Some people might not associate the word “trust” with artificial intelligence (AI). Stefan Jockusch is not one of them. Vice president of strategy at Siemens Digital Industries Software, Jockusch says trusting an algorithm that powers an AI application is a matter of statistics.
This podcast episode was produced by Insights, the custom content arm of MIT Technology Review. It was not produced by MIT Technology Review’s editorial staff.
“If it works right, and if you have enough compute power, then the AI application will give you the right answer in an overwhelming percentage of cases,” says Jockusch, whose business is building “digital twin” software of physical products.
He gives the example of Apple’s iPhones and its facial recognition software—technology that has been tested “millions and millions of times” and produced just a few failures.
“That’s where the trust comes from,” says Jockusch.
In this episode of Business Lab, Jockusch discusses how AI can be used in manufacturing to build better products: by doing the tedious work engineers have traditionally done themselves. The technology can help engineers manage multiple design variations for semiconductors, for example, or sift through routine bug reports that software developers would otherwise have to manually review to figure out what is causing a glitch.
“AI is playing a bigger role to allow engineers to focus more on the real, creative part of their job and less on detail work,” says Jockusch.
Also in the episode, Jockush explains how AI embedded in products themselves have already won over millions of people—think voice assistants like Siri and Alexa—and will someday become such a common component that people will barely talk about the value or the future of AI.
“I mean, how many discussions do you have nowadays about the value of Excel, of cellular calculation, although we use it every day?” says Jockusch. “Everybody uses it every day in something, and it’s so universal that we hardly ever think about it.”
Laurel Ruma: From MIT Technology Review, I’m Laurel Ruma. And this is Business Lab, the show that helps business leaders make sense of new technologies coming out of the lab and into the marketplace.
Our topic today is artificial intelligence and how it helps companies build products. With highly focused simulations that can be run in countless ways, an enormous amount of data can be collected, analyzed, and then used to make business decisions that help humans build better products. And that will be the difference in a highly competitive market. Two words for you: trust statistics.
My guest is Dr. Stefan Jockusch, who is vice president of strategy for Siemens Digital Industries Software. He is responsible for strategic business planning and market intelligence, and Stefan also coordinates projects across business segments and within Siemens Digital leadership. This episode of Business Lab is produced in association with Siemens Digital Industries. Stefan, welcome to Business Lab.
Stefan Jockusch: Thanks for having me.
Laurel: Could you give us a sketch of your background at Siemens Digital and what you’re working on now?
Stefan: Absolutely. Yeah, so our business is the technical software business in Siemens, and the software we make supports the whole process of the initial idea of the product to all the way through the manufacturing of that product, and then including the mechanical part, the semiconductor, the software running on the device, the sensors, and then also the operation of the product.
So one of our pieces of the portfolio is an IoT [internet of things] platform, where the product then basically feeds back information about its behavior. So, all of this. And what we like to think of is our software really builds a very complete digital twin of what we use every day. And the digital twin, as I said, includes everything from the idea to the design to the manufacturing process to the operation.
Laurel: So your days are very busy. How do you fit into this entire operation?
Stefan: Yeah, my own job is, as you said, for strategy. So in strategy, we, of course, look at the overall business plan for the business. We look at our competitors, we like to understand what they do. We look at the market around us, which is a very big and complex and very dynamic market. Also, of course, we have some initiatives at all times. We look at some aspect of our business, how it will evolve, how we might have to change our business model, how we have to transform our go-to-market model, how we interact with customers.
As you know, in the software space, there’s a lot going on these days, where we move away from having software that you install with a CD-ROM or a flash memory, and you more and more expect now to find yourself around the cloud. So, all these kinds of things are aspects of our environment that keep us busy.
Laurel: In a discussion about AI, it inevitably comes up that people are fearful about it, whether they’ll lose their jobs, whether it’s here to actually help humans or some Terminator situation. But we like to take an optimistic and a forward-thinking look at how artificial intelligence works. So, when we do discuss it, I like to always really set it in a scene thinking about humans and keeping humans in the loop. As AI learns and processes data, how do you then frame human-centric AI versus a more nefarious machine-centric AI?
Stefan: I personally have a huge privilege in that discussion, which is that I did my PhD work about AI and machine learning. And that is a long, long time ago in the mid ’90s. So in the mid ’90s, it was a big topic, all this whole thing of intelligence that’s encoded in these algorithms. And there was probably the same discussion back then, “Is this going to take over? Are we going to be so perfect in automation that we don’t need any humans whatsoever? And aren’t machines becoming not only more intelligent, but only even more creative than humans ever can be?”
