Data Sovereignty and Decentralized Data Science in Web3 with Dawn Song

everyone thanks a lot for being here so I'm getting started here thanks a lot for joining the talk my name is Dong Song I'm a professor in computer science at UC Berkeley the co-director of a new campus-wide center at Berkeley College Berkeley Center for responsible decentralized intelligence and also the founder at Oasis Labs so today I'll talk about data sovereignty and decentralized data science in web 3. as we all know data is a key driver for modern economy and the lifebloods of AI machine learning and more and more data it's been collected every day and personal and personal data is now a significant fraction of global GDP and it's increasing exponentially however a lot of this data is sensitive and handling the sensitive data has posed unprecedented challenges to both individuals and institutions so for example individuals have lost control over how their data is used oftentimes their data has been misused assault without their awareness and also a lot of the data has been sold and the premise that they have been anonymized but however a huge volume of research has shown that anonymization is insufficient to protect usage data privacy and also um for institutions they continue to suffer from large-scale data breaches and where hundreds of millions of users are even billions of users sensitive data or stolen by attackers and the Privacy regulations has also made its more and more cumbersome and expensive for business to comply and the biggest issue is it's still difficult to actually utilize data so the valuable data is now often said in data Sellers and it's not being utilized due to privacy concerns so all these posts uh significant challenges and they are only going to get more and more severe as we move forward in the digital era to the extent that if they are not addressed they can severely hinder Society progress and undermine human values and fundamental rights and hence there's an urgent need for a new paradigm that I call a responsible data economy so what do I mean by a responsible data economy um so first so there are a number of goals and principles so first we need to establish and enforce data rights the city rights service Foundation of data economy and prevents a misuse and abuse of data and two we need to enable Fair distribution and value created from data such as users should be able to gain sufficient benefit from their data and most importantly we want to ensure efficient data use to maximize the social welfare and economic efficiency so today we are in the area of web 2 where a platform maintains essential centralized control for at all layers of identities and data where identities are managed essentially by service providers data is maintained in centralized data silos and data is often processed without the knowledge of the users and to address the challenges I mentioned earlier we want to move into the Paradigm of web 3 where users are in control through this decentralized Paradigm where we can enable decentralized identity decentralized access control and that's something we call policy compliance decentralized computation such that we can control users can control the data usage without relying on intermediaries a number of risks new technologies that I call responsible data Technologies when putting together can help us to shift into this new paradigm so these responsible data Technologies include for example secure Computing that helps protect the computation process from leaking sensitive information such as using secure Hardware or cryptographic methods such as secure multiparty computation for the hormonal encryption and so on and then differential privacy that ensures the computation output from a only extensive information about individuals and the sensitive inputs and with fair to learning it ensures data States while data is stays on owner's device devices the different devices can together train machine models and do data analysis in a distribute manner and with distributed Ledger it helps provides a immutable log to ensure that to ensure users data rights and data usage as compliant so now uh at oecc we have been developing various Technologies to help ensure uh responsible data use and so on and I want to share uh a case study and uh a project that we have launched recently together with the Manhattan this is the first of its kind large-scale real world deployment for privacy preserving AI model fairness assessment and so right so in this case the goal is to do a fairness assessments for AI models used for example by Manta and Instagram they um so as we know with the AI becoming more and more ubiquitous and making automated decisions and it's important to ensure that these AI models are fair that's not biased by on gender race and so on so then it's important to actually measure the fairness of these AR models and there are different metrics for a um to measure air amount of fairness so for Simplicity essentially here we want to measure how the AI model influence results uh how they compare according to certain sensitive attributes such as uses race or gender and so on and so for example comparing the average inference results of users identified as uh you know different gender and so on so in order to do this computation it needs a lot of sensitive information and it's usually sensitive attributes and is they uses influence results and hence it's really important to be able to do this computation in a privacy preserving manner and this is where the OSU technology comes in where we have developed and combining different types of privacy preserving technologies that I mentioned earlier um by combining secret multi-party computation and homomorphic encryption and differential privacy and even some serial energy proofs we have together developed this new privacy preserving platform in collaboration with meta and this has been rolled out to essentially our Instagram users and um and then this data enables the measurements of AI model fairness for Instagram so um so this again as I mentioned this has been the first of its kind the large-scale real world deployment we have learned um also some important challenge lessons and challenges from this experience the challenge here is that developing this uh developing algorithms and protocols it takes some time but actually it takes much longer to actually do the real world production um Quality development and deployment so today this type of decentralized data science and responsible data used today is simply too expensive and there are a number of reasons why they are expensive essentially there are many challenges when you develop this type of decentralized Technologies to enable decentralized data science and so this is in uh essentially uh in analogy with for example for deep learning when before python tensorflow was developed it was actually very difficult uh labor intensive to even develop and train small neural networks because the programming abstraction wasn't there so similarly here for developing decentralized Technologies and essential data science we need better programming abstraction to make these things easier and also today the different types of decentralized systems they often stay in silos and are difficult to actually compose for example for factory training