The Open Secure and Verifiable Half to a Brighter Future

Transcript

All right, everyone. Good morning. So, today uh we'll be talking to you about the uh open, secure, and uh verifiable path to a brighter future. Who here wants a brighter future? Who here wants that future to be more open, >> more secure, more verifiable, >> more the nobody likes the Okay, that was like a memecoin from four years ago. So that's fine.

So technological progress um especially enabled by uh the rapid growth of modern AI is uh appending a lot of assumptions that we had often in our subconscious about how the world works. Right? Before you could assume that your software has only a medium number of bugs. Now all the bugs are being found. Before you could assume your TV is just a TV. Now, your TV might be spying on you. If you had a private conversation outside, that was a private conversation. Now, you can use AI and you can use powerful

Gabend here could have imagined 5 years ago that there was literally such a thing as a cancer vacc. Now it's here like something like halfway there to proving who here imagined five years ago that one day my slides

would not be so ugly. Now it's happening. Who here has played World of Warcraft? So I think a few of you are going to appreciate this, right? You know, World of Warcraft, the first expansion pack before you're at level 60 and then you could go up to level 70, right? And well, you could go up level 70. So, first of all, everyone rapidly levels up at the same time, right?

Second, monsters that before could be were very difficult to slay suddenly become very easy to slay, right? Before, you know, you had your super almighty crazy dragon and then you level up 10 more levels and then you just like go right back and you kill it within 5 minutes and you're like, "Oh, wow. It's so easy now." But also all of the equipment, all of the skills, everything you worked hard for gets completely zeroed out. Everyone starts training, everyone starts understanding this new world of people's new abilities from scratch, right? But there are now new monsters that we just have to deal with.

This is a lot of what the last year of AI has felt like to me. And it's a crazy game. And unfortunately, it's also a compulsory game. So in 2023 I uh was uh first developing this concept of DAC right and uh I had this three-sided triangle in mind basically how do you deal with the challenges of very powerful very enabling but also dangerous technology right and the three paths that you could take are basically one is deceleration basically ban it out of fear the second is uh domination basically out of fear put all of your eggs into one powerful organization that takes over and that

and that controls everything and everyone. And the third one is uh of course doom, right? And uh there was this idea that maybe if we develop defensive technology more if we make it open and we broadly distribute it then and we make it verifiable so people could make sure that it's actually defensive that we could actually yeah like find our path away from all three of these extremes. Now I think it's pretty clear deceleration is not happening right the world is accelerating and so what is the path to get a good outcome now so what is domination right I think uh domination is this sort of default path that I think uh the world is uh already going

toward and I think many people are kind of sometimes subconsciously sometimes consciously pushing toward right and domination is uh the philosophy of security that says says we are secure because we have the best tools and nobody else gets to have tools. So this uh this tweet I have above I think is like everyone else, all of the people who don't have the tools will start feeling resentful, will start feeling very entitled to find all the ways that they can to get tools themselves because everyone needs to protect themselves and each other. And potentially, you know,

you if this is the only approach to security, then you get a race, right? So, if we do not come up with a better path, domination becomes the default, I think, is uh one of my thesis, right? So what is the alternative in general? Right, this is where you get to sort of defensive decentralized acceleration, right? Basically aggressively improve our security and I'll get into what security means in many areas to be able to better handle this new environment to be able to better protect each other across all physical, virtual, cyber, info, biological, everything. a philosophy of security that focuses less on retroactive deterrence and more on

proactive prevention. And also I do think that there is a role for you know what people call pacing the frontier basically for not like not actually progressing all of these frontier models so quickly that there's not enough time to do any of this. So the classic problem with any kind of security is basically who do you trust, right? And basic and the solution here is openness and verifiability. Right? If you want a uh a path for us to be able to survive and prosper in this new environment, a path that does not depend on handing over the keys to everyone's castle to one small group of people.

Then if you want something that is uh simultaneously respects local self-s sovereignty um and also that enables cooperation then you want open source. You want innovations that are made in one place to be usable and improvable in other places. You want verifiability. You want to be able to verify that something that you are getting actually has the properties that it says it has. You want to be sure the TV is not spying on you. Formally verified properties like basically use mathematical tools to actually verify security properties of uh at least digital components. You want to minimize trust dependencies. You

don't want to have like devices that call into 14 different servers for no reason. By default, consumer software, consumer tech, consumer anything is basically horrible swap. And uh you know, we have to like actively push against that direction. And cryptographic proofs basically do not prove by revealing all of the information about yourself because then you're leaking your privacy and that itself makes you less secure. You prove by proving the specific statement about yourself that convinces someone else that you are someone that is safe and that is a good idea to interact with.

