Hi everyone, thanks for being here. My name is Allison Snider, I'm a managing editor here at Axios, and with me on stage... we're gonna keep going. This is Peter Koerte, nice to meet you. Thank you so much for doing this. He is Siemens' Siemens AG's Chief Technology and Chief Strategy Officer and a member of the Managing Board. So thanks for doing this, we really appreciate it. I think we're just going to dive right in. My first question for you is: Siemens has sort of long been associated with transportation, healthcare, energy, and you're trying to make yourself over as an AI company. How are you going about doing that and how important is that here in Davos this week?
Well, it really matters because if you think about what we need most is good employment, good productivity, good growth, and we believe that technology actually can bring that. So the manufacturing sector per se is really suffering from a shortage of labor. No matter if you look in all the industries from chemicals to automotive, all the way, we see that the people are missing that really want to work in an attractive industry. And we believe that actually these are the true heroes if we give them the right technology to do so. So at Siemens, we've done electrification in the beginning, automation, but now we go on the next level, which is all about digitalization, and of course digitalization infused by AI because of all the data that comes from all the machines, from all the buildings, from all the trains that we run.
Jake just said something provocative twice about AI and SaaS companies going extinct if they do not have data. Siemens is data rich, right? You have 300,000 employees working all over the world. What's your challenge though with the data that you have? Yeah, start there.
No, it is fair. We know in particular in the industrial space, we know that there are 17 billion devices that are actually receiving and sending data. 17 billion is twice as many people on this planet, which means fundamentally we don't have that issue that we are running out of data as training for the large language models. It's not the case in industry. We have a different challenge, and that is data access, where truly companies have to trust each other to share the data so that you actually can make use out of it. And second is of course no hallucination or anything has to happen, because if you think about it, if your AI gets you the tone of your email wrong, that's one thing. If you think about programming a machine, running a train, running a building, that's an entirely different thing. That's why we say industrial AI needs to be three things: it needs to be safe, it needs to be reliable, but most importantly it needs to be trustworthy.
So Siemens, working with Microsoft, you've developed something you call an industrial co-pilot. Can you take us to the factory floor, whether it be again transportation, healthcare, somewhere? What does it look like? What is it doing? Who's working with it? And how's it going?
Yeah, if you today are an industrial engineer or somebody who has to actually set up these machines, you speak in a very specific language. And it just so happens that most of these people that are able to speak that language are just leaving the workforce because we see all the boomers going out and the younger generations coming in, and for them that language doesn't mean anything and they have no appetite to pick it up. So Python and these kinds of environments where you can program in a state-of-the-art environment, that's what they're looking for. So we said, what if we could make it go away all together and you just tell the machine what actually needs to be done? So instead of writing automation code, you just tell the machine, the robot, to pick this glass and then put it over there, and then actually all the code is being generated. So that's for setting up the machines, which is one thing. But the harder part is, we just learned and I came across a statistic which really stayed with me: the average tenure of the factory worker level is dropping from 20 years down to three years, and that is only since 2019. So by and large, you have a lot of people that have no clue actually what's happening. So imagine you're working on a factory floor and it's only you, nobody else, and the machine in front of you breaks down, but you're the guy who actually is responsible for making it run. What do you do? And if you don't reach your manager or your colleagues, then you have a big issue. So we decided that this industrial AI, we can actually democratize that knowledge. We can bring it, we can learn it, we can teach it, and then really tell the worker whenever that is needed, 24/7, even on a Sunday morning graveyard shift, to say, 'You know what, this machine is broken down because that ball bearing maybe was out of center, and so therefore you have to take the following one, two, three, four steps.' And that is fantastic because with that we can capture the knowledge as people exit the workforce and infuse it as people come onto the shop floor.
And what is that in terms of productivity? What do you see productivity-wise, efficiency-wise? What's the payback for you so far?
It's early, as with most of these AI applications. We know that we have about a hundred companies that are working with it right now. We see that for sure the time it takes to troubleshoot any kind of failure is much, much faster, the quality of it is faster, but we are still lagging these large-scale studies that obviously give you the productivity increase of whatever percent that is. But at the end of the...
Day... um, you know, actually recently I think there was a WEF report that found that 41% of businesses said they were going to cut jobs because of AI. So there's a real threat to workers. How do you balance that with the optimization that you're trying to get, and what does it mean for them?
Yeah, the companies that we work with, actually it's quite the opposite. So the report I know is that until 2030 we're going to miss 85 million people in factory jobs or in work jobs, so industry jobs, which means that we have a shortage of labor. And it's not so much about the fact that AI is replacing them, because we believe that actually it more augments every factory worker. So we give that knowledge to the people that are coming online, and we haven't seen that yet in terms of that threat in particular where people say, 'You know what, the AI is going to take my job away.' You don't see it. I don't believe it. And I don't believe it for the very reason: if you think about let's say 20 years ago with robots, it was the same conversation. We talked about robots being on the shop floor and taking all the factory workers away, and then we have a dark factory. Run around the factories, the factories I know are definitely different than they used to be 20, 30 years, but there are still people around, except that of course the qualification has changed. And that is really now going into the areas where you need problem solving and all the skills we as humans are good at, as opposed to the repetitive tasks which obviously the robots, and to me AI is nothing else than robot in the digital age, is really doing so.
When you develop something like the industrial co-pilot, are workers at the table? Are they involved in those conversations? What's the input that they give?
It's both. So we co-developed it of course with the data in particular from the shop floor, and therefore also we needed to know in which environments they are being created. But then also now as we roll this out, we continuously refine that. For example, the question of that human-machine interface: how do you actually consume the data, where do you consume the data is a very important part. So do you go to your smartphone, do you go to the machine itself? Will we eventually see those glasses right that actually will tell you, you're looking at the machine and then of course AR/VR just prompting gives you the advice of what to do. These are the kind of things where it's very fluid. So we just launched it half a year ago, but that is all happening as we speak.
What do you need, or what's needed for industrial AI to fulfill its promise? What's needed technically in terms of AI advances?
So technically, the most important part, it comes back to it's the data. It's not the models, it's not the compute power, it really is the data. And for the data, we need to have high quality data, that means the edge cases, so that we have that ability to guarantee for a very reliable system. And that is of course after all something that, you know, there are circumstances. You have to envision to make it really industrial grade and hard. We have some of those operations. For example, we have streetcars that actually drive autonomously and they are AI powered, so we really took it all the way to the end. But it takes longer usually in industry than it takes in the B2C space.
