Rolling out hypothetical 5G infrastructure, optimizing manufacturing workflows, testing experiential new products, even building out a complete – constantly evolving – doppelganger of the entire world. Those are just a few examples of what digital twins make possible.

Digital twins, or representations of tangible assets or systems in the physical domain, are transforming industries because they enable organizations to virtually “tinker” and experiment.

Now, AI — and particularly rapidly evolving generative AI platforms — present the opportunity to rapidly advance these capabilities.

“Companies can see patterns they can’t see using traditional data collection and analysis methods, identify exceptions faster and respond more accurately,” said Mike Kuniavsky, technology R&D senior principal at Accenture Labs. He added that, “the world of digital twins is growing by the day.”

What is a digital twin?

At its simplest definition, a digital twin is a virtual representation of an object or system that spans its lifecycle, is updated from real-time data and uses simulation, machine learning (ML) and reasoning to help inform decision making.

Gartner analyst Rajesh Kandaswamy describes digital twins as a “technology-enabled proxy that mirrors the state of a thing such as an asset, person, organization or process.”

As he explained, digital twins are built in software — analytics, 3D models, CRM, IoT platforms — and their elements include model, data, unique 1-to-1 association and monitorability.

These digital representations are closely coupled with their physical counterparts so that elements can be examined and manipulated in ways impossible in the physical world, Kuniavsky explained. They are unified so that whenever there is a change in the physical object, there’s a corresponding change in its twin.

This can enable in-depth, for example, quantitative analysis of complex business processes (the design, manufacturing, marketing or sales of a new product) or real-time immersive visualization of a single object (watching a maintenance robot repair an inaccessible machine), he pointed out.

“Digital twins benefit organizations by allowing them to use all the tools we have developed for analyzing real-time digital phenomena — databases, filters, visualizations, pattern classifiers —to our physical environment,” Kuniavsky said.

Bringing digital twins to life

From testing and refining, to optimizing workflows, to creating sandbox environments for partner experiments, Kandaswamy has seen many organizations in “asset”-based industries — that is, with tangible physical assets — diving into virtual environments. This includes manufacturing, retail, construction, transportation and telecommunications.

Ericsson, IBM, Verizon and Nvidia are all making use of digital twins. Nokia, for its part, is studying 5G use cases with AI and machine learning (ML) algorithms coupled with digital twin technology. These monitor and assess the impact of 5G implementation and provide automated recommendations for next steps.

The company has also coupled with Bosch to research 5G industrial automation and control systems in manufacturing. This involves sensors throughout production that collect data, transmit it over 5G link and display digital twins of the machines to enable real-time adjustment of operating parameters.

Network digital twins also help with capacity management and process optimization, improve customer experience, ensure the reliability and traceability of the supply chain and enable agile DevOps. Other examples include RAN optimization and planning for “what if” scenarios (potential gaps between coverage and demand, for instance).

“Digital twins represent a fundamental shift in the scope and scale of engineering,” Nokia asserts in a blog post. “But they also enable a transformation in the scale of the human imagination.”

How AI takes digital twin insights to the next level

So how can generative AI help take digital twins to the next level?

As Kuniavsky explained, a number of AI techniques can be used to analyze the functionality of a digital twin object or process in ways that would be too difficult or time-consuming using traditional methods.

“AI is great at recognizing patterns,” he said. “AI tools can distinguish between a photo of a muffin and one of a puppy by learning patterns from trillions of pixels and their relationships to each other, then quickly learn about a new kind of muffin or dog breed from just a few images.

This kind of analysis can be applied to business processes, he said, with AI sifting through billions of data points (say, every quarter turn of every wheel of a bullet train) to identify otherwise unrecognizable patterns (for example, “when a wheel wobbles in this way, that means the track in that specific spot will need resurfacing in 2 months”).

Modern AI can then take these pattern-matching systems and invert them to create brand new patterns from what they’ve learned. For instance, a clothing company could create a “clothing possibility space” that produces designs that fit the season’s (or the week’s, or the day’s) style motifs.

More pragmatically, generative AI could rapidly accelerate traditional design tasks. As an example, Kuniavsky said, it could possibly take traffic flow data from different subway stations at various times of the day and combine that with the specific physical layouts of the stations (down to where every trash can is located). This could be used to create a GenAI that could instantly create new station layouts based on varying different criteria. Experts could then use this to compare how different factors could affect traffic flow.

While this kind of design has been possible with non-generative AI systems — known as “generative design” for the past 20 years — the methods to create models were labor-intensive and required significant specialization, Kuniavsky said.

As he put it, GenAI can now tell us: “If you have enough examples to analyze a phenomenon, then you probably have enough examples to create synthetic examples of the same phenomenon.”

“Generative AI allows us to go from examining a thing to dreaming up new versions of that thing, in a single step,” said Kuniavsky.

4 questions to ask before you venture into digital realms

With their myriad benefits, the use of digital twins is bound to accelerate: Appledore Research Group, for one, projects the digital twin market to reach $10 billion by 2025.

Still, enterprises must be cautious and thoughtful in approaching the technology, Kuniavsky advised.

Before adopting digital twins, he advised organizations ask themselves these questions:

1. What data do they already collect that could become the basis for a digital twin? Do they collect data in near real-time? Is there existing data that the organization can quickly convert to near real-time?
2. If the organization had a real-time digital twin and a way to identify patterns within it, what decisions could it make that it can’t today? Could it respond faster to certain positive patterns that it currently doesn’t know about until it’s too late? Could it avoid certain situations by being able to predict the outcome of a negative situation?
3. Would a generative AI meta-product description allow designers and customers to imagine new products they hadn’t considered before?
4. With digital twins enabling simulation and generative AI providing what-if scenario development, how can these two tools combined improve an organization’s decision-making and operations?

Kandaswamy agreed that while use cases are evolving rapidly, the technology is still relatively nascent — as is its pairing with AI tools.

“These are early days in this,” he said. “Firms must be careful of the hype. It is important to understand the purpose of doing a digital twin and evaluate multiple options, including generative AI.”