So that discussion is at least 25 years old, probably much longer. And nothing of that sort has happened. I would even say, after I was done with my thesis, the interest in all this machine learning stuff probably flattened out, and I would say in the last five to 10 years, it re-emerged.
And basically, that is because the compute power that we have today to do even simple things, very simple things like recognizing language or recognizing a face on a camera picture that this is very doable now. But in terms of computers becoming really more humanlike or dangerous to humans in terms of being able to be creative, I don’t think we have seen any of that. And this is now going on 25 years, so I personally believe we should be safe for another 25 years, at least.
Laurel: People will be very heartened to hear that. But it does bring up a good topic, which is trust. And where are we with AI and trust and what AI can even do today?
Stefan: There are very different opinions, I would say. And one of the reasons why the opinions are also different is that most AI algorithms don’t show you exactly how they reason. Basically, you present AI with tons and tons of data, with so much data that you cover every possibility of what you’re looking at. And if it works right, and if you have enough compute power, then the AI application will give you the right answer in an overwhelming percentage of cases.
So if you look at stuff like face recognition that’s now being used to even unlock your phone or stuff like that, so we just get to a huge reliability. And as I mentioned this example, we start to trust the technology so much that we give it jobs like recognizing identity, which is a very critical application.
So, there is a trust that’s really justified by statistics, if you want. So probably whatever company—I think it was Apple who first came with that face recognition to unlock your phone, they start trusting their technology after they really have been able to test it millions and millions of times and haven’t gotten more than a few misreads. So, that’s where the trust comes from.
Many people are still a little bit worried about it because you never can tell how exactly AI works, because you can say, “Well, it’s the information encoded in about five million parameters. This is how it works,” but you can’t exactly tell.
And I know a few experts who believe more in other learning paradigms that give you a more deterministic way and are a little bit skeptical about the classic machine-learning algorithms that others use. But frankly, my answer is as long as you know your data set and you can test it and you get statistically a hit rate of 18 nines after the decimal point percentage, then you can trust the algorithm.
Laurel: Excellent. So when we’re thinking about a company like Apple, which is probably the best example when thinking about human-centric products, how does AI fit into a product lifecycle now in 2020 compared to five or 10 years ago?
Stefan: Compared to five to 10 years ago, I try to think back myself on all we had and what we didn’t have. Because I would say in a certain modest extent, we probably had AI embedded in a lot of everyday products, again, without knowing that we have them. But, of course, that has increased dramatically, and we just briefly talked about this example of face recognition. You can say that all these smart assistants that we use today, whether they are called Siri or Alexa or Google or whatever their name is, but, of course, that’s a massive application of AI technology that we are actually getting used to.
So yeah, and it’s really becoming more and more of what identifies a product. I think that’s probably the big shift in the last years, where we really go after, what is that experience as a user? How does our product behave in a really smart and helpful and intelligent way? And that’s what ultimately, I think, creates a lot of our desire to have it and our loyalty to the brand.
Laurel: So if you are one of these engineers who are trying to build this smart and unique product, where do you see AI being integrated to help those engineers and product designers make the best product they possibly can?
Stefan: Yeah, that’s getting big, actually. So, AI is basically very good when it comes to taking over heavy-lifting type of work and to allow the engineer to focus on real creative work. And you wouldn’t imagine how much heavy lifting work an engineer has to do every day. One thing that actually we put in our software, which is a feature that watches a user and starts predicting what that user might do next—basically make a recommendation to saying, “Isn’t that what you had in mind of doing next?”
And that, of course, makes it much faster for the user to go through a certain work process. And maybe as for an experienced user, it’s just faster. And for an inexperienced user, it may save a lot of time, where that user isn’t really sure what the next step should be and starts digging through Help menus and the menus on the screen and so on, so doing all this unproductive stuff. So I think in short, there’s a lot of heavy-lifting work that AI is taking over.
Another example is what we use in our semiconductor design side is the semiconductor designers have a lot of boundary conditions and variation of their designs. They have to keep it in mind then when they make a change. So AI is already helping them manage variations and just supporting the engineer here.
