Filecoin & AI: Building the Next Generation of Data Storage | Clara Tsao - Filecoin Foundation

good afternoon everyone thank you for coming uh today I'm going to talk about a Hot Topic that I've heard all throughout the week at e thumber which is really around Ai and web 3 um and for those that may not know um I am the founding officer of a project called the filecoin foundation we do long-term governance of the filecoin network and we are also the largest decentralized storage Network in the world and so we def want to talk about what we're doing in this topic but also why it's more important than ever to have decentralized storage when we think about the growth in AI so first off um what you see behind me is actually a map of all of our data centers around the world the way that our decentralized storage network works is we have 3,300 storage providers today um in every single continent that are running their own storage operation these can be 100 person operations huge data centers some with even more security than going through the airport and others are more mom and pop shops people who want to really think about storing local data and doing so that doesn't really belong in big Tech so um this is just to show the scale of um our data centers worldwide and why AI well for those that are not as familiar AI is growing at an absurdly large rate Enterprises are adopting AI by 20 we're looking at a 34% growth and a huge Market uh out there uh in fact around I think um 11.4 billion so this is um I think it's worth even more today from that 2021 statistic behind me why is storage so important well this is a pretty obvious answer data is the fuel of AI um today the more data a model is trained out out of the more accurate um and AI data set is right when you are thinking about use cases out there um we want to make sure that the data most AI is trained on is coming from verified sources and so this is um these are critical reasons why decentralized storage is so important because we can really take the cryptographic proof um used in blockchain to to determine where probes of data today originate how and where data is stored we have the ability to prove where it's being stored and when it's being altered who can access them and for how long as well as also evaluate how AI models are leveraging this data and there are really three advantages to uh filecoin that I I will go through today the first is really the scale and cost um today AI is very very expensive to compute but also we're using more data than ever and that storage cost is also really high it's a barrier for a lot of companies today to innovate if they're paying hundreds of thousands of dollars on their AWS bill every month the second I'm going to talk about is trust through our proof of storage mechanism and then third is really resilience to Monopoly ownership who here has had to pay Apple more and more dollars every month because you've run out of storage and I don't know what that number will be when we're 50 or 100 years from now but I'm sure that data cost will continue to be higher and higher and so we definitely want to make sure that people are also storing their data in a place that isn't really hard to recover and something that users can afford to to access at any point in time so um I did want to talk about first just the scale of the amount of data that AI produces at any point on the left you see self-driving cars if you have ever visited San Francisco recently you'll see that these cars are everywhere but if you guys don't know this each self-driving car produces around 1.4 terabytes of data per hour and that is around 5.8 pedabytes per year just one car alone if there are traffic accidents or issues that happen this actually happened in San Francisco local officials will want to see that data they want to see where the cars are trained off of right so we're really talking about real world use cases today where uh storing And archiving data is really critical for all kinds of innovation Ai and then on the right um obviously um a lot of products in AI like cha GPT uh they are trained off of tons of different models and data sets and storing that uh definitely takes a lot of storage and compute power and this brings me to why filecoin or why decentralized Storage Solutions filecoin is really built to scale with AI um today like I said before we are the largest storage Network out there we have around 7.1 exabytes of available storage and we're only growing as we have more and more storage providers enter our ecosystem so um that that is enough to train chat GPT around 15 million times um but you can also Imagine the businesses built on AI that will need that storage and compute power the other thing about AI is trust in AI is actually uh declining um we were just at Davos two months ago where we met with a number of policy makers and people that are not necessarily crypto native but they're also not technology native and they are so scared about AI because it's really hard to trust um in fact 61% of people today are worryed about trusting AI we've seen scam phone calls by AI today there's been a lot of other use cases like deep fakes that really make people worry about using or interfacing with AI and only one in two people actually think the benefits of AI outweigh the risks so um this is where the trust layer really comes in how do we actually make sure that it's safe and safe enough that we can trust so one unique aspect of filecoin is we have something known as proof of storage which really proves that data is being stored somewhere and that is being actually checked every 24 hours so um this is something that has allowed us to really verify