玩偶姐姐

While an engineer for Digital Equipment Corporation in the early 1990s, co-invented the first digital twin of a CPU.

“When you build computer chips, there are billions of transistors,” she says. “You can’t look at every one with a probe and a wire. Wouldn’t it be great if we could have a digital twin model and artificially break it under different conditions, and use that to find manufacturing defects?”

Panetta, now dean of graduate education for the School of Engineering at Tufts University, is also an innovator in the world of AI with her pioneering work in computer vision. And today, digital twins and AI are being combined in a way that amplifies the potential of each while creating new opportunities to do things previously out of reach for businesses. For example, digital twins are now being used in conjunction with AI to model diseases and human organs, monitor major infrastructure projects, analyze IT infrastructure, and much more.

“Cybersecurity is going to be the first market that utilizes this in a very robust way,” she says. Companies create digital twins of their systems and use them to collect the data that comes in. “Now you can plug in different tools or vendors, and experiment with how people can break in.”

There’s a lot of potential in combining digital twins and AI, but many practical use cases already exist in enterprises today. Part of the value of digital twins is they’re a platform to pull in data from multiple systems and combine it in a way that makes sense. But the reality is things are still siloed, says Gartner analyst . “Despite everyone saying we have one view of the customer or patient, it’s still being pulled in from proprietary silos,” he says. “And that’s one thing the digital twin does well.”

Digital twins also incorporate domain knowledge and align with specific business needs, he says. Combining digital twins with AI takes us into the next evolution of the technology.

“At Gartner, we now call it intelligent simulation,” Velosa adds. “We expect it to be one of the top 12 disruptors.”

In April, Gartner predicted that intelligent simulation will underpin more than 25% of strategic business decisions by 2032, and by 2027, the first enterprise will be able to quantify $1 billion in operational savings by using simulation twin capabilities.

Furthermore, according to a Hexagon survey of over 600 executives released late last year, 80% of leaders — from automotive, architecture, and engineering to construction, oil and gas, and city planning — say AI has made them more interested in digital twins.

How AI super-charges digital twins

McKinsey says digital twins can be difficult to design and build, but LLMs can help write code for them, accelerating the development process. Gen AI models can also compress data while retaining critical information, allowing digital twins to better manage the volumes of data they require. Plus, they can be used to generate scenarios for the digital twin to simulate, or act as a natural language interface for the twin, making it easier for people to use.

Hexagon adds that, among other things, 59% of leaders say AI is used in processing front-end data, 56% to enhance the user interface, and 27% to make decisions.

Take for example the user interface of the digital twin. Although the twin is easier to look at than the raw data feeds themselves, that doesn’t necessarily mean anyone can immediately see what’s happening.

“Generative AI would be used to look at the entire simulation and turn it into a summary for humans,” says , professor at Carnegie Mellon University. “It could tell me things I might be missing, and summarize things in a way I can understand.”

AI could also allow a user to interact with it using natural language. A recent use case from a telecom network shows a digital twin simulating an outage so it can see how traffic is rerouting, using an LLM as an interface so it’s easier for an executive to use the digital twin.

“Before you had data scientists building simulations,” says , emerging technology leader at PwC. “Now you can have a lightweight interface to translate the human language into the levers the system needs. And because the simulations have reasoning built in and can plan, they can rewrite the simulation on their own to give you more permutations.”

One organization using AI-enabled digital twins is the 4J School District in Eugene Oregon where Ben Shapiro, senior network engineer, supports more than 16,000 students and over 50,000 devices. A digital twin makes it easy for even level-one tech to see what’s happening on the network. The district uses the Marvis Minis network digital twin from Juniper Networks, in conjunction with Mist AI. And a gen AI chatbot within the platform makes it even easier. “Marvis is a game changer for democratizing tech support,” he says. “My goal is for the low-level, front-line tech support to be able to figure out what’s going on without elevating it to me.”

Mavis Minis doesn’t yet support running simulations, though, but that might come in the future. “It would be really cool to leverage the data and telemetry that Mist has collected to make sure your network is future-ready,” he says. “It’d be nice not to just guess, but have something to back it up, and be able to quantify it for leadership.” For now, he says, he just relies on his intuition.

AI-powered digital twins of humans

The emergence of highly-intelligent gen AI models has opened up new use cases for digital twins that weren’t possible before. For example, companies have long been using ML to segment customers and predict what different kinds of customers might want. With gen AI, a company can build a literal avatar of an individual customer.

“The marketing or product team can now interview this twin and ask, ‘Would you buy this new book series or this cruise package?'” says Gartner’s Velosa. “Or, ‘Which would you prefer between these six options?’ And they’re talking to the digital avatar.”

