What if you could share data with partners, governments, and other organizations to drive innovation without breaking privacy laws? Wouldn't it be great if you could make better use of your company's carefully guarded customer data and maintain the highest standards of privacy and security? Imagine if you could create new revenue streams for your business by monetizing your data without compromising personal/confidential information? Such is the promise of synthetic data, which has the potential to revolutionize how the world uses and benefits from its data. In today's world, data really makes the world go round. It is the foundation of almost everything we do. And data becomes even more powerful and important when it is shared. Think about how much faster disease could be cured, or how much less waste, or how much more efficiently ecosystems could work if data could be shared freely. Of course, such an exchange is not possible today because we are limited to using our own data, which is well protected for good reason. What is synthetic data? Simply put, synthetic data is data artificially generated by an AI algorithm trained on a real data set.
The goal is to reproduce the statistical properties and patterns of an existing dataset by modeling its probability distribution and sampling. The algorithm essentially creates new data that has all the same characteristics as the original data, resulting in the same answer, but most importantly, any of the original data cannot ever be reconstructed from either the algorithm or the generated data. im synthetic data. . As a result, the synthetic dataset has the same predictive power as the original data, but does not contain the privacy concerns that limit the use of most of the original datasets. Here's an example: imagine a simple exercise in which you are interested in generating synthetic data about athletes, specifically height and speed. We can represent the relationship between these two variables as a simple linear function.
. if you take this function and want to create synthetic data, it's enough just to have the machine randomly generate a set of points that match the equation. This is our synthetic set. Same equation, but different values. Now imagine you're interested in height, speed, blood pressure, blood oxygen, etc. The data is much more complex and requires more complex non-linear equations to represent, and we need the power of AI to help us determine the "pattern.
" Using the same mindset as in our simple example, it is now possible to use trained AI to create data points that approximate this new, more complex "pattern" that we have learned, and thus create our synthetic dataset. Synthetic data is a boon for researchers. One example is what the National Institutes of Health (NIH) in the US is doing with Syntegra, an IT services startup. Syntegra uses its synthetic data engine to create and validate an unidentified copy of the NIH COVID-19 patient database of more than 2.7 million people screened and more than 413,000 patients who test positive for COVID-19. A synthetic dataset that accurately duplicates the statistical properties of the original dataset but contains no links to the original information can be shared and used by researchers around the world to learn more about the disease and accelerate progress in treatments and vaccines. While the pandemic has shown potential use cases for synthetic data focused on health research, we see the potential of this technology in a number of other industries.
For example, in the financial services industry, where restrictions on data usage and customer privacy are particularly restrictive, companies are starting to use artificial data to help them identify and address customer bias without violating data privacy regulations. Retailers are beginning to realize how they can create new revenue streams by selling synthetic copies of their customers' buying behavior that companies like consumer goods manufacturers would find extremely valuable, while keeping their customers' personal data a closely guarded secret. Business value: security, speed and scale While the use of synthetic data is still in its infancy today, massive growth is expected in the coming years as it provides companies with security, speed, and scalability in their data and AI workflows. Security: protectionand confidential information The most obvious benefit of synthetic data is that it eliminates the risk of exposing critical data and compromising the privacy and security of companies and customers. Techniques such as encryption, anonymization, and advanced privacy preservation (such as homomorphic encryption or secure multi-party computing) focus on protecting the original data and the information in that data that can be traced back to a person. As long as the original data is in the game, there is always the risk of being compromised or exposed in some way. Synthetic data doesn't mask or change the original data, it replaces it.
The most obvious benefit of synthetic data is that it eliminates the risk of exposing critical data and compromising the privacy and security of companies and customers. This is one of the highlights of the COVID-19 example mentioned earlier, and indeed it is a big argument in favor of the healthcare industry as a whole. Imagine if, from the very beginning, we combined all the data we have on everyone who has contracted this disease around the world and shared it with anyone who would like to use it. We would probably be better off, but legally speaking, there is no chance of that. The NIH initiative demonstrates how synthetic data can overcome the privacy barrier. Speed: fast data access Another big challenge companies face is getting their data quickly so they can start getting value out of it. Synthetic data removes the hurdles of privacy and security protocols that often make it difficult and time consuming to obtain and use data.
Consider the experience of one financial institution. The enterprise had a wealth of valuable data that could help decision makers solve various business problems. Yet the data was so well protected and controlled that accessing it was a difficult process, even if the data never left the company. In one case, it took six months to get even a small amount of data, which the analytics team used very quickly. Another six months followed just to get the update. To get around this access barrier, the company created synthetic data from its original data. The team can now continuously update and model the data, and generate constant, actionable insights on how to improve business performance.
