Surprises from Gen AI We Never Saw Coming

Welcome everyone. My name is Francesca Margherita Chifari and I'm part of the steering committee of the Future of Manufacturing that has been recently merged into the Industry for Open Europe Lab. So from now on, we all belong to the same organization. The only thing that differentiates us is the time zone, but nothing more. So I'm proud to kick off a series of webinars recorded in Europe. And today I have the honor of engaging in an open discussion, not with just anyone, but with a very experienced person in the field of digitalization. So, Peter Schauf, welcome, Peter.

Thanks a lot, Margarita. I'm very much looking forward to exchange today.

So you have recently embarked also on a new adventure, founded your own company. So first of all, congrats.

Well, thanks a lot. Yeah, it's quite an interesting journey, but I think it's a great time also to take such ventures. There are some discussions going on that soon potentially also small startups can scale significantly with, for example, the use of AI agents.

Yes.

Sam Altman, you know, the CEO of OpenAI, he even recently shared in a recording, in an interview, a bet running with his friends which one-person startup will reach $1 billion valuation at what point in time.

Hopefully it's yours.

I'm a little bit off from that yet, but it's still interesting times and I see a lot of potential of, well, ramping up business and exploring what is possible.

Cool. Yeah. So yeah, that's also the reason why I'm here, because I noticed that you're really into generative AI, which is a super hot topic right now. And I thought you would have been the right person to discuss the surprises from generative AI that we never saw coming, as the title says.

Perfect. Yeah, let's jump into that.

So, first of all, I would like to comment on its adoption, which has been incredible, right? Especially when it comes to the manufacturing sector, because we all know that implementing digital solutions takes It takes a very, very long time in this specific industry. And I mean, everyone is understanding the importance of collecting data, providing solutions, better service to your customers. But when it comes to that, it's always like a very, very slow process. And we do it already in automotive, for example, because our cars are connected. We do it at supermarkets. We can collect data, different information with loyalty cards. But when it comes to industrial machines, it's always scary for some reason for people. But something completely different, very, very special happened with generative AI because in a couple of months it became the technology that really everybody wanted to try out. So against all expectations also in the manufacturing industry. Why do you think that happened?

So let me also share some slides, some graphics once in a while to discuss something. So in this case, Just, I think this is a graphic everybody has seen in some form or the other. It just shows the speed of adoption of generative AI, and we will talk about different kinds of AI, hopefully, because I think that is always very important to discuss. And what you see on the screen is basically the comparison of how quickly a specific platform, that is ChatGPT, a specific solution of a company, has been adopted. By a broad range of people. And you see that 100 million users have been using ChatGPT in that case within 2 months, and that was tremendously rapid also compared to all these other digital solutions that had similar— not as well— similar kind of approach that scaled significantly globally, worldwide, because easily available through the internet. But you ask for industry, and I think, why, why is that important to compare and to see how rapid it was also for the private sector? Because first of all, I think here private sector and industry come together in a way that it's just so easy to be used, and people use it in their private life. And now, of course, they think about how can I use it in industry, how can I use it in manufacturing, for example, because I'm a guy working in manufacturing, and then I want to apply these things, these concepts that I know from my private life, also in that context. And one of the aspects that need to be considered here is also how easy it is to be adopted. And again, different types of AI— the classical AI was not easy to be adopted. You had to train it a lot within industrial settings for specific solutions. But in this case now, with the pre-trained generative AI, it comes with just a big pre-training basic foundation, and you can then apply it in different application areas very easily. That is something that is new. Yeah, you don't need to start from scratch. There's already a huge amount of knowledge available that comes along with these types of technologies. And that's, I think, it's very a big shift in that context.

Yeah, I would say it's a sort of revolution. And what I think was also very surprising is the very high number of different use cases that came up in so many different sectors or teams. For example, totally off topic, but I was recently listening to an interview with a teacher who was discussing how frequently now students are using ChatGPT to to complete their homework. And honestly, probably back in time I would have done exactly the same, especially to, to write essays. Um, but, um, she raised the question whether learning such a vast amount of information is still as beneficial as it once was. I, I still believe it is. Um, but I mean, we, we could discuss these for many different hours, whether it's positive or not, how to solve it, because that's what they were trying to do as well. But let's translate that to the, to the manufacturing world again. Again, um, while this could be, for example, a problem at school because new generations don't develop specific skills that we did, um, at some point I also learned writing essays myself. Um, it's a huge, huge advantage when it comes to training your workforce, for example. First of all, it could be exploited when it comes to training new employees. You could have all the internal knowledge base centralized in only one application and interact in more than 1,000 languages. And apart from training purposes, generative AI can be also applied to the shop floor level, to the customer service, to the sales force, can help the HR team as well. What interesting use cases do you have in mind?

