Synthetic Data, Real Insights

by Alexander Webb Advisor, GoPoint Ventures

Synthetic data may be artificial, but it has real value. What is synthetic data and what is it used for?

Synthetic data is artificially generated information. Although it has many uses, synthetic data is particularly valuable as a way to preserve privacy in sensitive data, and as a way to model rare “black swan” events.

For example, Buzz Solutions, a GoPoint portfolio company, uses synthetic images alongside existing proprietary data to help train its cutting-edge AI tools which can detect faults in critical infrastructure. “Obviously we cannot light an electricity substation on fire,” says Buzz Solutions Co-Founder Vikram Chaudhry “so we created a 3D model and trained our AI based on this synthetic data.”

This approach has worked well, with Buzz Solutions’ AI performing with significantly greater accuracy than any other tested model in an evaluation run by the New York Power Authority. But, Vik explains, it’s a rookie mistake to think you can easily train a model with only synthetic data. “What is most important is that we have an unrivaled set of real-world data, and our synthetic data is built on top of that stack.”

A host of potential users have taken notice, with the United States Army in recent years ramping up its use and acquisition of synthetic data, highlighting conflict as one of the most notable “black swan” use cases for artificially generated information. Moreover, synthetic data using satellite information is becoming increasingly useful. Healthcare is also an industry embracing synthetic data, with the BBC reporting on AI models which help find cancers that doctors would otherwise miss.

Synthetic data is not new, but it is becoming increasingly important. As early as 1987, an early self-driving car prototype was trained on 1200 synthetic images. Only a few decades later, firms are able to harness vastly more data. By 2025, enough data will be created every day to fill 212 million DVDs. But today’s era of data abundance allows for much more sophisticated use cases, including using synthetic financial data as part of risk-management or quantitative trading models.

Understanding the use and creation of synthetic data is key. In the wrong hands, synthetic data could even be a double edged sword. Before the 2008 financial crisis, many analysts thought a sustained, dramatic decrease in real estate prices across the entire United States was unlikely or impossible. Had these assumptions been programmed into a risk-management model, it could have exacerbated the financial crisis by giving regulators and banks a false sense of security.

A range of businesses, from banks to healthcare providers, find themselves with crucially important data which cannot be legally shared with other actors. Instead, a synthetic dataset which preserves the character of the original set, while maintaining anonymity, can allow for insights that respect privacy.

Synthetic data has many advantages, from being cheap to produce and also allowing greater privacy, a big plus in regulated industries. Many companies find themselves with crucially important data which cannot be legally shared as-is. A synthetic dataset which preserves the character of the original set, while maintaining anonymity, can allow for insights that respect privacy. Take Anthem Health Insurance’s 2022 partnership with Google Cloud to detect health abnormalities and fight fraud, while preserving patient privacy in line with laws and ethical regulations.

Unfortunately, the creation of new synthetic data and images also presents opportunities for fraud and misuse. This will likely create new business opportunities for companies which can detect the fakes. Insurance companies may be particularly interested in detection use cases.

Ultimately, synthetic data is a powerful tool, but it is a complement, not a substitute, for real data and real understanding. Google’s Gemini AI debacle, in which its AI model infamously created both false and offensive outputs, such as diverse Nazi soldiers, dramatically shows how synthetic images and data are only as realistic or reliable as the models and assumptions which create them.

Even “real” data, if incomplete or biased, can lead to erroneous results. Consider the famous “blue zones”—areas where locals supposedly live unusually long lives. A study by academic Saul Newman largely attributes the phenomenon to poor data collection with a side of pension fraud.

If Newman is correct, the use of error-ridden blue zone data to create a synthetic data model about human health and longevity could have produced nonsensical, or perhaps even dangerous results.

Regardless of the truth about blue zones, the debate underlines how users of data, whether synthetic or “real”, need to have a strong understanding of the underlying data, and a detectives’ appreciation for what could be misinterpreted or misunderstood. It also requires a judicious and educated approach to using data to make decisions.

When even “real” data depends on careful interpretation and analysis, a business that believes they can rely on synthetic data alone is likely fooling themselves. At the same time, today’s wave of synthetic data companies need to find a way to stand out amidst an increasingly competitive field where image and data creation may soon be commoditized.

There’s no question that demand for synthetic data will rapidly rise in many industries. As AI technology improves, foundation models proliferate, and compute cost continues to decline, synthetic data will be used almost everywhere. However, widespread adoption will be partially a function of easy-to-use interface layers and workflow solutions which make synthetic data–as well as its creation and analysis–accessible.

Just as the internet only became practically usable for millions after the advent of modern web browsers and graphical user interfaces, many synthetic data tools and consumer-facing use cases may only see widespread adoption once an easy to use workflow solution is implemented.

Share This Post

More To Explore

Optimizing Inspections for Wildfire Risk

What does it take to better align utility inspection processes with wildfire risk goals? Buzz is working with utilities and partners to tackle this challenge.

Buzz Solutions Announces Strong Growth as Power Utilities Accelerate Adoption of AI to Modernize Inspections

Buzz Solutions announcement of major milestones including the launch of PowerAI Basic to help utilities begin adopting AI tools.

Investing in the Age of National Security: How Geopolitics Shape Today’s Startups

In the early 2000s, Silicon Valley saw massive growth as a largely unregulated internet allowed for a “move fast and break things” era of innovation. Billions were made by taking legacy businesses and digitizing them, while startups faced minimal compliance and overhead costs in a period of laissez-faire legislation. Globalization encouraged founders to frame their company as eventually serving every nation on earth. National security concerns were largely an afterthought.

All that has changed.

Buzz Solutions Harnesses AI To Prevent Wildfires And Blackouts

Buzz Solutions is highlighted on Forbes.com for “harnessing technology to act as a force multiplier for grid operators struggling to maintain their transmission and distribution systems in a drier, more combustible post-climate world.”

Corporate Software Solutions – What’s the Right Lens?

There are many terms thrown out in the venture world like SaaS, workflow solutions, productivity tools, analytics suites, data orchestrations, and smart integrations. Here are 5 key factors that drive our scorecard when we evaluate a new venture in software.

Synthetic Data, Real Insights

The Electric Grid’s Growing Adoption of AI

Productivity tools will allow power companies to shift skilled labor away from mundane tasks. Faster and better inspections will allow companies to maintain higher run times and mitigate tail risk. Localized assets like the standby generator in your home and smart thermostats will work in tandem with advanced modeling to reduce blackouts during times of peak demand and reduced supply. And behind it all will be numerous artificial intelligence platforms.

Alignment and Collaboration Between Founders and Investors

A good venture partner should be a collaborative advisor and sounding board, helping founders sort through the various strategic paths as the company grows. And good founders should know how to leverage their VC’s knowledge and experience while still being a decisive leader.

Summit-based Company Adds Extra Shine to Car Wash Experience

Spark Car Wash’s founder Kyle Van Decker. sits down with NJ Biz to discuss the start of the Summit-based business, the features, recent milestones, plans for expansion and more.

Buzz Solutions Raises $5M to Accelerate Adoption of its AI-Powered Platform for Visual Grid Infrastructure Inspections and Monitoring

Buzz Solutions announced it closed a $5 million round of growth funding. GoPoint Ventures led the round, with additional participation from Blackhorn Ventures and MaC Venture Capital.

Adopting AI Is Just The Beginning For Utility Companies

As utility leaders begin to execute their 2024 digital strategies, now is the time to be crystal clear about the roles, responsibilities, and risks associated with digitizing such critical legacy systems.