“All models are wrong, some are useful” a famous quote by George Box. However, people's presumptions about digital twins are that these twins will lift us to a much higher level. The emerging interest in digital twins can be seen in many different fields. Especially with the increasing capabilities of AI, the expectations are sky-high. This was the opening of Susan M. Sanchez, one of the panel speakers at the Winter Simulation Conference (WSC’23, San Antonio, Texas) on digital twins and artificial intelligence.  Sanchez, an experienced senior researcher in simulation for many years, humorously presented Google hit statistics to highlight the significantly greater interest in the recently introduced term "digital twin" compared to some well-established simulation terms that have been with us for decades.

So, what makes this digital twin an actual twin? And how far are we ahead? 

The panel discussion (for the panel statements, see Taylor et al. (2023)) centered on applications in manufacturing and production. For labeling these different levels of smart, we borrow some terminology from an older paper (Fuller et al., 2020) and dive into digital models, shadows, and twins. Did you know that a digital model doesn’t need to be an exact replica of reality? In fact, sometimes, quite the opposite! Digital models are mainly useful for strategic decisions, as Marcus Rate opened his pitch. You could calculate what the effect is of a change in recipe or investment in new equipment.  The digital model’s main goal is to answer, “what if”, rather than “what should I do” questions.

“So, what about tomorrow?” Marcus continued. How can we answer these ‘what should we do’ questions that factory operators are facing when for example a machine breaks down? We stick to the terminology from Fuller et al. (2020) and move towards a digital shadow. Importantly, this raises different requirements, now we do need a digital copy of the real system. Importantly, the digital shadow captures the current state of the system, allowing us to simulate the near future before making manual decisions in the real system.

Exploring the various levels of smart, the question arises: what remains in the realm of digital twinning? Where the digital shadow only requires the uni-directional data connection, the twin needs the connection to go in both directions. Achieving identical behavior between the model and the real system demands substantial adjustments in current manufacturing systems. As per the insights from the panel speakers, such synchronization is not yet evident in the real world. However, this seems quite likely in the (near?) future. Take only as an example how chatbots amazed us this year. How unlikely this true digital twin situation may sound now, the fast development of generative AI has demonstrated how quickly technology can surpass our expectations.

So, what's the takeaway for us as WSC attendees? Charles Macal was explicit in his panel talk, offering distinct advice for researchers, engineers, and simulation vendors. For researchers, he had great news, there are a lot of research opportunities ahead! Developing digital twins comes with many problems, but problems are great for research. As food for thought, he suggested considering a brief pause in current research to dive into deep generative AI and explore it’s possible application to your specific research problems.

Concluding this piece, I'd like to share two quotes. Macal advised, "Don't focus solely on the situation of today and what AI cannot do for your problems now; instead, envision where we will be in the coming years and ensure you can keep up the pace." Lastly, Sanchez's metaphor, accompanied by an image of a surfer breaking the world record for the highest wave, being: "We simulation enthusiasts are the surfer on this image, and the wave is AI; we better ride it."