to impressive results in areas such as image recognition, speech processing and robotics in recent years. Large language models, such as ChatGPT, Llama, Gemini, and Le Chat from Mistral, have demonstrated that AI is capable of generating human-like text and solving astonishingly complex tasks.

The popularity and importance of AutoML has grown significantly in recent years thanks to various open-source packages and cloud services. While it was still a hot topic just a few years ago, it is now evident that it has arrived in productivity environments. AutoML makes it possible to automate the development of AI models, thereby increasing development efficiency. It supports developers with little AI expertise as well as highly qualified AI experts in automating tedious and time-consuming tasks in order to find better solutions faster.

Another important aspect and trend that research in Germany is focusing on is human-centred AI. This places people at the heart of AI development and aims to create AI systems that are transparent, sustainable and user-friendly. Human-centred AI requires a rethink in AI development, moving away from pure automation towards augmentation and enhancement of human capabilities. In line with European values, this is also the path that AI must take in the future.

German AI research in international comparison

In international comparison, AI research in Germany is well positioned, especially in the field of basic research. Germany ranks among the top countries in terms of the number of AI publications in relation to the number of researchers. Germany has a long tradition of AI research and boasts excellent universities and research institutes, such as the AI Centres of Excellence, as well as other strong centres such as hessian.ai, the AI Centre at RWTH Aachen University and the L3S AI Centre in the Hanover and Braunschweig region. In Germany, there is also a focus on the ethical and sustainable aspects of AI, which will be essential for AI to gain widespread acceptance. In addition, Germany has a strong industry and a leading position in the medical sector – areas in which AI solutions are already being used today. Germany has advantages over the USA and China in these sectors, as large amounts of data and expertise are available here. This potential must be leveraged.

In contrast, Germany's weaknesses in international comparison are particularly evident in investment, risk appetite and bureaucracy. Compared to the United States and China, Germany invests less in AI research and infrastructure. In addition, German culture is too often characterised by a high aversion to risk and error, which inhibits the development and application of AI innovations. Last but not least, bureaucracy in Germany makes it difficult to establish and grow AI start-ups.

Above all, investment in AI research and development must be significantly increased in order to keep pace internationally, for example by promoting AI clusters and providing venture capital for AI start-ups. Education and training in the field of AI must also be expanded across all disciplines in order to meet the demand for AI experts. Ultimately, the ethical and sustainable aspects of AI must also be taken into account in all phases of development and application. This can be achieved, for example, through the further development and implementation of AI standards and the promotion of research in the field of ethical and sustainable AI.

Harnessing the potential of AutoML, human-centred AI and sustainability

In order to get ahead in the global AI race, Germany should take targeted measures and make proper use of its potential. Automated machine learning (AutoML), human-centred AI and sustainability are important aspects of the future of AI in Germany.

AutoML can accelerate and simplify the development of AI models, which is particularly beneficial for small and medium-sized enterprises, as it eliminates the need to build huge AI teams. By automating tasks such as model selection and hyperparameter optimisation, companies can develop and deploy AI solutions faster and more cost-effectively.

Human-centred AI ensures that AI systems meet people's needs and values and contribute to improving quality of life, both on an individual basis and for society as a whole. Rather than replacing humans with machines, the aim is to develop AI systems that complement and enhance human capabilities.

Sustainable AI solutions can help to overcome the environmental, economic and social challenges of our time. AI can be used, for example, to optimise energy consumption, increase resource efficiency and create sustainable supply chains.

The future of AI in Germany is promising. Through targeted measures and by harnessing the potential of AutoML, human-centred AI and sustainability, Germany can take a leading role in the global AI race and use AI technology for the benefit of society. A national AI strategy that promotes cooperation between politics, business and society is key to realising this vision. Only by acting together can we ensure that AI is used responsibly and for the benefit of all in Germany.