Welcome to the OpenDP use case Q&A series, where we explore how our collaborators in the community are using and integrating OpenDP in their work. First up, we have Oblivious and we’ll be chatting with Jack Fitzsimons to learn about all the great work they're doing in the PETs community.
Q: Why don’t you start by introducing yourself?
Hello there! I'm Jack, one of the founders at Oblivious, an Irish-rooted company that specializes in privacy-enhancing technology. I lead a team of developers, security specialists, and product designers. We are dedicated to our mission to maximize the opportunities of the world's most impactful data sources.
My journey into privacy tech actually started with a background in machine learning and data science. Too often, we couldn't get through the red tape required to work with the data we needed to do our jobs. So we decided to focus on building solutions for our fellow data-alchemists.
Q: What is Oblivious’ mission?
We're trying to build a world where data respects its boundaries, trust is brokered via reliable technologies, and privacy is the default, not just an afterthought.
Today, inter-organisational trust is managed by terms and conditions, privacy policies and compliance certificates, none of which are particularly foolproof. Big tech tends to build their own systems whenever they can and there is a massive advantage in favor of industry incumbents based on their track record. Data scientists rarely have access to the most sensitive data sources and when they do, they often get access far below the level of granularity required to do their job.
We want to change this by integrating privacy enhancing technologies seamlessly into how every data scientist and developer already works. With minimal changes along the way and decent gamification to incentivize our tech-oriented colleagues to get started, we think the snowballing R&D of the last 20 years is ready to go mainstream.
With the support of the thorough and thoughtful colleagues at OpenDP, we're moving these concepts from daydreams to delivery.
Q: How has Oblivious used the OpenDP Library and SmartNoise (OpenDP/Microsoft partnership)?
OpenDP and SmartNoise have really been game-changers for us as we endeavor to reshape data access and collaboration. They've been the backbone of our efforts to deliver dynamic data disclosure controls which we use across all kinds of projects.
The starting point was our engagement with the UN Privacy Enhancing Technologies (PET) Lab, a collective of national statistical offices with a shared mission to use PETs for good. We undertook an interesting pilot to reconcile international trade data. A task that might appear straightforward; yet is marred by complexities and sensitivities at granular levels, with no countries wanting to share fine grained logistics information with one another.
We combined the power of trusted execution environments with OpenDP/SmartNoise to carry out record linkage within the trusted execution environment (TEE), keeping the data passing through completely opaque while maintaining transparency of the processing. This approach, coupled with the use of OpenDP/SmartNoise for output disclosure control, enabled the safe dissemination of statistics.
The project culminated in a comprehensive PDF report, watermarked and signed with the TEE's attestation document. This ensured that the generation of the report was transparent, and any amendments post-generation would be detectable.
Our collaboration with OpenDP extended to the private sector with a major US telecom company. Their need was to extract dynamic and private insights from highly sensitive customer data without compromising the integrity, confidentiality, or security of the information. OpenDP's potential shone bright in this scenario, outperforming traditional manual data disclosure controls in cost, timeliness, and theoretical security rigor.
Q: What are some opportunities Oblivious has created to engage the PETs community?
It’s a shared belief in the PETs community that progress is most potent when knowledge is collaboratively shared and collectively built upon. With this in mind, we teamed up with the CeADAR Institute in Ireland to develop a hackathon for the Human-Centred Artificial Intelligence Masters students across European universities. This platform gamified the learning and usage of differential privacy and PETs in general, enabling budding data scientists the opportunity to handle more impactful data sources.
Encouraged by the positive outcome of the hackathon, we led a dedicated PETs stream at the 2022 UN PET Lab Hackathon. We developed a simple API-based interface allowing users to interact with OpenDP, SmartNoise, and DiffPrivLib on a dataset without direct access. The unique competition format placed the onus of balancing privacy and utility on the data scientist. We had over 300 participants from 33 countries join in, with on-site events in institutions like Harvard, Toronto Metropolitan University, and ETH Zürich.