Yeah, first of all, let me again show you something that was— is it already shown yet? So no, just one of the LinkedIn posts I did about the amount of use cases. And just see, let me show it to you here. It's a little bit better here from a quality point of view. So this is a study by, for example, Harvard Business Review, and it just shows the amount of use cases that there are. And it's not everything, yeah, from business applications to private life. And I think it just shows how much is possible because all of these 100 use cases are use case clusters, basically categories. And if you add them to themes, I think it's important to understand it's not about the finding the use cases. Yeah, because companies that really have a strategic approach to that, they have identified 1,000+ use cases partially in terms of how generative AI and AI as such can be applied. But I think it's important to understand, okay, what kind of themes can be used for these types of AI? And therefore, I thought this list is quite interesting. For example, it kicks off here with technical assistance and troubleshooting. Obviously, in the industry, that is a big topic, yeah, which is a day-to-day topic that needs to be addressed. And if you suddenly can help your engineers in the field with what you said already, the access to information, relevant information that helps them in troubleshooting and fixing problems, that's already a big step. What I think is also relevant, and we come to that, for example, when we look at the first category, the main category that was identified, creating ideas. It's in general about content creation and editing. What I think in industry is very interesting here is that you can have much more up-to-date and tailored also documents, manuals of how you run machines, applying different languages and making it very easy to transition from, for example, one manual that is for one specific device or machine to others in variations. And something like this will help also interaction and just better utilization of assets.

I mean, you almost don't need manuals anymore. I know that you have to print them, to create them, and deliver them to your customers because of the law. But if you can have all the knowledge base in one system and it really doesn't matter what language it's written in, then it's a huge advantage.

Absolutely.

Yeah.

And also, this is what generative AI can perform and can offer is this easy interaction with knowledge and making that knowledge much more available in a broad range. So that's, I think, is really a game changer because you suddenly, if somebody on the one side of the globe has an issue and that gets fixed, everybody else can through AI access this knowledge much more easier because with a manual you would always need to decide what can I put inside and what not. And now suddenly there's no limit to basically the knowledge that you're able to share because AI is able to share in a very dedicated way. But I think when you look at these lists, it's really interesting when you go just to the second one, which I was surprised about already. The second category is therapy and companionship. That is something I'm not yet so much into, I have to admit. Yeah, I'm exploring AI a lot. I was surprised to see that on second place. But as such, yeah, it's really worthwhile going through that, I believe, and deciding. But talking about industry, I want to look specifically at another industry post, maybe from LinkedIn. It's by Jeff Winters. Like Geoff, he's a very dedicated industry expert in terms of digitalization. He posts a lot of own content, and he made this very nicely, these categories for the industry specific. Yeah, very cool. And when you see that here, it's about connectivity, but also what he means with connectivity is customer-facing chatbots, for example. Yeah, what we just talked about, making knowledge easily accessible. It talks about generation, prediction, analytics, detection, control. And what I think is important in that context is these 2 types of AI. Maybe that's a good point to discuss because many people, when they hear about AI, they say, well, we had that in the past many years, decades even, of what we— when we addressed already and tried something out with AI. So what changed suddenly? And what changed was basically this generative AI, these pre-trained transformers, which just, I mean, they are based on knowledge that is created on a scientific paper in 2017. And since then, it really kicked off and is now having a strong impact in only, you must say, kind of these 8 years from a research paper to a global industry with billions of investments. And the difference again is the interplay. I mean, the classical AI, and we had to train a lot. And for example, when you look at, let's say, sentiment analysis, that was one of the areas where classical AI was trained with lots of datasets in terms of, well, what's the sentiment, for example, of a customer email that you receive in service or else? It's very valuable if you receive a lot of emails to structure them in terms of Are the people happy or are they really angry about the service? And to train on your dedicated use case was quite difficult because the machine builder had different service incidents and people writing a different way than, for example, an insurance company. So the transferable way to apply, for example, sentiment analysis was limited. So every company nearly had to train some significant amount by themselves. Now suddenly, these pre-trained transformers, they have so much knowledge, so much understanding about language, that sentiment analysis is now easy. They can do that. So these pre-trained transformers have basically substituted classical AI in that field. In other fields, it's not that automatically given. For example, in predictive maintenance, when you to analyze IoT data. That's not yet, at least, an area where generative AI is performing very strong and very reliable because there's still a lot of hallucination issues with this new type of AI. But I believe it's very interesting to go through these different sections, these areas for manufacturing, for example, and think about, do I want to apply classical AI? Do I have the data to do that? Can I test generative AI with all this knowledge that is available? Yeah, and just explore within these fields of manufacturing use cases.

