The "Linux of AI": Building Personal Intelligence

The perfect AI dream is to own a personal AI whose loyalty ensures it works in your best interest and protects your private data while you interact with it. In this powerful Mindplex podcast, our host Dr. Mihaela Ulieru launches a lively conversation about decentralised AI, welcoming Raza Rassool, serial entrepreneur, inventor with over 75 patents, and founder of Kwaai, to explore the urgent mission of building the "Linux of AI."

The episode discusses the distinction between Personal AI and Personalized AI: one is truly yours, the other serves the vendor's interests.

Transcript

Copy Transcript Copied

SPEAKER_00 0:00

So um there's this uh massive wealth transfer that's happening right under our noses. It's it's going on. If we look at what success would be for um Singularity net or Quai or any of the movements that are trying to resist this trend, it would be let's say in a few years from now, a decade from now, we have the Linux of AI. We have an alternate where we've got a fabric uh or an infrastructure, a digital public infrastructure that's stitched together with all of our computers, where each one of our machines, when we're not using it, when it's idle, in its idle cycle and still burning electricity, it's now available to to participate in um in AI jobs or any compute job. And then each each of these nodes would have its own wallet and it would earn a passive income that it could spend on that same network, and so we're not burning tokens forever and and continuously becoming subscribers and renters of our own um intelligence. We are now part owners of a digital public infrastructure built by the people for the people. Gosh, I'm on a soapbox, I need to step down. Yeah. Is this model aligned with human values? That is just an intractable problem to solve because whose human values are you talking about? All you can do is focus on the individual, and the individual saying, I want to make sure that this computer system is loyal to me as a loyal agent. It's not a double agent, in fact, working on behalf of a vendor, um, which you sometimes get the impression that you know Google is doing that. When you ask Google a question, you know, the first the first results that come up are sponsored results. So, hey, so you weren't actually acting on my best interests. You were you were giving me the advice of the the the highest bidders in your in your marketplace. Um so we want a loyal agent, one that's truly acting on our behalf and representing our best interest. I think that's the exciting part is actually engaging, having a digital twin that engages with you to figure out well what are your interests, what is what is your what are your better aspirations, which is which is I don't know, computer systems don't don't do that, they they behave like you know inanimate tools like my calculator, and um it's it's not asking what you know what what does success look like for you. Imagine if a system did that, um your personal assistant, what does success look like for you? Um and it might be freeing up time, it might be it's and it's different for every person. And so now once you are able to engage with this ongoing conversation about what is success for you and what are your aspirations, you've now got a an agent, a digital twin that is aligned with that personal mission for you. And now if everyone is armed with that personal assistant that can give them a 10x improvement, rather than corporations you know always seeking their 10x improvement. If if we could each of us get a 10x improvement, imagine what that does for society.

SPEAKER_01 4:25

Today we continue to meet the speakers of the upcoming AGI conference to get a glimpse into their work in anticipation of hearing the keynotes in San Francisco in July 2730. And that will be, yes, 2026, this year, coming very soon. For this episode, we are joined by someone who has repeatedly transformed emerging technologies long before they became mainstream. Reita Rasul is a serial entrepreneur, inventor, and visionary whose career spans AI, digital media, imaging, semiconductors, and distributed systems. He holds, I mean, he is a legend when it comes to inventions, more than 75 patents, has built technologies that earn both Oscar and Emmy recognition, and has successfully led companies acquired by Google and other technology leaders. So, Reza, welcome to the podcast.

SPEAKER_00 5:24

Thank you so much, Michaela. Um, thank you for having me on your podcast. I'm looking forward to the AGI conference. Thanks for your generous introduction.

SPEAKER_01 5:35

No, really, we are so enthusiastic, and thank you for accepting our invitation as a keynote speaker at the conference. Uh, you know, after you made it, achieving all that, you retired and founded Quay, an ambitious open source initiative which you coined as the Linux of AI. What an exciting retirement project. Like with singularity that our founder, Dr. McGurzel, did bet his whole career as a father of AGI on deploying the Linux of AGI. So you are definitely aligned, as you probably know. But but before we delve into it all, you know, I'm sure our audience, especially the younger ones, would like to learn from your experience as a successful inventor and entrepreneur. So I'd like to start uh by inviting you to tell us a bit about yourself, your background, your drive to invent breakthrough technologies. And also, like, you know, if you can sneak in, how did you meet Ben? What's the origin of this name, Kwei? Like, okay, tell us everything.

