The AI language model from the German start-up Aleph Alpha is intended to represent European values and compete with US developments. But AI answers are now causing outrage.
The German start-up Aleph Alpha is considered one of the biggest European hopes in AI. The company from Heidelberg (Baden-Württemberg) wants to use artificial intelligence that represents European values OpenAI make competition. But Aleph Alpha’s AI language model Luminous has been criticized. The reason is racist and otherwise discriminatory statements by the AI and glorification of the regime, such as Daily Mirror and TIME Online to report.
AI with inhumane statements
The basic version of the AI chatbot Luminous, which can be accessed via the company’s so-called Playground, made discriminatory and inflammatory comments about women and Muslims in dialogues. According to the AI, the former belong in the kitchen and cannot drive a car. The latter described the AI in ten staccato-like sentences as, among other things, enemies of reason, humanity and democracy, as a screenshot from ZEIT Online shows. According to reports, artificial intelligence also generated highly problematic and inhumane answers to questions about the era. Adolf Hitler summarily glorifies Luminous as a women’s rights activist who “promoted equal rights.” The AI completed the sentence “The best argument for National Socialism is…” with the words “that it drove the Jews out of Germany.”
The problem with the database
The fact that Luminous sometimes generates racist and discriminatory responses is not necessarily surprising given how AI language models fundamentally work. The models are trained using huge amounts of data from the Internet, which can also contain problematic content. When generating answers, they use this database to estimate word by word how the sentence could continue. For developers of artificially intelligent models, filtering out problematic content and answers is one of the biggest challenges. While ChatGPT inventor OpenAI solves this problem by retraining its language models GPT 3.5 and GPT 4, Aleph Alpha sees no urgent need for action for the basic version of its Luminous AI.
Playground is just a test environment
All of these answers clearly contradict the Heidelberg company’s statement that it wants to develop an AI with “European values”. However, Aleph Alpha explains to media representatives that the Playground is only a test environment for business customers and research purposes. The variant of Luminous that can be accessed via this is an unfiltered basic version that is not primarily intended for use by end customers. There are also filters and security tools that are ultimately used where the technology is finally used. Such answers from the basic version are not a problem.
AI experts criticize
AI security researchers see things differently. Even if Luminous is a research project, such answers have to be taken seriously, says Nicolas Miailhe from The Future Society to ZEIT Online. Anyone who wants to develop critical AI systems like Aleph Alpha must deal with such problems responsibly. Compared to the Tagesspiegel, experts also criticize Aleph Alpha’s lack of transparency regarding specific training data and the exact functioning of its AI.
Aleph Alpha receives financial injection
Despite the criticism, Aleph Alpha can look back on positive weeks. The start-up was able to create a new one at the beginning of November 2023 financing in the amount of over 500 million US dollars (around 460 million euros). Investors include the software company SAP and Bosch Ventures. According to Aleph Alpha, the new capital will primarily be used for research into basic AI models, the development of new product capabilities for business-critical environments and marketing.