You can also cross different use cases because, for example, you were talking about sentiment analysis but also supporting the customer service. So, you could actually see how the feelings of your customers, the satisfaction, sorry, of your customers improved over time while relying probably on generative AI to formulate the questions to the customers.

Exactly.

I will now take inspiration from something that you mentioned. So you were talking about companionship as the second category of use cases. And something else that I noticed that really, really surprised me is how much we do trust this technology, considering it's always available, it's easy to address, it's cheap, it doesn't judge. Some people ask for personal advice and ChatGPT or similar technologies are now seen as a sort of substitute of the psychologist. And this blind trust in technology can be dangerous in some contexts, for example, but it has also a silver lining when it comes to applications in manufacturing. And why is that? I mean, because we humanize generative AI, like we humanize technology. While, for example, maybe people were a bit skeptical about chatbots, I mean, I was spending time developing some tool. Now they really rely on the answers provided by virtual assistants based on the, on the GPT algorithm. And for example, customers are now happy to interact with it, and the customer service can save a lot of time because they're happy to interact with a virtual assistant, with a, yeah, an avatar rather than just a real person, because they perceive it as a real person as well. But what do you think about it? Like, why do we trust generative AI so much?

Yeah, I was also quite surprised, I do admit, in some contexts how much trust people already place in these types of AI because at the beginning, I mean, everybody is kind of careful, but then the answers are so good and so convincing. And there are some research papers out there which already said that AI is as good as people in convincing people, others, about certain topics. And so it's kind of this being very well related and very well spoken and having a lot of knowledge that kind of builds up trust. And for example, there's this platform Character AI where people spend hours just talking to AI. So that, again, this companionship topic, I couldn't yet realized it for myself. Yeah, I tried it out, but I'm always lacking the time to really get engaged with one of these characters.

Make it better.

Exactly. But I use it a lot for idea creation. Yeah. And also what new ideas it's providing to me, that's of big value. And when you transfer that to the industry, I do think there is not the absolute trust yet. Yeah. Because companies, and especially in manufacturing, for example, and factories, they need to have very reliable solutions. So that is just key. And this is the generative AI, which is more like a human person sometimes making things up and sometimes being very creative about solutions that might not be the best way to go about it. That is something where you need to be very careful. So it is important to understand where can I use this new type of AI? Where do I need the more classical, more reliable AI potentially that can give you a clear indication? I'm certain about 95, 85.6% that this is the right solution. Genetify is a little bit more a black box, I would say, for now. It's a big research topic to understand these types of AI better. And let me again show you something very interesting, I think, from a research perspective.

Yes, please.

I've seen that graph the other day, which you can see now. That shows very interesting research from Boston Consulting, but together also with Harvard Business School, MIT, and lots of universities. And what they did, they asked a lot of their consultants to use generative AI in that case and to see and evaluate how good they then performed. And what you see here is that especially in creative product innovation, their already very highly skilled, um, consultants, they performed significantly better. I mean, 40% is really not, not something minor or marginal. It's really a significant improvement of their highly skilled employees. But on the contrary, if you see on the right-hand side, for business problem solving, that is very interesting. The people who used AI performed worse than the ones using, not using it. And why is that? That was indeed because they trusted the result that the AI generated so much that they picked it above their own conclusions more often than not, and it was very often wrong. So it— I think it's very important to understand when can I use it, when can I not use it, where do I need to be careful. But also, I mean, this, uh, that's, that's actually a very interesting discussion about Can I use AI for business problem solving? And I mean, here it says basically, well, don't do it, it's dangerous. But I do think also there are ideas, and also we are following an idea where this can be actually also addressed and how you can then improve problem solving with AI. But for the time being, in any case, don't rely on the output of AI. It's rather kind of— it's a coach, it's an idea giver, it's a benchmark that you can use. Because it's so good in what it explains and how it explains it that people can fall for it, right?