Or another example is, when you develop software, you get these bug reports and you get hundreds of them and you have to read them all and manually figure out which component of your software is responsible for that bug. So that’s another function that is now being automated by AI because it’s another piece of work that’s really a lot of tedious, detailed work. So, I think AI is playing a bigger role to allow engineers to focus more on the real, creative part of their job and less on detail work.
Laurel: Yeah. And that’s a bonus and a benefit for everyone, right? More creativity and less tedious work.
Laurel: So when we bring this up a level and we think about sharing data and connecting systems within a modern organization, how does this idea of sharing data and sharing scenarios and simulations and experiences help the organization actually start that evolution?
Stefan: Yeah, I think the simple answer is as everything is becoming digital, so every organization is more dependent on data than it probably was 20 years ago. So we live off data. And as we just started talking about, if you want to take any use out of AI, you need lots of data. You need so much data that ultimately your AI can extract something meaningful out of it.
And the problem is, of course, that historically as every business has become more digital, we have created these islands of data basically because we solved one problem first. So we created product lifecycle management, which is the place where you hold the data for design, but then we have also the ERP system, enterprise resource management, which is like SAP, which holds all the business data. That’s a different data repository.
And if you really look closely into complex manufacturing companies, they have dozens and dozens of data repositories and they are all disconnected. And that’s a challenge.
It’s the next level of what has to happen is that we’ll start bringing together these very disparate, these islands of information, and we start connecting them because ultimately when you hold a product in your hand, all of these data from different sources are in there. So, after we have figured how to put anything we can into an electronic database, the next step is going to be to bring those data sources together.
Laurel: So, in your experience, why is this valuable? Have you seen anything particularly exciting come out of disparate databases brought together for business decisions or just something surprising that helped a client or a customer do something interesting?
Stefan: Yeah. You put it very positively. I think I have a lot of negative examples where a seemingly small change in one of the islands of information has a huge impact, but there’s no chance to see it without knowing the other data. In the automotive industry, like the mechanical design and the electrical design, it basically was born independently, and it’s only right now, automakers are figuring out better and better how to bring these worlds together because they have to.
Just as an example, if you develop the electrical system of a vehicle, you might think that at some point, “I need an extra wire here—I can’t solve it differently.” So I add a wire to my wire harness and just make it a little thicker. So it may look like a fairly modest change where you are sitting, you’re just saying, “We’re changing from a diameter of 0.8 inches to 0.82 inches. That can’t be so dramatic.”
But your mechanical colleague has probably figured out where to put this wire harness in the vehicle, and he might’ve already ordered the tooling to do metal bending and really to build a cable channel that will exactly fit 0.8 inches, but not 0.82. This kind of problem still occurs in that industry.
And the background is really, a lot of the products that we use today, automobiles, but also electronics, cell phones, and so on, they are very highly optimized. If you open the hood of a 20-year-old car, you see a lot of space in there where you could put stuff. If you pop the hood of a modern car, there’s almost no space, there’s no wiggle room.
And because of that lack of wiggle room, it’s really more critical today than ever to understand if I change a little thing in my world, what happens to somebody else’s world? And I think this is where you see I have lots of examples of what can go wrong if you don’t take this into account, but there are, of course, certainly a big potential also of avoiding these mistakes.
Laurel: Yeah, lots of opportunities there eventually, but that challenge is bringing all that data together. So, when we think about this, obviously new, but necessary attention to data, machine learning, and AI, how will it help spur on competition and accelerate a company’s product offerings?
Stefan: Yeah. As I said, I think, as consumers, as most of us who could buy technical products, we more and more, I think, get excited by these very smart types of functionality. And you probably agree, I mean, the day that a really reliable and affordable self-driving car hits the market, we will be very interested. And I think we are already interested if somebody tells us, “OK, this car can actually parallel park without you touching anything.”
That is super exciting. So I think in general, AI-driven functionality will probably have a big part in differentiating what a business can offer, probably be a little, even as exciting or more exciting than the looks of a product or the aesthetics or other parameters like this.
I think also AI and electronics in general is coming into more and more types of products that haven’t been so heavy in electronics before like, imagine running shoes or sports equipment are getting smarter year by year and more and more things have chips in them. So I think overall, it’s becoming more and more of a differentiator and a way to attract people and also build these ecosystems of intelligent applications that get people hooked.
Laurel: Yeah, it’s excellent for the consumer. You can see that. For the person who builds it and the engineer in the production of it, how will AI help keep that human in the loop? How will AI help a person’s job? We talked a little bit about improving creativity, what else helps?
