The AI revolution will be ‘virtualized’

This article’s insights were developed based on two papers by Dr. Will Roper: “There is no spoon: The new digital acquisition reality” (United States Air Force, October 7, 2020) and “Bending the spoon: Guidebook for digital engineering and e-series” (United States Air Force, January 19, 2021). The two papers discuss the increasingly blurred lines between digital and physical systems and the implications for public sector procurement. This article expands the perspective to encompass private sector companies as well.

Right now, we’re inside a computer program?
—Neo (The Matrix, 1999)

The 1999 science fiction film The Matrix blurred the lines between the real and the virtual. Less than a quarter of a century later, dramatic advances in AI and digital simulation are disrupting how technology itself is created and used.

Today, companies can and do achieve The Matrix–like simulation realism.¹Will Roper, “There is no spoon: The new digital acquisition reality,” United States Air Force, October 7, 2020; and Will Roper, “Bending the spoon: Guidebook for digital engineering and e-series,” United States Air Force, January 19, 2021. It’s now possible to create digital representations of components, products, and processes that precisely replicate the behavior of their real-world counterparts. This “digital thread” gives engineering, manufacturing, testing, and operations teams the perfect playground in which to solve problems, explore the limits of their designs, and create innovative solutions. And, freed from the shackles of the physical world, they can do all that at breathtaking speed.

An illustrative example is Team New Zealand’s pioneering use of digital techniques in sailboat racing, which helped it win two consecutive America’s Cup competitions in 2017 and 2021. Before the team launched its 75-foot AC75 in 2021, it knew how the monohull would perform and exactly what it needed to do to bring home the title. Using an AI-powered design simulator, the team “e-raced” many lifetimes’ worth of virtual America’s Cups on thousands of virtual prototypes, learning to glide at 50-knot speeds on AI-perfected hydrofoils despite complex forces and knife-edge sailing conditions.

Such physical mastery was possible only because the sailing simulator was functionally equivalent to reality—a digital thread connecting all relevant aspects of sailing inside a computer. Its prerace digital boats live in the same sailing metaverse as its race-day diagnostics. The data are consistent, so the outcomes are consistent, but they are much faster to create digitally than physically. As one of the designers noted, “Configuring the foils, which used to take up to several days, now takes hours.”

In addition to speeding up the learning curve, Team New Zealand’s example reveals another digital-thread epiphany: “It’s not one big thing that makes a difference. It’s a lot of little things put together that enable big performance improvements,” as one designer reflected. The team’s AI systems could navigate a vast landscape of possible configurations of all those little things, zeroing in on an optimum best set of parameters for its design. It’s the same approach that’s helped computers conquer games like Go.

Authoritative ‘virtualization’

The game may be similar, but the playing field is wholly different. A Go board is a 19 by 19 grid of lines, easy to represent in a digital model. Real-world problems, such as racing yachts and factories, are much more complex. Creating an accurate digital thread to represent such systems can be easier said than done, and many companies have invested significant resources into “virtualization” without capturing value.

Value in digital transformation begins with what can be called “authoritative virtualizations” that can replace or truncate real-world activities. In this context, “authoritative” doesn’t have to mean “perfect.” Working with physical systems already involves working with less-than-perfect data. Failure modes and likelihoods are estimated to fall within an accepted measurement and sampling error. Engineering teams shift their performance and failure estimates conservatively to account for the cumulative error in their measurements. When those estimates indicate acceptable real-world risk, systems can be certified for use.

The same thinking applies to the virtual world. By working backward from certification requirements, companies can determine the level of model accuracy that may authoritatively substitute for physically collected data.

Authoritative virtualizations are digital models of systems that are certifiably predictive. They can take many different forms, from look-up tables and simple statistical analyses to complex multiphysics models that take years to perfect. Whatever the source of the data, the key to replacing or truncating real-world activities with digital ones lies in managing acceptable cumulative error.

Such virtualizations are being used in many industries. From agriculture to motorsports and aircraft maintenance to e-commerce, these digital oracles have threaded into real-world operations, generating value by accelerating the emergence of previously unimaginable businesses. For these companies, the line between the digital and the physical has dissolved completely. Let’s look at some examples from digital pioneers.

Boosting throughput at Tata Steel

Superheating steel alloys to precise casting temperatures is a delicate metallurgic process. Violate heating conditions, even for a moment, and the batch must go through time- and energy-consuming reheating. It’s a throughput killer across the industry.