Yeah, definitely that. Well, you can take inspiration from it but not trust it completely.

Exactly.

What I'm saying—

There are ways— sorry, but actually there are ways how you can then establish trust. For example, RAG is some technology aspect, Retrieval Augmented Generation it's called. That means that you base the data generation of the large language models, these generative AIs, on your available data that you define. And this gives it much more grounding and reliability than if you just ask it. And there are things like you could do prompt templating. I think it's a very interesting and still neglected area where you set up a template, and in the background of this template, which is then very easy to use, you already provide some examples to the large language model of what the output should be. So you can have much more reliable results when you do something like that. And obviously when you work with industry partners like 40 Factories, for example, yeah, which is using industry knowledge, which is having several use cases already implemented and a dedicated solution to industrial purposes, I think that is also helping a lot.

Yeah, definitely. I mean, this is going to be maybe another topic for another webinar, but also how many roles, how many different jobs is this new technology creating, because I've seen some job descriptions about, um, people creating prompts, and this is something completely new. Uh, but let's go, let's go back on track and, uh, let's move to, to the next topic. Um, what we also must be careful about is how we mix the, the virtual and, um, and the real world, because they're really slowly merging. And as we were saying, like, just before, there are people spending hours, for example, talking to the virtual systems. Um, this is also what's possible to be done on Character AI, for example, that you showed me a few weeks ago. Um, and you can interact with, uh, with avatars on the most disparate topics. I had a look and you can talk about and plan your holidays. You can prepare to start— pretend to start a war between Batman and Superman. I mean, uh, it's definitely fun at the beginning, but no, not healthy in the, in the long term. But as you can imagine, also the gaming industry has been completely revolutionized. And I have another fun example in mind before shifting again to manufacturing. I heard about, how is it called, Akihiko Kondo. It's, I don't know if you ever heard about this guy, but it's a Japanese man who got married to an avatar. So that's the level that we reached by mixing Real world and virtual world. This sounds like madness. Well, it sounds like madness to me, but not when it comes again to manufacturing. Because, for example, always if we talk about the customer service or what happens at the shopper level or during the training, when you mix virtual and the real world, you can, for example, provide suggestions while wearing a pair of a pair of smart glasses and you can perform a maintenance activity. So you could train your muscular memory and learn how to perform this activity. And then on the maybe left side, you have another screen and you can interact with your assistant. And this is definitely very, very useful. What do you think about it? Do you have any other use cases in mind? I mean, I also remembered a few A few months ago, you were mentioning how this can also be used by spreading the right culture within an organization, and you can create an avatar as well. So just tell me more about it.