These experiences have underscored the value of robust feedback loops between data scientists and PET providers. By partnering with organizations such as OpenDP, IBM and Microsoft, we are contributing to the advancement of privacy-enhancing technologies. We're excited to continue this journey, developing the next generation of secure, privacy-focused data solutions.
Q: Oblivious recently announced Antigranular. Could you tell us more about that?
Antigranular is our competition platform that's turning data privacy into a fun and interactive experience. We're taking the idea of data competitions to a whole new level by focusing on the secure use of data within Trusted Execution Environments (TEEs) and limiting data handling to interactive differential privacy operations.
We've integrated a selection of differential privacy frameworks, like OpenDP, SmartNoise, and DiffPrivLib, into the mix. We're planning to add even more in the future, ensuring that Antigranular evolves with the data science and broader PETs community.
It's not just about winning competitions, though. We're looking to build a community of data scientists who understand and value Privacy Enhancing Technologies. The aim is to encourage learning and stimulate conversations about data privacy, turning it from an obligation into an intriguing challenge.
We've also introduced an exciting scoring system that marries accuracy with privacy. So, the better you are at creating effective models while preserving privacy, the higher you climb on our leaderboards.
And here's where it gets even more interesting. User input matters to us, and we see it as an integral part of improving PETs. Antigranular is as much about creating feedback loops between data scientists and PET developers as it is about competition. It's a space for collaboration, learning, and growth, aimed at enhancing the secure and private use of data. You can even directly share your Jupyter Notebooks so others can learn, enhance and gain inspiration from your work to date.
In short, Antigranular is our way of gamifying privacy. It's about making privacy exciting and accessible while building a community that's passionate about responsible data use. We can’t foresee it being a smart “business” endeavor, but we do think it’s a crucial step in making privacy enhancing technologies the new norm.
Q: The first Eyes Off Data Summit is planned for later this month (July 20-21, 2023). The summit will showcase notable PETs experts from industry and academia. What is the impact you’re hoping to achieve and plans for the future?
The upcoming Eyes-Off Data Summit, which we're hosting in Ireland, isn't aiming to be just another tech event. We're setting our sights on something different - getting a diverse crowd of security folks, legal professionals, and data enthusiasts under one roof. Together, we hope to tackle challenges and brainstorm on ways to make Privacy Enhancing Technologies (PETs) more accessible and secure for everyone.
We've got two main parts planned for the event. "Words" is all about sharing insights, with talks and discussions from stakeholders knee-deep in sensitive data. "Actions" is about getting hands-on with PETs through workshops and a hackathon, running both onsite and virtually.
We really hope that it is more than just a business gathering - but rather a community event. The tools we'll use in the hackathon will go open source, integrating frameworks from OpenDP, SmartNoise, and DiffPrivLib, and possibly more. We’ve also invited a broad range of organizations who will showcase alternative PET frameworks.
We've managed to get some fantastic speakers onboard - from organizations like the UN, the OECD, the Royal Society, the Information Commissioner's Office UK, various national statistics offices, to big multinationals like Intel, Microsoft, and Mastercard. We even have a few exciting members of the OpenDP community including Salil Vadhan, the Co-Lead of OpenDP and Raphaël De Fondville and Pauline Maury-Laribiere, both from the Swiss Federal Statistics Office and OpenDP Fellows.
We hope everyone attending will come with an open mind and be ready to contribute, regardless of their role. As I always say, "a rising tide raises all boats". The Eyes-Off Data Summit is about advancing the dialogue around data privacy and providing hands-on experience with PETs.
If anyone in the OpenDP community would like to get involved in the inaugural event, please do reach out at hello@oblivious.ai, on LinkedIn or Discord.
Want to be featured in a future use case Q&A? Email us at info@opendp.org and include details of how OpenDP helped support your project!
