Digitalisation & Technology, 05 June 2024
AI made in Germany
Cologne and Heidelberg instead of Silicon Valley
Thanks to Open AI and its ChatGPT, AI applications have been experiencing unbridled hype for months. Much of the focus has been on generative AI and the big tech companies, particularly from the USA. Almost unnoticed, however, “AI made in Germany” has also developed into a seal of quality - and has a lot to offer in many important areas.
The flagship AI companies in Germany do not come from the start-up centres in Berlin or Munich, but from North Rhine-Westphalia and Baden-Württemberg. Nor do they cater to the growing demand for content that is generated by artificial intelligence instead of humans, but instead aim to solve very specific problems.
DeepL
ChatGPT, Midjourney and Google Gemini would certainly be among the most frequently mentioned AI tools in a spontaneous survey of known AI applications, but DeepL? Probably not. Yet millions of people around the world already use AI translation from Cologne - both privately and professionally, and the trend is rising. The development of DeepL can certainly be described as astonishing and was also worth an article on //next.
The translation service, which emerged from the online dictionary Linguee in 2017, has not occupied a new market, but has taken on the heavyweight Google and its translation service. Despite this notable competition, DeepL has quickly made a name for itself for high-quality translations and established itself in professional circles.
Since January 2023, DeepL Write, an AI writing assistant, has also been added to the Cologne-based AI company's offering. The special feature: unlike ChatGPT, DeepL Write does not write itself, but optimises texts that have already been written.
Aleph Alpha
The city of Heidelberg's ‘digital citizen assistant’ goes by the name of ‘Lumi’. It provides visitors to the city portal with information about the city's services, provides information about planned measures to reduce CO2 emissions and helps interested parties to plan a visit to Heidelberg.
Lumi is the first milestone of the Heidelberg start-up Aleph Alpha, which is developing an innovative AI system. Unlike GPT-4, for example, which uses OpenAI for ChatGPT, the AI model is to be based on European data protection regulations. To this end, Aleph Alpha is developing its own LLM called ‘Luminous’, whose responses are intended to be traceable and transparent. Also unusual for the billion-dollar AI industry: Luminous' code is publicly accessible and Aleph Alpha is not aimed at the mass market of end users, but at companies and public authorities.
There are already very specific examples of AI applications. Even in areas that one would not necessarily associate with digitalisation. Baking bread rolls, for example.
AI as a concrete solution for everyday life
Many discussions about AI revolve around the almost infinite possibilities offered by algorithms, machine learning and deep learning. However, much of this remains stuck in theory and future prospects. Yet there are already very specific examples of AI applications. Even in areas that one would not necessarily associate with digitalisation. Baking bread rolls, for example.
The bakery trade is not only struggling with high energy prices - the shortage of skilled labour has also been a problem for the industry for years. For bakeries with several branches and sales outlets, calculating the optimum product quantities is a further difficulty: if too little is produced, customers inevitably migrate to the competition; if too much is produced, overproduction causes unnecessary costs.
This is precisely where the start-up BäckerKI comes in. It relieves bakeries of the often time-consuming task of calculating the optimum product quantities and optimises production planning. Bakeries save time and at the same time optimise their income and expenditure by producing neither too little nor too much baked goods. This also increases sustainability, for which the overproduction of food is particularly problematic.
BäckerKI is not only a successful example of the practical application of artificial intelligence, the development process could even serve as a blueprint for AI applications with direct practical benefits. The AI system was developed in close cooperation with various bakeries and is therefore perfectly suited to their requirements.
So instead of developing an AI solution and then looking for suitable problems, the founders of BäckerKI took the opposite approach.
What has actually become of the robots?
The Star Wars robots C-3PO and R2-D2, WALL-E or Ava from Ex Machina - they all embody a futuristic vision in more or less humanoid form. In the real world beyond the Hollywood studios, in-tensive research into robots has also been going on for years. However, communicative interaction with humans, as presented to us in science fiction films, has so far been too great a hurdle. NEURA Robotics could now overcome this. The start-up from Metzingen in Baden-Württemberg has devel-oped cognitive robots that not only perform various tasks thanks to AI, but also understand and learn from them. Neuro itself describes this development leap as a ‘new era of collaboration be-tween man and machine’.
How does today's weather forecast affect the number of curry sausages ordered in the ERGO canteens in Berlin? Do colleagues in Munich eat more salad before the public holiday? How do people in Cologne feel about bibimbap on the first day of the school holidays? At ERGO, the artificial intelligence (AI) of the start-up Delicious Data provides reliable answers to these and other questions in real time - and the quantities produced are immediately adjusted in the kitchens. “Thanks to the use of Delicious Data, we can organise our operational production processes more effectively and thus avoid food waste,” says Barret Jähn from the insurer's catering subsidiary, ERGO Gourmet. The team already won an internal competition in 2020 with this idea to avoid food waste.
AI language models depend on language
Does a German or European language model even make sense as an alternative to ChatGPT, Google PaLM or Meta Llama? After all, the well-known AI models already work very well, even in the local languages. Nevertheless, the answer is yes, an AI language model from Europe would have several major advantages. Firstly, it would not only be based on English, but also on German, French and Spanish, for example. After all, a language is always an expression of the respective culture with all its peculiarities, values and norms.
Secondly, a European LLM would also be a technological statement: ‘Look, we have the knowledge and the will for our own AI model!’ And if it also bears the ‘AI made in Germany’ seal of quality, all the better. Becoming less dependent on US technology companies is certainly not a bad idea. This is all the more true as it is questionable how compatible the US AI models are with the EU's “AI Act”.
Text: Falk Hedemann
Your opinion
If you would like to share your opinion on this topic with us, please send us a message to: radar@ergo.de
Further articles