And why stop at a single avatar? Digital twins can be used to simulate entire focus groups. It’s kind of like trying out a trial strategy in front of a mock jury, says Simon James, managing director for data strategy and AI at Publicis Sapient, a digital consultancy. If a company is testing out potential advertisements, it could run them past a virtual focus group, with digital twins of different representative customers from different demographic groups. Publicis Sapient, for example, might get a request for information from a potential client.

“I might have a panel of different 玩偶姐姐s,” says James. “One might be really cost-conscious, and another into the weeds who understands the technology. You get them to read your document and come up with questions.”

He says it’s too early to quantify whether the firm has been able to win more work as a result of the process.

“We’re looking for every advantage,” he says. “I think it gives you a view of some perspectives that maybe you wouldn’t have if, say, you were working in isolation or on a small team. Sometimes getting an external perspective is useful.”

AI can also be used to model individual employees, says Carnegie Mellon’s Lightman, such as workers in a factory.

“In manufacturing, you’re modeling and analyzing human activity,” he says. “You have maintenance workers coming in, operational workers.” But there are risks associated with this particular use, however. For example, do the workers have any recourse? Lightman asks. Can they say no to being simulated? Maybe they don’t want their activities to be modeled.

“There might be hesitancy associated with some actions I’m trying to digitize,” he says. “I think organizations have to be much more transparent about why they’re collecting data and what they’re doing with the data.”

Digital twins of workflows, or entire companies

In order to integrate AI into business workflows, an enterprise has to have a good understanding of what those workflows actually are.

“We don’t have that understanding of how the enterprise operates,” says Jon Knisley, ABBYY’s product marketing manager for process AI. “And that’s where I think the process digital twin can become incredibly valuable.”

AI could soon be used to figure out what people actually do in a company, something that’s been difficult to do in the past. Process mining was supposed to help, but actual business processes are often inconsistent, undocumented, or just wrong.

“Most of our work is around process digital twins — looking at log data and user activity data, and converting it into a representation of the work that’s going on,” Knisley says. “We’ve got great security, marketing, and finance data but we’ve never had great process data.” And the lack of it is the last big gap preventing companies from truly becoming data driven, he says. “That’s the holy grail.”

With AI, a company can put intelligence to work on the task, and also combine transactional and log data with other sources of information. For example, one team might tackle a particular task differently from a team elsewhere in the same company due to different resources and constraints. Or consider a process where individual actors might have some discretion, says Publicis Sapient’s James.

Digital twins are commonly used to simulate supply chains or distribution centers, he says. “What AI brings is that rather than being a rule-based system, you can give some discretion to the actors in the chain to do different things,” he says. “You can put sensing and thinking nodes into your system.” A store manager in a college town, for instance, might see 40% of foot traffic on Saturday morning. “So I don’t want to be out of stock,” he says.  And the system as a whole can be more intelligent.

“With the AI-driven approach, you can let the framework figure out the best strategy,” James says. “That’s been an emerging quality of deep learning networks for a while. You don’t explicitly teach it how to play chess, but it can figure out what the ultimate strategy is to win the game on its own.”

How AI benefits from digital twins

And it’s not just AI making digital twins better. The digital twins can also make for better AI.

“We’re using digital twins to actually generate information for large language models,” says PwC’s Likens, adding that the synthetic data is of better quality when it comes from a digital twin. “We see opportunity to have the digital twins generate the missing pieces of data we need, and it’s more in line with the environment because it’s based on actual data.”

A digital twin is a working model of a system, says Gareth Smith, GM of software test automation at Keysight Technologies, an electronics company. “It’ll respond in a way that mimics the expected response of the physical system.”

This means the synthetic data the digital twin can generate for training AI models can be more extensive and richer than traditional synthetic data techniques.

Another potential use case for digital twins that might become more relevant this year is to help with understanding and scaling agentic AI systems. Agentic AI allows companies to automate complex business processes, such as solving customer problems, creating proposals, or designing, building, and testing software. The agentic AI system can be composed of multiple data sources, tools, and AI agents, all interacting in non-deterministic ways. That can be extremely powerful, but extremely dangerous. So a digital twin can monitor the behavior of an agentic system to ensure it doesn’t go off the rails, and test and simulate how the system will react to novel situations.

Today, 75% of large enterprises actively invest in digital twins to scale AI solutions, McKinsey said in an April report. “In the next decade, successful Fortune 1000s will run their businesses and test their boldest strategies using digital twins,” says McKinsey partner Alex Cosmas.

These digital twins will be replicas of their operations and simulations of their full value chains. As Cosmas says, they’ll be living, breathing, expanding assets that solve a growing number of problems.

“Leaders will test myriad decisions and a range of possible scenarios in simulated environments, without the risks of disrupting their core business,” he adds. Best of all, they should be self-funding and ROI-positive right from the start.

More on digital twins:

• Digital twins: 5 success stories
• 5 best practices for digital twin implementation
• Siemens grows its digital twin strategy into life sciences market