Digital twins, linked with AI, could remake networking. The combination enables a whole new level of automation.
A recent report by Hexagon looked at data from 660 enterprise executives worldwide to identify the ROI that could be achieved by digital twin technology. It found that only half of companies reported an ROI above 20%, which is roughly where CFOs tell me a project can get quick approval. Interestingly, this is almost exactly the ROI enterprises tell me they can expect from the AI projects that are approved. Why is that? I think it’s because digital twin technology and AI technology have a critical and unrecognized symbiosis. Doing one without the other is like playing the field without a glove.
As I noted in my last blog, enterprises say that getting the most from AI requires introducing it into their current application workflows, rather than “copiloting” individual workers. This workflow-component strategy couples AI with the real transaction flows that represent business operations. Workers rarely even see those flows, except in the form of reports or action instructions created by applications that reflect their results. For AI, context matters.
For digital twins, context is really everything. A real-world system is a whirling mess of elements, some mechanical and some human, and it all has to fit together to work. To exercise any control over such a system requires the software have a complete understanding at the system level, which is where digital twins come in. But enterprises tell me that creating a digital twin, and making decisions about how it’s used to control real-world systems, is something that ordinary application development often can’t handle. This tends to limit the use of digital twins to simple systems, and simple systems may have few or no direct interactions with workers. Since it’s worker productivity that creates application ROI, a lot of solid business cases can’t be made.
So how about combining them? There are three ways that AI and digital twins can be linked, and each could magnify the common value proposition, make more and better business cases, and raise ROI.
1. World foundation models
One way is the notion of “world foundation models,” which can model a real-world system and be used both to create and train a digital twin. These models can take various forms, including those modeling real-world systems that are largely human. How do you get an autonomous vehicle to avoid other human-driven vehicles or pedestrians, even pets? It has to be trained, and rather than setting up an enormous bumper-car arena and putting lives and property at risk during training, feed the system a video of a scenario and let it respond to that.
2. Closing the control loop
A second strategy is to use AI to do what industrial process types call “closing the control loop.” A real-world system is connected to a digital twin via some sort of sensors, and these sensors synchronize the model with real-world conditions. To exercise control/influence, an application component needs to assess the state of the model (and through it, the state of the real-world system), and then decide what steps to take. Where the system is simple, and in particular when a lot of naturally disorderly human elements aren’t involved, this can be done with traditional programming techniques—edge computing. But complex systems have too many variables to handle that way, which is why it’s been difficult to apply digital twins there. AI could fix that; a machine learning process or even a form of generative AI could be used to close the control loop by making optimum decisions.
3. Link AI with transactional workflows
The third point of cooperation is the linkage of AI with transactional workflows. Companies already have applications that take orders, ship goods, move component parts around, and so forth. It’s these applications that currently drive the commercial side of a business, but taking an order or scheduling shipment doesn’t load a truck or label a package. Digital twins have historically been linked to the real world via sensors that detect actual movement, work. How do they capture the commercial applications, the transactions? An open data framework like OpenUSD is a possible answer, and if AI also supports OpenUSD, then there’s a mechanism that not only lets AI “read” workflow data from existing applications, but also generate work to be introduced into those flows.
All of this, in combination, makes AI and digital twins a partner in a business at both the transactional and real-world functional level. The new combination introduces a whole new kind of automation, automation of an entire business. Automation that can touch every process, every worker.
Remember that our current IT spending is justified almost entirely by enhancing the productivity of the 60% of workers who are involved with the commercial/transactional side of things? That’s left 40% of the workforce out of the productivity picture, and their value of labor is actually a bit higher than that of the 60% of workers we’ve already reached. What kind of IT spending could reaching these workers justify? Digital twins, combined with AI, could totally remake IT, and totally remake networking.
Why networking? The 40% of “functional” workers, the ones out there pushing boxes, driving, even fighting fires or protecting the population, will need their own data collections to be empowered successfully. We’ll need to know what’s going on in the real world at a level we don’t even approach today. If you believe that autonomous vehicles can navigate the streets, then you believe that we can build a digital twin of those streets and an AI controller to get the vehicle safely (and legally) to the destination. If you believe that, then you believe we can understand and optimize the movement of goods and the providing of services in the same way.
What we’ve learned with AI and with the IoT/digital-twin initiatives so far is that low apples don’t provide the best ROI, and trying to find opportunities for new technologies individually is less likely to be successful than planning to use them cooperatively in order to address the complexity that’s a regular part of our work, our lives. And the good news is that, behind the hype, companies are starting to offer the tools needed to harness all the good stuff that we’ve talked about, and businesses are starting to see how to adopt those tools. So don’t be distracted by all this AI hype, or by vague claims of autonomous operation. Reality may be closer, and better, than you think.
