At Rise of AI 2025 in Berlin, a panel with Elisabeth L'Orange (Deloitte), Kenza Ait Si Abbou (FIEGE), Dr. Annika von Mutius (Empion, KI Bundesverband) and moderator Nikolaus Roettger (FutureStory) discusses how AI can benefit society beyond the circle of people already working with it. The panelists trace their own paths into AI, from academic mathematics and a university neural-network class to consulting and company-building, then debate whether the wider public needs to be excited about AI or simply needs enough trust and literacy to use it well. Von Mutius argues most business users do not need deep technical competence, only clear economic value, citing Germany's roughly six to seven percent AI adoption rate against far higher figures in the US and China. Ait Si Abbou stresses that legacy IT systems, not lack of enthusiasm, are the real barrier for many German companies, and that involving end users in building tools drives adoption. The panel also discusses Klarna's reversal on replacing call-center agents with AI, the risk of AI-amplified disinformation, and closes with differing personal optimism scores about AI's impact on society.
Key takeaways
- Kenza Ait Si Abbou says legacy IT, like FIEGE's 151-year-old AS400 warehouse system, blocks AI value more than lack of enthusiasm.
- Annika von Mutius cites a roughly six to seven percent AI adoption rate in Germany, versus about 60 percent in the US and China.
- Ait Si Abbou notes cancer diagnosis accuracy rises from about 60 percent for new doctors to 99.26 percent for a recent AI model.
- Ait Si Abbou says Klarna's reversal on replacing 700 call-center agents with AI shows adoption will 'zigzag' rather than progress in a straight line.
- One panelist says the disinformation network Pravda got ChatGPT to produce false claims about Russia in about 30 percent of cases reviewed.
Detailed notes
How the panelists arrived at AI
Von Mutius started working with AI systems around 2016 while finishing a mathematics master's and PhD, then worked in Silicon Valley before founding Empion in Berlin, a recruiting system that quantifies values, company culture and personality traits to match people to companies; she says she chose Europe over the US because of its value system, despite enjoying the enthusiasm for venture capital in San Francisco. Ait Si Abbou traces her interest to a university class around 2000 where a professor used cancer diagnosis with neural networks as an example; she wrote her 2003 thesis on clustering algorithms and says she was told at the time that AI meant an academic career, not industry. As CTO and board member at FIEGE she is responsible for IT, data, AI and information security, and describes her main work as reducing system complexity and building the infrastructure that AI needs to scale. L'Orange studied law, wrote a master's thesis on algorithmic trading using Goldman Sachs's algorithms, then founded a company, worked as a VC investing in deep tech, was a CFO, founded another company, and now leads end-to-end AI transformations at Deloitte.
Does the public need to be excited about AI?
Von Mutius argues that excitement is not a precondition for using AI, since many business users do not even realize AI is inside the tools they use, particularly for predictive systems; what matters, she says, is that a tool has demonstrable economic or societal value, for instance better time-to-hire and retention in recruiting. Ait Si Abbou counters with medical diagnosis figures: doctors leaving medical school diagnose cancer correctly about 60 percent of the time, experienced chief doctors about 87 percent, and a recent cancer model reached 99.26 percent accuracy, she says. She warns of a widening gap between AI-fluent people and those left behind, and cites a figure that 85 percent of pilot-to-project transitions in companies fail because organizations and employees do not adapt. Von Mutius adds that Germany's AI adoption rate is around six to seven percent, well below roughly 60 percent in the US and China.
Legacy IT as the real bottleneck
Ait Si Abbou describes FIEGE, a 151-year-old company, as running its warehouse management on an AS400 system programmed in RPG, so startups' plug-and-play AI pitches assume an API that does not exist; building that infrastructure has to happen before any value creation starts. L'Orange says the pattern is widespread in Germany, recalling a fund she worked for that still used 22-year-old CRM software and colleagues on Word 97 only a few years ago. Both frame the legacy debt as an opportunity: healthcare systems in parts of southern Germany that never digitized paper-based processes could skip straight to AI rather than going through a horizontal digitization phase first.
Building the bridge between the AI bubble and the public
Asked what connects the AI bubble to the general public, Ait Si Abbou points to writing accessible books, starting in 2020 while at Telecom, after running internal 'AI for everyone' workshops where employees arrived afraid and left relieved once she explained AI as applied mathematics such as regression. She recommends the free 'Elements of AI' online course developed by the University of Helsinki and translated into most European languages. Von Mutius adds that showing AI's weaknesses, referencing an AI incident database of failures, and comparing generative AI's output to a philosopher like Sartre, can reduce public fear of AGI or superintelligence scenarios she considers still far off. Ait Si Abbou also stresses involving end users, such as call-center staff, in designing tools from the start so adoption spreads through peer advocacy.
The Klarna reversal and non-linear adoption
Ait Si Abbou cites Klarna's announcement, around six months earlier, that it would replace 700 call-center agents with AI agents as part of its IPO story, and its recent partial reversal to rehire staff, arguing roughly 20 percent of call-center cases will always need humans and that running proprietary voice AI is too costly for many companies, which instead pay per-call fees to providers such as ElevenLabs, pushing large firms like Lufthansa and Deutsche Bank to build their own synthetic voices. She concludes that AI adoption will 'zigzag' rather than progress in a straight line, and that predictions of imminent AGI or superintelligence from tech executives are overstated.
Closing outlook on AI and society
L'Orange highlights AI's potential in healthcare and in translation tools that let people follow television or school curricula in their own language, including for migrants and children. Von Mutius is most optimistic about reallocating the German labor market, where many people sit in the wrong jobs despite high employment, and about AI's role in defense technology using real-time data, calling the topic more than critical for Europe currently. Ait Si Abbou expects AI to help address climate change, migration and resource scarcity by finding new materials and efficiencies. Asked to rate their optimism about AI and society from zero to ten, one panelist answers ten, L'Orange declines to give a single number, saying AI is a tool whose outcome depends on who uses it, and another panelist says five, citing a case where the Russian disinformation network Pravda got ChatGPT to generate false claims about Russia in about 30 percent of cases reviewed.
Quotes
It's not going to be a linear situation. It's going to be like zigzagging.
Kenza Ait Si Abbou
AI is not something that is just there because it's AI, it must have an economic value.
Annika von Mutius
We shouldn't forget that we are in the driver's seat.
Elisabeth L'Orange
Questions answered in this talk
What is blocking AI adoption in German companies?
According to Kenza Ait Si Abbou at Rise of AI 2025, legacy IT infrastructure, such as her company's 151-year-old AS400 warehouse management system running RPG, is the real barrier, not a lack of plug-and-play AI tools.
How accurate is AI at diagnosing cancer compared to doctors?
According to Kenza Ait Si Abbou at Rise of AI 2025, new doctors diagnose cancer correctly about 60 percent of the time, experienced chief doctors about 87 percent, and a recent AI cancer model reached 99.26 percent accuracy.
Why did Klarna reverse its decision to replace call-center staff with AI?
According to Kenza Ait Si Abbou at Rise of AI 2025, Klarna announced replacing 700 call-center agents with AI ahead of its IPO but later said it needed to rehire staff, since about 20 percent of cases still require humans and running proprietary voice AI is costly.