Yeah, exactly. Also my topic and very interesting, kind of combining this virtual and real environment and especially virtual worlds. Now, you might have heard also about the metaverse, and this has been discussed kind of from a consumer and gaming perspective. And people have been quite disappointed about it. Yeah, because, um, I mean, Facebook, everybody knew it, suddenly called Meta. Mark Zuckerberg really put a lot of focus on that, and his stock price plummeted to an all-time low because he, he burned so much money for this idea which didn't really take off yet. However, I do think that the combination with AI and these virtual worlds Also, especially from an industry perspective, it has a lot of value. And I know about that a lot because I've been with Siemens quite a long time, yeah, until I then left. And maybe let me also quickly show you here the combination here with Siemens, also again in LinkedIn, yeah. So you might know those 2 people, Roland Busch from Siemens, and I hope I say him right, Jensen Huang from NVIDIA, which is of course the player when it comes to these new types of AI because they provide the CPUs, the GPUs, and all the compute power around it. And while this metaverse, these virtual worlds which are then supported by AI in the creation, in the interfacing, I believe they have, have a very high value. in terms of, well, you can suddenly simulate things that otherwise would be costing you a lot of money in such an environment. But when I— when we talk about it, I always have the feeling that people say, well, okay, that's nice for big companies like Siemens and NVIDIA, but for example, when you talk about the industrial metaverse— let me show you that one again from Jeff that he posted the other day— the foundation of the metaverse is actually digital twins. There are different types of digital twins, and every company can think about what types of digital twins could already help me now in terms of being prepared for such a future, which is still out there. It's not yet there, that's for sure, this industrial metaverse. Just to run you through, it can be a digital twin of a part because you have a 3D model, but not only 3D model, but also the physics behind it and maybe how it's manufactured and the electrical part of it. Then these parts go into products, larger systems, which are then implemented somewhere, maybe in a plant. They need to be manufactured, but also a plant is full of machines. And again, this can be by combining several digital twins, combining also with the processes of how they are running. develop in such a metaverse environment. Yeah, so therefore I'm, I'm very excited about combining this AI with virtual worlds. Yeah, that's also actually what my PhD is about still. I'm editing now my first draft, so it's not yet done, but I hope until the end of the year I'm, I'm finalizing with that. So this augmented reality and how do I use that, but also for organizational development, which you just asked me. Yeah, so personalized AI, it's again a big topic, and I think from now, from a timing point of view, I would love to dive into that a little bit more, but maybe we should shift that to also other discussions going into that direction of how people can not only use data which is already available, but rather use AI as a personal coach and guidance within a company that focuses also on culture and how this develops. So I think Very interesting topics, and I think we have a lot of topics actually we could touch upon, so therefore I would shift that a little bit to next time maybe.

We actually have so many, but well, unfortunately it's already half an hour and our webinar is coming to an end. And yeah, the time is too short unfortunately, but my last question would be, please give 3 suggestions to the audience for implementing digital solution based on generative AI, since it's also what you do for a living. So what would they be?

So one thing I believe for everybody to— it's important for everybody to understand, how do we differentiate classical AI, generative AI? And actually what is also interesting, RPA is robot process automation or just process automation. In the workshops that I do, I normally ask the people to bring in their ideas about AI use cases. And then it's very interesting for them to understand, okay, we have those 3 buckets and it might not be AI at all. It might be process automation for the time being. Then if you look at AI agents, that's going to be the way forward potentially, but not yet.

Yeah.

And also to differentiate what is classical AI, what is generative AI. So these 3 buckets, I believe companies and people need to work on differentiating those. to really have a better understanding when and how to use AI. That's one. Then basically start using it. That's definitely the second one. I have it always open in a browser window in terms of when I have questions, when I try to generate new ideas. It's so helpful already and you start exploring what it can do and what it cannot do. Sometimes I do too complicated requests and it actually breaks down. GPT-4, for example. It's interesting that there are some limits in terms of what it can do also from a complexity point of view, and then you need to formulate it more simple, break it down into different steps. So I think this is just an experience everybody should have. And then as a third item, do it in an ecosystem. Yeah, so companies should not do it just by themselves. I really think that exchange topics like platforms, for example, the Industry 4.0 Club, great example, and there are other clubs like that and also service providers that have experience in the industry, 40 Factory for example, and else. And so I think this collaboration exchange around this topic is very important, and I'm really hoping that we in Europe— we are already lacking behind, I do it, I must, we must be honest with ourselves. Yeah, the United States with all these models, they are far, far ahead, and also China is pushing a lot. And so in Europe, considering our normal economic strength, we are really lagging behind. So we should get going, we should be active here, we should not lose time.

I think we're doing that. We are riding the wave at the moment, especially on generative AI.

We do that in any case. Yeah, so very good. And I'm looking forward to these exchanges. Yeah, I mean, in Since last week, you can— I changed my LinkedIn profile a little bit. You can book appointments now directly with my LinkedIn profile. That's cool. Margarida, we have our MachineBuild X event coming up in June, I think.

Yes.

So this is something where you can also, the people from the audience, get engaged with us, meet us, talk to us. Let's do that together.

Exactly. On June 6th, it's a non-promotional event on servitization and digitalization, and I guess Peter will be one of our speakers again as well. So to conclude, and thank you so much, Peter, and everyone in the audience who joined us today. Don't forget to follow the Industry 4.0 Club on LinkedIn. You can find many different interesting webinars on Industry 4.0, women in manufacturing. There is fireside chat, people in society, And if you would like to join the community or have something to share as a guest speaker, don't hesitate to contact us. And well, I wish you all a lovely evening.