For Tata Steel’s Kalinganagar plant managers, the task was to raise productivity, with digital transformation offering new possibilities to challenge current operating practices. The company started with data, which historically had been used mainly to monitor operations. Believing that new forms of analysis could reveal far more, Tata hired data scientists to design predictive algorithms—in effect, a form of authoritative virtualization. The idea was that the algorithms’ output would yield more precise control, enabling employees to raise blast-furnace-productivity- and -casting rates.

It worked—in principle. But in practice, many staff had little digital expertise. Without skills in data science, data engineering, and analytics, they had no way to improve the algorithms as production conditions changed.

The solution was company-wide training. Leaders established an introductory analytics academy to raise plantwide digital literacy. It trains employees to understand, modify, and even develop their own analytics models.

On the shop floor, virtual production cycles were soon outperforming their traditional counterparts. Applying the analytics-driven approach took first-time success rates to over 90 percent. Combined with more than 20 other analytics-based projects across the plant, Tata Steel generated millions of dollars in yearly value. It also earned recognition as a member of the World Economic Forum’s Global Lighthouse Network.

Danone’s digital dairies

Tata Steel’s Kalinganagar work encompassed 11 major optimizations to a single plant. It illustrates digital transformation’s compounding effect as it empowers people to identify and optimize myriad details.

The potential is that much greater across an entire network of plants, as at dairy company Danone. But so is the challenge. Danone operates 40 dairy plants across Europe in different eras and with differing layouts. Production complexities were hindering growth. Upgrades to the sprawling network’s physical infrastructure were unlikely to be cost-effective. Danone therefore turned to digital transformation to rein in the complexity and cost, starting with a plant in Opole, Poland.

Again, step one was using data. Years of dairy operations provided the raw material to train AI algorithms that could optimize production equipment performance. Likewise, step two was plantwide employee training. But Danone took a third step: investing in digital connections for its machines and new enterprise software for its workers.

The results were double-digit performance gains and a 40 percent reduction in energy consumption. With digital tools connecting workers, changeover times fell dramatically, and machine data connected to enterprise software boosted labor productivity by 50 percent.

Altogether, the digital transformation reduced costs by almost 20 percent, raised efficiency by more than 10 percent, and improved quality and carbon emissions by nearly 50 percent. Danone is now rolling out the same broad approach to its other 39 sites, with the Opole plant earning a place in the Global Lighthouse Network.

Harvesting data at John Deere

Starting more than a decade ago, farm equipment manufacturer John Deere saw the potential impact of real-time data for farming efficiency. Since then, the company’s products have evolved from pure hardware into hardware and software platforms, forming an “Internet of Farming Things.” More than 130,000 interconnected farming systems collect more than 15 million measurements every second, all uploaded to a cloud platform.

Farmers can now monitor real-time performance, weather, and cost. They can subscribe to predictive algorithms that enable pinpoint management of planting, watering, and harvesting—a farming digital twin. AI-based advances enable equipment to detect weeds and precisely deploy herbicides, reducing waste by up to 80 percent while doubling yields. Another system monitors grain harvesting to autonomously adjust cut patterns, increase efficiency, and recycle organic waste as fertilizer.

Digital transformation when lives are at stake

While digitally transforming steel or agriculture involves some degree of risk, the unknowns are to a large degree manageable. In healthcare, lives are at stake. Yet powerful modeling is fueling breakthroughs in life-extending, quality-of-life-enhancing treatments for millions of people.

Consider the design of an artificial heart pump. Historically, optimizing blood flow without damaging cells was intractable computationally. Simulations involved millions of elements, with transient conditions that were difficult to model. Even comparatively well-understood inputs, such as blood pressure readings that change, made it difficult to predict performance, optimize battery consumption, and prevent blood damage. When new devices were finally tested, installation was challenging. And even then, device–body interactions changed over time.

Machine learning, data, and high-performance virtualizations have flipped this paradigm. One heart pump manufacturer’s digital thread reduced design optimization time by over a factor of 1,000 by using highly accelerated deep learning to replace simulation. This company can now execute 100,000 design variations in a single day.

Formula One World Championship racing breaks the speed barrier in design

The preceding examples show what can happen when factories, projects, or companies digitally transform. But what about an entire sport? Formula One World Championship motor racing, where cars whiz around tracks at speeds up to 400 kilometers per hour, has a 70-year tradition of adopting technology for competitive advantage. But rising R&D costs threatened to narrow a diverse competitive field. The subsequent introduction of cost caps, intended to constrain R&D, unleashed even more digital innovation.

