The Ethical And Sustainable Aspects Of Industrial AI Innovation At Siemens: Dinis Guarda Interviews Michael May, Head Of Company Core Technology Data Analytics & AI, At AI With Purpose Summit 2024

Dinis Gaurda interviews Michael May, Head of Company, Core Technology Data Analytics & AI-Siemens at AI with Purpose Summit 2024. In this episode, Dinis and Michael discuss the transformative role of AI in shaping industries, Siemens' innovative approach to integrating AI across its global operations, and ethical considerations. The podcast is powered by Businessabc.net and citiesabc.com. 

Dr. Michael May is a technologist and researcher, especially in the areas of AI, data analytics, machine learning and big data architectures.  He is heading the Company Core Technology Data Analytics & AI at Siemens. In Siemens Corporate Technology, Munich, he is responsible for fourteen research groups in Europe, US, and Asia. 

At AI With Purpose Summit 2024, Germany, he is on panel discussion on Challenges and Opportunities of AI Governance. Speaking about AI governance and regulation, Michael told Dinis:

“AI regulation is a topic you must deal with, if you want or not. You will not be able to, in a short period from now, deliver products to customers which do not somehow give an answer as to how do I make it safe, how do I make it responsible, how do I meet the privacy requirements attached to that. You need some level of governance that you have to address. You need to have answers why your system is secure. 

Siemens, for a couple of years now, is working on that - preparing for regulation, but also trying to seize the opportunities that give you to build systems which are industry grade, i.e. comply with the upcoming regulations.

If your customers know what you’re building is safe, and it does comply with all the regulations, then they might be inclined to buy your products or use your products with more confidence.”

Speaking about striking a balance between regulation and too much of it, he further emphasises:

"On the other side, of course, we have to see that we do not overdo it. If there's too much of it, then it can become a burden for innovation. It can make the production process much more lengthy, more expensive.

We need to be absolutely clear about whether we need to have regulation which is good for innovation still and on the other side you keep things trustworthy. Build things or aim to build things which are safe and trustworthy even without external regulation.” 

Sustainable AI: Simulation and other AI-powered industrial solutions by Siemens

Siemens is harnessing the power of generative artificial intelligence (AI) to help industrial companies drive innovation and efficiency across the design, engineering, manufacturing and operational life cycle of products. Michael discusses prominent products like simulation software enhanced by AI. He highlights the application of this technology for designing and optimising the processes in automotive and oil & gas industries.

“Siemens is selling simulation software, for example, for the automotive industry, or for the oil and gas industry. We use simulation to build better products which are more energy efficient. Before you build it, you do a lot of simulation to make sure that what you build is in fact what you wanted it to be. These simulation systems are costly in terms of computation time, so you need a lot of computation - many hours, many days, many weeks, and in some cases until you have the results, because there's very complex mathematics behind what is used for the simulation. 

Now the question is, wouldn't it be, maybe, better if we can speed up the whole process, for example, by making it faster, so that it consumes less energy, because it doesn't compute that much. Maybe, it even finds solutions that could not be found before, and so, in the end, you get a product, maybe a more environmentally friendly product.

This is what we have been doing at Siemens. We are now selling a product that uses AI to predict using simulation. Using these techniques, you can filter away many of the potential solutions that have to be calculated, and then you focus just on the promising one. Maybe this is how we humans are good at thinking, so we don't evaluate millions of possibilities, but just a few because we know where to look, where to focus on the right part, and there the AI helps us to focus on the right part.”

Michael also highlighted the recently unveiled integrated generative AI-powered assistant in a production machine: Siemens Industrial Copilot helps automation engineers speed up the generation of code for programmable logic controllers (PLCs).

The evolution of Generative AI solutions at Siemens

During the interview, Michael outlines efforts to develop "smaller" AI models tailored for specific industrial tasks, significantly reducing energy consumption while maintaining efficiency. He told Dinis:

“An active area of research is to make these models smaller. Some people call it a small large language model or small language model and this is a very important area of research that is coming up with the same level of sophistication or quality of the model with being much much smaller in the number of parameters. This directly translates into energy consumption. Over the last 18 months, we have seen a lot of progress with some of the open-source models which have good results while being drastically smaller.”

He also highlighted that these fined tuned models also reduce the chances of hallucinations by AI:

“AI is not just generative AI. It is not just machine learning or deep learning but has many facets.  There's the one which is more, what is called inductive or probabilistic, based on inference, and  which might sometimes be wrong. And then the more logic based part that can give you these guarantees and one very practical way we do that is to bring in the domain knowledge.

Industrial companies have a lot which we structure in the form of what is called knowledge graphs. A Knowledge Graph is a formalism where you can model terminology knowledge, ontologies for example, and where you can also structure fact-based knowledge in a way that if your representation is correct then all the inferences you make from that are also correct because it's based on logic and not so much on probabilistic.”

Looking ahead, Dr. Michael envisions a rapid evolution in AI capabilities, particularly in multimodal models integrating diverse sensory inputs:

“The development will go on quite quickly and you can well extrapolate from now. I mean now we have these large language models, we have Vision models, we have first forms of multimodality, and I think step after step we will cover all the other modalities which are relevant like thinking in Time series, thinking in terms of space and geometries. So, it will be even more like our human brains. 

We will also have all these modalities we can talk, we can hear, we can see, can think about space and time, step by step will be incorporated into these multimodal foundation models and that will be a big big step forward in terms of capability. Not just adding one by one and one, but it's exponentially more powerful if you have all the combinations of reasoning and seeing and speaking. 

I'm always a friend of seeing AI as an intelligence amplifier, not as a replacement for a person. So, I think for the time to come, the most successful teams will be those that consist of humans and machines with different functionality, with different tasks and we need to learn. There's a lot to learn for us how to take best advantage of the machines and I better do it myself. 

AI, I think, will determine the future in the next few years as to how the industrial world will be working. I think it's not about replacing humans. We see how difficult this is to do the last bit so the last percentage or milli percentage of automation for all our processes”, said Michael.

Concluding the interview, Michael shares his inspiration for AI and technology with Dinis:

“For me with all this AI, it's still the question with which I started. I still wonder how learning works, human learning, and learning in machines and that has somehow kept me curious all the time all the years. Maybe you should look for that so you have a motivating question and then see how through all the years all the iterations and evolution steps of uh technology you can get that answered and I'm still wondering what the final answer would be.”