there are over a dozen different types of federal union Frameworks and each one settings is on Silo and it's actually very difficult for developers to use and so on so this motivated us as a research project at Berkeley we have developed a new system called co-link and that actually provides a system that essentially provides these decentralized programming abstraction it has built-in programming abstraction for decentralized identity private storage secure communication and decentralized protocol encapsulation so when we put out this together it essentially makes it much much easier to develop new decentralized applications and solutions and systems so so for example the example that I mentioned the work with the meta using this new framework instead of taking into many months to develop the system it would reduce the work to just a couple weeks and also here's another use case example where I can use in this framework it's very general so you can develop lots of different types of decentralized solutions so this is a command level supervision for SSH where you can simply use in the system just very quickly you can develop a supervised sh where um the server owner can observe remote guests recommend to run and can approve or reject certain SSH commands so without this together this programming abstraction together we also developed what's called colon is a unified platform and that integrates different state of the Arts cryptographic protocols for further learning for secure multi-party computation and even for Key Management and so on and so with this colon platform it enables a new open source platform for making decentralized data science easy and helps bridge the gap between crypto protocol design and real world deployment and to making all this magnitude cheap and faster to to deploy develop and deploy uh cryptographic decentralized cryptographic protocols and also make these protocols composable in a unified framework and builds a preset of carefully designed and standardized decentralized programming abstractions and to again make decentralized data science and other decentralized applications much easier so with the coding it can help accelerate the research development and deployment of decentralized data science it helps researchers and protocol designer and to have faster and easier deployments and to lead to Greater impact and for application developers it's much easier for application developers to utilize these protocols as building blocks in their applications oftentimes they can incorporate this Essentials protocols with just one line of code and also helps the community come together and to enable better reproducibility better reusability and build an ecosystem so with this then it helps individuals to better maintain better control of their in their personal data to provide stronger privacy and security guarantees and also can help them gain more benefits and fair value from their data and helps company organization to better utilize their data and build better products and services so without that uh the um so again today we are in the web 2 world I have it with new regulations it also helps and it actually helps users to gain more rights on their data for example one of the rights in gdpr is called rights to data portability where it's a essentially makes it required and also makes it easier for users to download their data from different third-party services and then users essentially can bring this data from third-party services into their personal Data Vault and uh so with the with the platform for decentralized data science and also the technologies that Oasis has been building we can build what I call data Commons for decentralized data science where did the owners and producers they can provide their data in encrypted form uh in this decentralized platform and register their encrypted data sets with data use policies and distributed that decentralizes the catalogs and to the Consumer cell analyst can search these catalogs and find relevant data and then they can write their data analytics and machine learning programs over these different data sets and data sources to to try to utilize the data and then they can submit these data analytics and machine learning programs to this decentralized data science platform which then essentially provides secure Computing to utilize the data and ensure the data is utilized in a compliant manner with the desired policies and to produce in the end the results so with this uh with this approach we can significantly reduce the friction of data usage remove data silos and provide stronger protection for security and privacy so I strongly believe that in 10 years this type of data try sending the comments will become predominant in 10 years this address into the comments will become predominant ways of utilizing diverse sources of data enabling ownership economy where users benefits from their data as owners and partners and in 10 years huge the data stewards for dishes and Chassis will be a new class of entities important in the city ecosystem to help manage and protect users data and grow its value and in 10 years huge economic value will be created through these new platforms after the trust and Commons other magnitude higher than today's Marketplace so with this a new approach we hope that we can really enable this Paradigm Shift shifting from centralized control in web 2 to user control in in web 3. so uh with this approach we can help make decentralized programming and decentralize data science easy so that's about how we can actually build public goods with the stated comments and essential data science so I also want to just briefly make an announcements that we um so at Berkeley also with RDI we are along we have just launched a new ckp and web3 hackathon and it started March 1st the sign up open summer March 1st ends on May 2nd and this is in collaboration thanks to many partners and sponsors sponsors with up to 200k Plus in prices and many leading researchers and cryptographers as judges and so on and has a number of tracks in the hackathon for to enable decentralized collaboration for the community to come together to build the key through knowledge proof Technologies and it's hosted by Berkeley RDI where I'm also a co-director which aims to advance the Science and Technology I've worked through decentralization and decentralized intelligence and you can learn more about the RDI also and the research at rti's website and what I want to mention is here is many of you if you are interested you can check out the moocs so we have Let's uh in collaboration within other institutions we have a lot a number of moocs including the Define mooc the entrepreneurship in web 3 mooc and also currently we are running the zkp mooc you can go to ckduction learning.org to learn more about the the mooc and Serena's proofs and we have over 3 500 people enrolled across the globe and that's also why we are holding the [Music] as well and for people who are interested in building new Ventures you can also check out the Berkeley blockchain accelerator which has incubated 85 teams with a total follow-on funding for over 450 million dollars so with that again you can you can by going to the uh check out the barcode for the work the the zkp hackathon and with that thanks thanks everyone