To avoid the high costs of building physical prototypes and wind tunnels, digital twinning became an industry-wide means to refine all aspects of racing. Every car component and its associated physics—down to the literal rubber meeting the road—was painstakingly virtualized and anchored by authoritative test data. These digital models became so powerful that even computation is now capped, spawning even faster digital innovation.

Data-driven agility drives Formula One World Championship’s learning curve at the edge of digital possibility. Real-time feedback improves approximately 85 percent of each car during the racing season—one part every 15 minutes. Left undeveloped, “If you took the car that qualified fastest at the first race of the season. . . . by the time you got to the final race, that same car would be the slowest,” said McLaren Racing CEO Zak Brown.

The industry’s compounding learning is powered by 300 onboard sensors that stream 100 gigabytes each race to conjoin physical and digital cars in a continuously improving, looped digital thread. Teams create hundreds or thousands of track-optimized digital twins before each Grand Prix. By the time that physical cars hit the next track, simulator practice ensures that their drivers know every handling detail.

How to ride the digital wave

Formula One World Championship has achieved a near-complete virtualization of racing. Conjoined with—and learning from—the physical world, its dizzying pace of innovation is a harbinger of what’s to come.

From casting steel to the farm equipment made from it, the food produced by it, and the artificial heartbeats circulating its nourishment—whether sailing, driving, flying, or watching orbits overhead—few business, government, or military endeavors can escape this wave of digital transformation. And if the only real choice is to ride the wave, the following steps provide a broad road map for a successful digital transformation:

Inventory your organization’s data stockpile and IT

Any virtualization requires a threshold amount of sufficient-quality data to mirror reality. Likewise, threading together digital and physical realities requires enabling IT, often across the value chain. You either have them (and a clear path to acquire them) or are likely at a digital dead end. And the constantly changing technology landscape means continual reassessment of both data availability and IT system capabilities.

Establish a set of ‘digital lightning bolts’

Energizing efforts force head-in-the-clouds digital concepts to connect with real, measurable, feet-on-the-ground implementation. Properly selected, these digital lightning bolts become the engines that power the digital transformation and lead to a series of innovations.

The challenge is to steer clear of three common failure modes: pie in the sky, mired in the weeds, and perfect defeats good. In failure mode one, company leaders pursue so many transformative ideas that little value results. Nebulous concepts end up in fuzzy strategies and unscalable experiments. In failure mode two, perfectionism and risk aversion smother digital ideas, propping up the status quo rather than disrupting it. In failure mode three, leaders and executors pursue the final goal without completing incremental deliverables. Interest in the transformation wanes as the future never becomes the now.

All three failure modes develop their own momentum, making these initiatives tricky to manage. Digital lightning bolts require equal parts evangelism and technical accomplishment—half process reform, half product development. C-suite engagement is critical to ensure that the first lightning bolt hits its mark and, in so doing, is a scalable template for future lightning bolts. Small groups of handpicked innovators and experts need resourcing and authority to navigate around resistance.

But there’s nothing like early success to motivate and empower fast followers to replicate digital-transformation templates across companies as the new norm. Tata Steel’s first superheating project and Danone’s Opole plant are real-world lightning bolt examples: small, innovative teams plus digital experts and all-in leadership support led to rapid, replicable results. These templates are then scaled across the organization so that early disruptions become part of the new status quo. Then the process is repeated.

Own your ‘virtualization building code’

When replacing real-world activities with digital ones, error is inevitable. Containing it requires knowing the allowable error for physical certification and use. Further specifying error levels for each virtualization component provides a kind of “digital building code” that determines whether a model is authoritative (or accurate) enough to thread with other virtualizations—including reality as a source of truth.

People demonstrate their faith in building codes every time they enter a building. Designing a building to code—initially in a virtual form, on blueprints—is the first step in safely translating a structure from idea to physical reality. The analogy for digital virtualization is to build it to a code that verifiably matches digital results to the physical world, within an acceptable error range. At that point, the virtualization can be trusted as authoritative and safe to use.

The complexity of determining allowable error for a multitude of data points depends on industry and application. Companies decide what’s strategically important to retain in house and where an ecosystem of partners would be more effective. For example, SpaceX developed its own enterprise-resource-planning tool, WarpDrive, which tracked every facet of its operations. Requirements for running physical factories were decomposed into subrequirements for each virtualization component—supply chain, inventory, production, and finance—until the virtual factory matched the physical for SpaceX’s decisions.

