AI — The German Gap.

We need to talk. Talk about the growing gap in artificial intelligence in Germany.

A point of view on AI every 6 months called “We need to talk.”. In November 2024, talk about expansion and hindrances of AI in Germany. By Laurent Burdin, founder of trend specialist Space and Lemon Innovations.

The J curve is well-known in economics. It describes how a trend sharply falls down and then bombastically goes up again. It follows the letter J form, down and up again (1). This is where we are today with AI in enterprise.

A critical moment for the expansion of AI in society and in companies of any size, particularly in Germany.

GPT-4o

Fatigue is installing.

We need to talk and sort out a few things about how AI is being treated by people and companies — in a German perspective where GDP growth is sluggish — flat to negative estimation for 2024 (2). We’ve experienced a formidable adoption of AI since the launch of ChatGPT at the end of November 2022. In society, in universities and in companies. Many have started tests and put teams in place, mainly for obvious simple use cases like text or image generation, or search for documents with Microsoft Copilot, ever so present in German companies.

But after a while people are getting bored by it, sometimes disappointed with the results of the tests. Just using it for simple questions and answers. An AI fatigue has installed itself. There has been a bit of confusion as well: unclear measuring of results, low data quality, no dedicated teams, very few investments on literacy compared to 2023, and many new priorities when growth is the key conversation end of 2024.

A tough moment because AI and LLM are the promise of a next productivity gain and above all, because technology is making huge progress. End of 2024 applications of AI are all about enterprise — this is where the market is, B2B, not B2C. Fatigue on one hand and acceleration on the other. At the bottom of the J curve. However, we have no choice but go through the J curve and up again.

The litany of “responsible AI”.

At society and political level the attention for AI has completely melted. This is very German I observed (being French and living in Germany for so long): if they are not into the race in an area, and they aren’t, they bring weird arguments like “what about data protection” or “responsible AI”.

I recently attended an AI conference at the Hamburg Chamber of Commerce — a good one by-the-way. Then came the Hamburg minister for “Economy and Innovation” (Dr. Melanie Leonard) and the only thing she had to say is “Responsible AI, responsability, be aware of responsible usage of AI”. Just this, nothing else. No imagination on how AI can transform industries, bring better results and liberate employees from repetitive, tedious tasks. What AI is actually all about.

The EU AI Act is a regulation that enforces responsibility. Please, can we talk about something other than “responsibility”? For example about what AI will do for the better in the economy and society. Like the weekly Der Spiegel did with its latest cover “How AI cures diseases and holds us healthy” (26.10.2024).

In the meantime, an acceleration of technology.

Everyone has seen how impressive Large Language Models (LLM) perform, especially for those who were acquainted with the former models like NLP. Remember the 2018 chatbot era? Some still think “LLMs are just chatbots” — this is wrong. They are so much stronger and getting faster better. End of 2024 we’re in a race between 5 players: Open AI (with Microsoft), Gemini, Claude (with Amazon) and the two open-source models Meta Llama and Mistral AI. I’m excluding the China efforts led by Alibaba (3). A fantastic acceleration seen in the latest Open AI ChatGPT o1 model (4) that can build a chain of thoughts, reasoning it is.

In Silicon Valley the music around AI is all about “when will AGI (Artificial Intelligence) be reached and what about the next step, superintelligence?”. OpenAI co-founder Ilya Sutskever gathered one billion dollars to exactly build this and called his newco “SSI — Superintelligence”. This is exactly where x.ai is exploring with $6.5 billion gathered early this year.

AGI is in sight. A network of AIs that can reason and connect with each other. On paper it is feasible and OpenAI is marching in this direction. I enjoyed reading “Situational Awareness” by German-American AI specialist Leopold Aschenbrenner, giving great insight on the development steps ahead (5). One thing is sure, AGI will come with what we call AI-agents and the agentic economy. Agentic, for agent something.

A growing gap.

The gap between fatigue and acceleration is there in front of us. If we add the repetitive warning of “responsibility”, the gap will grow further. It is a specific German issue — a large economy with too few AI start-ups (Helsing in defense, DeepL in translation and Aleph Alpha as an integrator). Aleph Alpha presented itself as a contender in the LLM race in 2023 and has been catched-up by reality (no performant LLM, less fundings than claimed). Aleph Alpha is not a champion today, but does good work presenting down-to-earth use cases and bridging the gap.

The gap is also in many companies that have put a lot of effort into 2023 but have lowered priorities there. At the exact moment where it is getting exciting and where AI is extremely productive. With the AI-agent logic. Companies need to react and learn to overcome this difficult moment, the bottom of the J curve.

Shall we really call it agentic?

Agentic is the word of 2024 in AI, a family of AI-agents in enterprise. The agent overtakes tasks done by humans. A process, digitalised, automated by an AI that learns. There are two kinds of agents. The simple ones, the “non-agentic” that retrieve knowledge from a database. In some industries where there are tons of regulation, tons of forms it makes sense to have a toll to search and find fast. The other type of agent is “agentic”. They can build a chain of thoughts and a chain of actions. Gather knowledge, re-work it and pass it to another agent or a human. The idea of an agentic economy is a world of small well-trained agents that will connect to each other to perform a task.

A thing for companies, back-offices and later in the decade robots in factories and warehouses.

This is the next step of productivity — the AI-agents. A simple thing to build — a process flow. digitalised and bound to data based and a logic of actions. This is what’s ahead of us.

Key is dedicated resources.

A fatigue + an acceleration = a gap, at a moment where it gets really interesting for business. This is the main issue of the German economy and companies and deciders and every one of us.

Make AI a priority again in companies. This is the only way. In 7 points.

1. learning and updating every 6 months — everyone from CEO to employees
2. a dedicated team for AI and not diffused around the company. It’s not an IT or a data thing
3. dedicated time and resources (IT, data, budgets)
4. define and freeze use cases (small) and develop agents for that
5. a process to evaluate results
6. publish a “responsibility guideline” (with clear guidelines so that the subject doesn’t come back again and again)
7. never doubt that AI will be the next step of productivity gain. Agentic AI. And it will.

A major competitive advantage for those who’ve understood this. And up the J curve again. “I think that most people are underestimating just how radical the upside of AI could be.” says Dario Amodei (6), founder of Anthropic, the acknowledged alternative to Open AI.

Let us be happy and successful with AI. And love it, like I do.

Hamburg, 05.11.2024

Links

1. J curve. https://en.wikipedia.org/wiki/J_curve
2. Growth German economy. https://www.zdf.de/nachrichten/wirtschaft/iwf-weltwirtschaft-konjunktur-deutschland-100.html
3. China and LLM, Laurent Burdin. https://www.linkedin.com/posts/burdin_chinatech-china-llm-activity-7181290693363924994-u4yV
4. News OpenAI GPT o1. https://openai.com/index/introducing-openai-o1-preview/
5. Situation Awareness, Leopold Aschenbrenner — June 2024. https://situational-awareness.ai/leopold-aschenbrenner/
6. Machines of Loving Grace — Dario Amodei- October 2024. https://darioamodei.com/machines-of-loving-grace