Synthetic data removes the hurdles of privacy and security protocols that often make it difficult and time consuming to obtain and use data. In addition, with synthetic data, a company can quickly train machine learning models on large datasets, which means faster training, testing, and deployment of an AI solution. This solves a real problem that many companies face: not having enough data to train the model. Access to a large set of synthetic data gives machine learning engineers and data scientists more confidence in the results they get at different stages of model development, which means faster time to market with new products and services and, ultimately, more value. Faster. Scale: Sharing to Solve Bigger Problems Scale is a by-product of security and speed. Accessing data securely and quickly allows you to expand the amount of data you can analyze and, consequently, the types and number of problems you can solve.
This is attractive to large companies whose current modeling efforts tend to be quite narrow as they are only limited by the data they own. Companies can, of course, acquire third-party data in its "original" form, but this is often prohibitively expensive (and comes with associated privacy issues). Synthetic datasets from third parties make it much easier and cheaper for companies to augment their own data with additional data from many other sources so they can learn more about the problem they're trying to solve and get better answers - without worry. compromising someone's privacy. Scale is a by-product of security and speed. Accessing data securely and quickly allows you to expand the amount of data you can analyze and, consequently, the types and number of problems you can solve. Here is an example.
Each bank is obliged to identify and stop fraud itself and regulatory bodies. And each bank is on its own search, operating independently and devoting significant resources to the case because it is required by regulators and only the bank itself is allowed to look at its data for suspicious activity. If banks used syntheticdata, they could share information about their investigations and analyses. By combining their synthetic datasets with peers in the industry, they could get a holistic view of all people interacting with banks in a particular country, not just each bank, which would help simplify and speed up the discovery process and ultimately eliminate more fraud using fewer resources. Why doesn't everyone use it? The benefits of synthetic data are compelling and significant. But implementing them requires more than just hooking up an AI tool to analyze your datasets. Properly creating synthetic data requires people with really deep AI knowledge and specialized skills, as well as very specific, complex structures that allow a company to confirm that it has created what it set out to create.
This is a critical moment. The project team must be able to demonstrate to the business (or regulators or clients, if necessary) that the artificial data they create does indeed represent the original data, but cannot be associated with or disclose the original data. install in any way. It's really hard to do. If it does not match, important patterns in the original will be missing. This means that subsequent modeling efforts may overlook potentially great opportunities, or worse, lead to inaccurate conclusions. There is also the issue of bias, which can easily creep into AI models trained on human-created datasets that contain inherent historical biases.
If a company creates a synthetic dataset that simply copies the original, the new data will still have the same offsets. Therefore, you need to make complex adjustments to your AI models so that they can account for bias and produce a fairer and more representative set of synthetic data. And it's not easy, but it's possible. Properly creating synthetic data requires people with really deep AI knowledge and specialized skills, as well as very specific, complex structures that allow a company to confirm that it has created what it set out to create. Synthetic data can also be used to create datasets that are consistent with a pre-agreed definition of fairness. By using this metric as a constraint on the optimizing model, the new data set will not only accurately reflect the original, but will also meet that particular definition of fairness. As a result, this new valid dataset can be used to train the model without the need for bias reduction strategies such as algorithmic fairness, which can lead to accuracy trade-offs.
For example, Mostly.AI demonstrated its effectiveness on the well-known COMPAS recidivism dataset, which fueled the algorithmic results of racial discrimination. The Mostly.AI approach narrowed the gap between high COMPAS scores for African Americans (59%) and Caucasians (35%) to just 1% with "minimum trade-off in forecasting accuracy". In addition to verifying that the actual mechanics of creating synthetic data are robust, most companies also need to overcome cultural resistance to the concept. "It won't work in our company." "I don't trust it - it sounds unsafe.
" "Regulators will never go for it." We encountered this at a North American financial services firm we worked with. When we first broached this topic with some of the company's executives, we had to do a lot of work educating them, as well as risk and legal departments, about how synthetic data works. But now that they've changed their minds, they can't be stopped. Moving Forward: Education, Purpose and Skills For companies that want to efficiently create and use synthetic data, there are three main considerations to keep in mind to capitalize on these benefits: EDUCATION OBJECTIVE SKILLS Synthetic data is a new and complicated concept for most people, and comes with a lot of misconceptions. Before implementing any synthetic data program, it is important that all senior management, as well as risk managers and legal professionals, fully understand what it is, how it will be used, and what value it will bring to the enterprise. Looking to the Future: The Economics of Synthetic Data? The thirst for data to solve all sorts of problems is not going anywhere.
If institutions, universities, governments, and companies open the doors to their data—but in a synthetic way—the potential for the future is exciting. This could lead to the development of a prosperous synthetic economy.where parties create, buy, and sell data—or, in some cases, give it away for good reason—without worrying that individuals or companies might be compromised in any way. The greater availability of synthetic data will lead to federated learning, which will allow organizations to build intelligent systems trained on other organizations' datasets, democratizing data for the common good while respecting privacy and security. The thing is, if you can create synthetic data from your own data and it makes sense to share it, Synthetic data has exciting potential and many viable use cases in every industry imaginable, but it is still at the forefront of data science. How quickly it moves from its current state to real-world applications remains to be seen. But there is no doubt that organizations that can figure out how to create it and use it effectively will see significant benefits.