Data breaches are on the rise. This fuels growing demand for synthetic data: artificially generated information not based on real-world events. One key advantage of synthetic data is that it’s a high-quality alternative for when real data is unavailable or restricted due to privacy concerns, says Dr Uzair Javaid,
In addition, real data is significantly more expensive and time-consuming than generating synthetic data. “A painting by an artist may take one week, but AI can now generate it within a few minutes and with the same visual fidelity. You do the math there,” he analogises.
To bridge the gap, he co-founded the synthetic data-generating platform Betterdata in 2021. The company’s current product was launched in 2023, and it has already secured contracts with clients such as the US Department of Homeland Security and Kajima Corporation.
Betterdata offers two types of generative models. Non-promptable models are intended for users with data expertise as they require structured input data, and allow for more controlled training and generation of synthetic datasets. Promptable models enable users to generate synthetic data using natural language input regardless of technical expertise.
At present, only the non-promptable generative models are utilised as Betterdata will only be releasing its promptable model in Q3 this year. The programme will allow users to generate synthetic data using natural language for all data scenarios, even when zero data is available. Because Betterdata’s models are so advanced, Dr Uzair says, they are able to preserve the data quality of their synthetic data while making it impossible to reveal a customer’s identity.
Originally from Peshawar, Pakistan, he earned a master’s scholarship at Karlsruhe Institute of Technology in Germany before embarking on a PhD scholarship in computer science at NUS. He’s candid about being something of an overachiever. “During my PhD, I explored data privacy and security, and deep-dived into privacy engineering.
“Fast-forward to 2021, I had 15 publications under my name when I needed only three to graduate,” he recounts. He has also ethically hacked the Ethereum blockchain by reverse-engineering its encryption algorithm, gaining access to 670 wallets.
It is a testament to the raging fire in his belly that he clung to his convictions in privacy engineering. He hopes to serve the community by finding a better alternative to encryption and rethinking advocacy in data privacy.
Thus far, Dr Uzair’s biggest challenge lies in the probabilistic nature of data, which can be difficult to control because generative models work on probabilities and estimates. Starting with probabilistic outputs before making them as deterministic as possible can be a complex endeavour. That said, deterministic data is crucial as it can be further developed.
“If you get a new outcome each time, you’re taking a step in every direction and going nowhere, which is a supreme waste of time.”
Dr Uzair Javaid on a disadvantage of probabilistic data
In his experience, most of us do not care about privacy until we lose it. It is his hope that we become more proactive rather than reactive about privacy issues; proactivity implies assuming the worst will happen and building appropriate systems to prevent it, whereas reactivity refers to fixing privacy issues after they arise. Reactive people usually suffer five to ten times the time and cost losses, he points out.
To Dr Uzair, privacy is a universal right, not a privilege. Protecting our data isn’t just a checkbox exercise, but a multi-layered issue with legal, technical, ethical, and political components. Individuals and societies are affected alike.
“And yet, it is too often relegated to regulators or niche startups. We need a shift in mindset. Like it or not, everyone from data creators to data consumers has a role to play.”
Photography Mun Kong
Art direction Chia Wei Choong
Hair Yue Qi using GHD
Makeup Keith Bryant Lee using Clarins
Photography assistant Melvin Leong
