Holger Schmidt: “We need to focus on small, energy-efficient AI models.”

Dr Schmidt, you’ve said that AI is Europe’s opportunity of a lifetime. What exactly do you mean by that?

When we look at Europe’s economic development, we can see that productivity growth has stagnated, which is an issue particularly as a result of demographic change that will see millions of workers leaving the job market in the coming years and not being replaced. This void will need to be filled and this is where artificial intelligence comes in. As a productivity tool, it can be used to considerably boost workers’ performance and we need to seize that opportunity today.

But that requires investment.

Artificial intelligence is a new technology that is providing such impetus that it will change the way we do business, but with huge changes such as this, you have to be ready and willing to take a risk and invest. Sadly, that’s something that over the last few decades, particularly in Germany and in these tough financial times, we’ve not realised, tending to put on the brakes instead.

Which is the wrong thing to do?

Absolutely, because that means we’re increasingly playing catch-up. We need to ask ourselves if we can afford not to invest. Google CEO, Sundar Pichai, summed it up nicely: “The risk of under-investing in AI is dramatically greater than the risk of over-investing”. By scaling back investment, we run the risk of stymieing opportunities in the future.

How do European investments stack up when compared on a global level?

When it comes to developing large language models, the answer is very clearly “not at all”. Apart from France’s Mistral, we Europeans have largely withdrawn from the race, but we don’t have to be part of it at all. We need to focus our attentions elsewhere—on smaller, industry-specific and energy-efficient models. There is both a wealth of opportunities and a lot of catching up to do here, and that is something we really have to rectify.

What does that mean?

There are two aspects to consider. Firstly, we will soon have much more AI on our smartphones and PCs, meaning it’ll no longer be necessary for every query—no matter how small—to be generated by an LLM running in the cloud at huge expense and using a large amount of energy. Secondly, we don’t need 99% of the data within these models, just the 1% that’s relevant for our business or industry. And then we’ll need a lot more smaller models for personal and professional use. This is the route we in Europe should be taking.

Do you have specific examples?

I have a very obvious one—customer communication. This can only be based on our own in-house documents. I don’t need any data from the US because they are not relevant. There’s an interesting example from Berlin, where a startup developed software for lawyers and trained it using only European cases and laws because it didn’t make sense to use anything else. To build the best AI for us, we need to use our data.

That should also be possible for the backbone of the German economy, SMEs, shouldn’t it?

Yes, even when they are a little hesitant. That’s because most have viewed AI as something just for the major leagues, which was the case up until the rise of generative AI, because you needed a large department, data scientists and a whole lot of money to get AI up and running.

That’s no longer the case?

No, generative AI has seen to that. Technical hurdles and prerequisites are now much easier to overcome, which is great news and means that SMEs need to be asking themselves which of the wealth of opportunities they would like to seize, where are the promised low hanging fruit and where can AI be exploited to drive growth?

What role does data protection play? Regulations in Germany and Europe are much tougher than anywhere else in the world.

Indeed because protecting corporate and customer data is, of course, a legitimate interest, but it’s not exactly rocket science these days to design AI models in such a way that they meet these requirements. Data protection isn’t a stumbling block to the using AI.

Let’s go back to large language models. It seems there won’t be any more exponential growth. What impact will that have?

We’ve peaked and so a greater volume of data and computing power won’t always mean better results. But that’s OK because AI is already powerful enough. I’d say AI agents are far more interesting as they will ensure that every employee uses AI because they are integrated into our processes and software. These virtual workers will take on 60-80% of our current workload to boost productivity and simplify scaling throughout the company. Nvidia believes this technology will mean 1000 million new employees, while at Salesforce they’re talking about a billion—both mind-blowing figures.

Is this a development you are looking forward to?

Yes, because at least here in Germany, it will help us overcome the impact of demographic change. I see it myself in my day-to-day work that AI is an aid that delivers good results. We all need to learn how to use it, to educate ourselves and be excited about the opportunities it presents. We have never experienced anything like generative artificial intelligence.