Stefan: Yeah. As I said, my skepticism of humans really being replaced is not so high than it might be for some other people. And I think as we have started talking about the most AI applications that are now basically going into supporting the workplace actively, again, they’re mostly focused on making the human more productive and being an assistant and taking care of detail work of heavy lifting that humans aren’t as good at and they are also not as interested in.
And, of course, you can always make the argument, “Yeah, if I make humans 10 times more productive, doesn’t that mean I can let go nine of 10 workers if I achieve this?” So theoretically, that’s true. I would frankly say, that’s really not what we have been seeing in the past. For some miraculous reason, before covid-19 started, unemployment in the US was continuously going down for, I would say, ever since 2008. So whatever productivity was achieved in that time, did not really lead to job losses.
And if you look at technical professions, I think there’s still a shortage of engineers, which you would think, “OK, if I make engineers more productive, shouldn’t I lose a lot of engineering jobs?” That’s really not what we have been seeing.
And I think one of the explanations is, number one, the more productive we get, the more sophisticated products become and the more there’s at some point consumption growth.
And secondly, you always need experts to deal with the latest technology you come up with. I mean, before we had cars, we didn’t need car repair shops. Now, we do. So it creates a new profession. And I think with AI, you will probably create professions that will be about really making it work, making the application stable. And so to me, it’s very hard to predict if it’s really going to hurt things like job markets. I would say, that’s really not what we have been seeing.
Laurel: Great. So when we think about AI, benefits of it, how can AI be an invisible aid to these people building, designing, producing?
Stefan: I would say, for the most part, it probably already is, because again, I think we had a few examples. Many times you don’t really see where it is in action or not. Of course, if your computer makes a recommendation, what you might do, you will think, “OK, some intelligent instance is there helping me.” But in many other areas, you might not even realize it.
There are of course, some very exciting spaces, where we make it very visible or we get to see it very visible because we are actually getting better and better at completely automating the creative work, for example, of creating structures that are biologically inspired. So as you know, today, there’s a technology in manufacturing called 3D printing that is very flexible in what the shape looks like that you build. And there are technologies that can really take the boundary conditions of a design and then let the computer figure out what the right geometry is. And what comes out of that usually looks like a vegetable or a plant. So, very funny structures get out of this.
So in that case, I have made this intelligence very visible, and I have really taken the whole job of designing out of the hands of the human being, and then the machine comes back with something that you almost think it could grow in your garden.
So those are the very visible things that I think are extremely exciting, because again, if a machine can do it and can do it even in a more elegant way, why should the human then be bothered by it and not think about the more creative aspects?
But on the other hand, I think there’s a lot of functionality already that we really don’t get to realize that AI is supporting it. And I think over time, I almost think 10 years from now, we might even not have so much discussion about the value or the future of AI or how it will evolve because we will be so used to it. I mean, how many discussions do you have nowadays about the value of Excel, of cellular calculation although we use it every day? Everybody uses it every day in something, and it’s so universal that we hardly ever think about it.
So, to me, there are two possibilities 10 years from now: either we are so used to it that we barely ever talk about it, or we hit another wall, like that was hit in the late ’90s, where you figured out there’s so much it can do, but we don’t have strong enough computers to have it do more. So we forget a little bit about it. And then 20 years later, we have yet faster computers and we, again, get all excited.
Laurel: That’s amazing, I love the idea of somehow acquainting AI to be as commonplace as a spreadsheet. You’ll just use it and you won’t even know it, but your life will be better because you have it. Stefan, thank you so much for joining us today in what has been an intriguing on Business Lab.
Stefan: Absolutely. Thanks for having me.
Laurel: That was Stefan Jockusch, a vice president of strategy for Siemens Digital Industries Software, who I spoke with from Cambridge, Massachusetts, the home of MIT and MIT Technology Review, overlooking the Charles River.
That’s it for this episode of Business Lab. I’m your host, Laurel Ruma. I’m the director of Insights, the custom publishing division of MIT Technology Review. We were founded in 1899 at the Massachusetts Institute of Technology. And you can find us in print, on the web, and at dozens of events each year around the world and online.
For more information about us and the show, please check out our website at technologyreview.com. This show is available wherever you get your podcasts. If you enjoyed this episode, we hope you’ll take a moment to rate and review us. Business Lab is a production of MIT Technology Review. This episode was produced by Collective Next. Thanks for listening.