Create the simplest digital thread of value

Once you successfully create an authoritative virtualization, you must digitally thread it with other virtualizations, or reality, to unlock value. But across most industries and sectors, integrating disparate models and operational tools remains a significant challenge. For example, many still struggle to fully connect design to manufacturing or to the supply base across all parts. Yet the connection of these disparate systems is often where the greatest value lies.

In technically advanced industries, such as automotive and aerospace, model integration is often more challenging than component virtualizations. Thousands of in-use models and simulations, each representing their own thin slice of reality, don’t integrate easily today. Stitching results together—with consistent physics and acceptable error levels—remains beyond today’s plug-and-play technology. Teams of experts focused on the task must do the stitching.

Digitally upskill your workforce

Companies often undervalue, underspend, and then underwhelm in their investments in human capabilities. Having an upskilled digital workforce takes time and tools. For many industries, digital transformation tools, foundational for training, are lacking and require significant development.

Understandably, many companies opt not to go it alone, particularly early on. Learning from experts greases digital lightning bolts, avoiding early missteps to build team confidence.

Yet over time, tangible leadership commitment to long-term training becomes vital. Tata Steel could easily have outsourced its analytics and IT, especially when early customer demands outstripped the initial digital model’s capabilities. The company’s commitment to ongoing training to improve these models internally, though slower than outsourcing, created a data-centric workforce and supporting culture that’s sustained its continued improvement.

Be deliberate and active in building a digital culture

The procurement leaders at one government agency recognized that a reactive maintenance cycle—in essence, waiting for components to fail before fixing them—was no longer viable. But achieving the reduced cost and downtime of predictive maintenance required more than new technologies and procedures. Instead, the top team encouraged start-up behavior: adopting a founder mindset, drafting a compelling vision statement, and empowering local leaders. Through an intensive communication program, accompanied by substantive changes in how people work, the agency gradually built a grassroots digital culture.

This energized ecosystem implemented predictive maintenance on a scale no top-down process could otherwise achieve. Similar culture-driven revolutions transformed software development and investment enterprises.

What good looks like

Once digital lightning bolts and a digital thread are in place, the driving question becomes, “How do we know our digital transformation is succeeding?” These indicators, in some number, are generally a good sign:

• Easy integration and parallelization emerge. Parallel efforts, even reversing formerly sequential ones, become possible as integration simplifies. Some functions may even be automated and baked into virtualization tools themselves, much like software development platforms today.
• Cycle times compress. With formerly physical activities now virtualized and parallelized, processes accelerate—sometimes by as much as two or three times. Software-like practices begin showing up everywhere. Team New Zealand’s foil configurations look more like software sprints than mechanical endeavors.
• Leapfrogging design delays becomes possible. Physical prototyping and testing are still needed for systems that push a model’s boundaries. But for systems in the virtualization sweet spot, digital design and testing start to eclipse their physical analogs. The old “fly before you buy” adage becomes “e-create before you aviate.”²Will Roper, “There is no spoon,” United States Air Force, October 7, 2020. After digital transformation, the process becomes the product as companies shift toward an environment of no physical prototypes or no more paper.
• A lower-cost mosaic arises. In addition to the direct cost savings of each preceding indicator, digital transformation aggregates myriad small cost savings of otherwise optimized industries. Tiled together, broader efficiencies are found as direct and indirect costs become more precise. Danone’s digital dairies saw this.
• The gaps among R&D, production, and the wider value chain start shrinking. With innovation happening inexpensively at digital speeds, software-like iteration becomes affordable for creating physical systems too. Connected to cost-saving production technologies, such as containerization for software and software-enabled manufacturing for hardware, digital threads enable smaller-scale production. This production continually evolves, eventually outperforming cheaper-but-fixed mass production. The medical-device industry is a heartbeat away from this lifesaving paradigm today.

Industry 4.0’s wave is cresting. In its disruptive wake, all technology becomes software, improving the physical world by coevolving in its digital twin. With such an abundant source of faster-than-real-time data, AI will soon learn to steer many physical industries. The result may not be the dystopia of The Matrix nor the utopia of tech-industry fantasy. It’s more likely to be somewhere in between. What matters now is being on the wave, not beneath it.