These are all things that with uh zero knowledge proof technology and uh in many cases there's even a rollover for blockchains can help with. So in what areas do we need more ability to trust? We need to be able to trust our operating systems. There's projects uh some have been around for a long time. Graphine OS, Cubes OS that are trying to make operating systems that are more secure. Um, others that are newer, um, increasingly even projects to formally verify either the Linux kernel or alternative kernels, minimize the number of lines of code that bugs could potentially be in. Uh, so not your operating system, not your application.

Uh, we need to be able to trust our hardware. So we it is not acceptable for us to have TVs where we have no idea if they're spying on them. It is not acceptable for us to have hardware wallets where we have no idea if there was a SIM card stapled on top that's uh giving away your private keys to like possibly North Korean hackers, not your silicon, not your life, right? We need more open and more verifiable silicon. Um we need to be able to trust the things that are most relevant to our health. Right? So this on the left uh this is a UK critter. This is a device that uh like people in our community have been working on basically fully open- source um measurement of all kinds of air quality indicators and it also

has a fun Tamagotchi game. Uh so I think our goal is to really get to make air quality something that people can also really uh understand better and and uh understand you know where the air is good where the air is bad how to make the air quality better and and uh how to survive in that environment. Also with health you need to monitor the data outside of yourself you need to monitor the data inside of yourself and you need privacy for both. So on the left air quality monitor there's there's even a pilot project to use FHE to basically do decentralized science on air data without compromising any person's privacy. And then we need still need

more projects for your personal data. Most smart fitness devices are the exact opposite of being designed with personal self- sovereignty and personal privacy in mind. We need to be able to trust our sources of information and our sources of connection to other people. Today, social media is close to useless for forming opinions on controversial issues in society, for figuring out what to invest your money into, medical advice. Um, who here gets their advice on what coins to buy from uh Twitter?

Okay, I see I see zero hands up. Uh, so that's good. uh like keep not buying coins based on what you see on Twitter. Now, all three of these areas, they're adversarial, right? Lots of people who are running lots of bots have an interest in you, you acquiring an opinion that is in their interests and not in your interest. This is the problem. So, this is where we have been so far, right? So, one is uh decentralized social media. So these we've started to see uh projects uh such as forecaster we've started there is a lens uh there is uh also more things outside of the ethereum world I see uh I

see people from the the protect the >> well actually I I think you guys are uh making a very important point here which is uh how can we do better right basically it's okay for people to make opinionated communities but if we're making opinionated communities people need to have real and meaningful choice in which communities they are participating in if they're unhappy with one to self or to be able to collectively organize and make their own and so we need in decentralized social

media better interoperability. Um, so Globs is one ERC that does that. There are other approaches too. We need more user control. Users need to have direct agency over what things they see in uh social media uh protocols. Um, and uh even user moderation and not even not just relying on communities. And we need it to be we need the ability to create a new community, something that is actually practical and feasible to do. and that is not gated by being able to set up a whole new set of infrastructure. Making continuing the journey and making decentralized social media even fully decentralized and open is important. Um

also verification features um are important. being able to have better tools so that you can see what things people are saying are true, what things are false, what things are contentious, what things are believed differently in different communities. Um, also we need to be able to trust our communication tools, right? So what we don't need is closed software that has hidden back doors. We do not see we do not need uh back doors that get hacked and then become back doors that are controlled by somebody else. uh we do not need like this for example disappearing messages that turned out to actually not be disappearing and got and and kept being saved for years. What we need is open-

source secure formally verified ends to end encrypted messengers that protect your privacy. Users being able to control how their data is stored and also don't forget metadata privacy. It's not just about what you're saying, it's also about who is talking to whom and when. We need to be able to trust our AI. Everything is dependent on AI. Who here has used AI in the past month, in the past hour? AI is super important. It's basically becoming a second brain. But you don't want your second brain's thoughts to be going into a a third party corporation's servers and uh potentially being being used as training data, potentially being

hacked, used in arbitrary ways. If you can use a local AI where needed, you can use a remote model. You can use a remote frontier model in ways that minimize that frontier models authority. And I think there's interesting work on that. There's things like sandboxing. There's the idea that you can use a local model in order to sanitize the uh like rewrite remove private information from requests that go to a remote model. And then when requests come back from a remote model, make sure that those requests are actually in your interests. There's a lot of work to do here to try to get the power of powerful AI and more of the security and privacy of local AI at the same time. Um, privacy is a super important factor in security and trust,

right? And uh we really need to treat data as toxic waste by default. If you can, don't collect it at all. The answer is not to trust no one. The answer is also not to trust powerful corporations. The answer is also not to trust charismatic influencers who claim that they're a good and righteous rebellion against powerful corporations. What is the answer? That is something we need to figure out together. Thank you.