where data is coming from and how it's being altered over time on the right you can see here this is actually a project that we funded um around a year and a half ago um a group called Starling Lab at Sanford University was using our technology to really make sure that journalists and everyday citizen journalists on the ground in Ukraine they can actually use photos around what's happening in the war and let that photograph be evidence used in international criminal court one day and as you guys may be aware of um disinformation misinformation has been online for a while now and most people want to make sure that they can prove that an image is what it is and that it wasn't taking photos from a war from 20 years ago and repurposing that in front of a headline so those are just use cases of our technology but also all kinds of decentralized storage why it's so powerful than ever because people don't necessarily know what to trust anymore they don't know if an image has been altered or if it's generated completely by Ai and we can actually prove this out and we are working with a ton of different startups and companies to do exactly just that the other aspect is transparency and provenance and so um a lot this is a big debate in AI right now over models and making sure those models are open and accessible by everyone today we store uh the largest essentialized network of open and public data to date and we want to make sure that we can also keep different AI models open so that people can check for the safety portions of what models are being run out of here is just a couple of examples of data store today on a on filecoin um we've worked with a lot of groups like CERN um NASA we are actually backing up MIT open learning data so we really believe that a lot of content today needs to be free and accessible for everyone and so those are just a couple of examples a lot of these institutions you see behind me they will have to spend normally a ton of money to just back up data that they want to offer to the public and so this is why um having a decentralized network and also the cost being a fraction of AWS or other centralized providers really allows us to also um be able to offer data as a public good to everyone using it in in the internet so the third point I want to to make is data today is owned in the handful most of the data we see and use today is owned by four major companies I think all of you guys know who they are and this is only going to get worse with AI um all the major AI companies today actually have a large stake invested by most of big Tech and this is really really scary I mean Amazon U put in $4 billion for anthropic as you guys are well aware Microsoft has a huge investment in open Ai and we also see this as well in a lot of AI companies being owned by China right and so this is really scary because these companies will have more and more of our data ever before and sometimes this is not always the most visible to most users and by having a decentralized uh storage Network we do allow and make sure that the data that you want to store is actually accessible by the users that want to store it and also being able to prove that and there is a lot of risk for centralization of AI um if you guys remember on the left a few years ago an incident called Cambridge analytica took place where Facebook was using a ton of data and companies were taking advantage of this to really Target users for election security and there is also a question of is there another cambrige analytica in the age of AI that will happen right so we definitely want to make sure that AI can be decentralized as much as possible and also not necessarily in the hands of just a handful of companies and so um I've already touched about this but decentralized storage networks allow for a lot of resiliency to data ownership um all of our storage providers are like many entrepreneurs around the world and they can really design um data storage to be Enterprise great in terms of of storage they can design particular areas of storage to be only accessible by you know a few people they can determine the time that data is stored right so this is really really amazing to have a network of different people be able to be storage providers and I wanted to also talk about something really exciting we've been working on as an ecosystem the last two or 3 years is really decentralized compute Beyond just AI training models um today compute and compute power is actually facing a supply chain problem um Nvidia and a number of companies um they have to wait months if not years to even get the right parts to grow uh and meet the demands of what most companies need for compute and when we think about having a Global Network of data centers it really solves the problem of not necessarily having those uh Hardware having to be localized in a particular part of the of the world in fact you can have it almost anywhere and so I wanted to leave with a couple of quotes here you guys can read it yourself but a ton of people um definitely are bullish on decentralized storage centralized compute and um we truly believe that filecoin is going to be a market leader in uh really thinking about the next generation of AI as we we think about the challenges that AI faces when it comes to having all your data in the hands of just a few companies so thank you so much I'll stay around uh to answer any questions I don't know if there's any and I think I have some extra time so any questions from the audience and if if folks want to keep in touch my email is on this first slide so definitely reach out if you're interested in collaborating working with us and thank you so much for having me