SPEAKER_00 6:42

I was born in a um a place called Cape Town in South Africa, the tip of Africa. And um the I was I was born at a time when um a regime called apartheid, that was a governing um set of laws uh that that was a represented a dark time in the country's history. And um I come from a long line of I guess troublemakers, people that are um that resist oppression and um and and take a risk. Um my dad was a member of a teachers' union that became banned under that government, and so he had to leave the country, and he went to London to become a teacher in the East End of London. And my mom brought me and my two sisters out of South Africa to the UK, and hence you can tell from my accent, I still have the British accent, so I grew up there, uh, had my education, I studied physics at King's College in London. Um, but then quickly um moved into the computer science uh world, uh, even though my first job was leveraging my physics degree, the after that it was more computer science, but always close to the bare metal, close to the hardware, um, I was fascinated by the algorithms and the deep laws of um systems, and always tried to find the underlying um theory, as I did in you know, studying physics. Um so I was if if we if we tried to impart any wisdom to young um entrepreneurs and young inventors, um there was a there was a good um saying that I heard um someone give at a um uh at a commencement speech. Uh they s they were cataloguing um smartness and they were cataloguing luck. And he wraps it up by saying, so um the recipe for success is is not to just be smart, it's to be lucky. And and that's a difficult thing to repeat, but be lucky. Um I I was lucky, I was lucky to have been smart at the right time. There are lots of people that uh are smart, but that the the window is not right for them, and so when um the movie industry was becoming digitized, it bec it moved from editing movies on celluloid and moved from uh videotape to disk drives and computer processing of images, uh images and and audio and uh so on. That's where my expertise um took hold and and I happened to have been in the right time, the right place. So behind me you see this console, this uh peripheral, which was the which was the input device for a movie editor called Lightworks, and that's that's the the device that I still have on my shelf there. It reminds me of you know one of the first innovations that allowed us to edit a movie on a computer, a full-length feature film, and that uh that took Hollywood by storm, and uh it was on the strength of the success of that company that I moved my family from London to Los Angeles, and we've been here ever since. Um I raised a family of four boys with my wife. I'm still married to the same girl that I met when I was in high school, so and so she um uh she and I have had a very um fortunate life. We were also very lucky, um, and we've had four children and we raised them all up, and we were in the same house that we were that we bought when we came to Los Angeles. Um, and then I I had the choice of cool audio-video projects that I could work on, not only in the movie industry, but also in the biomedical space. Um my fondness of being close to the hardware meant that I I was uh uh able to work on the cochlear implant project for advanced bionics as a DSP engineer. And uh then their sister company, um, Second Sight, uh I became director of software in a project um to build a bionic eye, and um that was a successful, at least in first attempt at trying to restore sight to blind people um using a retinal implant. Gosh, um just at the right place at the right time, uh I was then invited to join a group of uh engineers in Seattle, um, and we formed what was now known as Widevine Technologies. It was acquired by Google 10 years later. It's the technology that secures the internet television that we watch, and I think 40% of over-the-top television secured by Wide Vine. Um, so that was a um uh another stroke of luck, I guess. Um, but a lot of the patents that were developed were developed while I was at Widevine, and it was basically Google uh acquired it for its patent portfolio. Um and so I realized that that was an important thing. You you should not only be developing product um for its own commercial success, but you should also try and encapsulate the innovation in defensible intellectual property that becomes an asset of the company, so that if the product was hey, too soon, which Wide Vine's technology was too soon in 2000, um then at least you've got the intellectual property as an asset that becomes saleable, and that's and that was really the the transaction to Google. Gosh, I can rant on like this because you know I I continued on uh built another company that also built up a patent portfolio, sold that to Google. Um, then I became um CTO at Real Networks and uh helped pivot that company from its streaming media roots to AI and then retired on the strength of um an exit there and um and then I thought, well, AI is heading in a dangerous direction and I need to do something about it. And so I formed Kwai. Kwai is a word that actually comes from the tip of Africa, its original Dutch meaning is angry or wild, but it's taken on a hip-hop connotation, which means cool, just like in the the word sick or ill uh means or wicked means cool. Um quai means cool. So we now have over a thousand Kwai volunteers that are trying to be cool in AI. Well, we're trying to make AI safer, faster, and greener. And how did I meet Ben? I cold called him, I sent him an email. I said we'd love to have you as a keynote speaker at our summit, and that was back in March, and he graciously accepted, and we became quick friends. Uh he's also got a physics background, and when any two physicists meet, they share their theory of everything, and we we we got so annoying to the rest of the people that we were um uh that that were in the audience, you know, people suggested we go get a room because it was just geeking out on stuff. But I'm I I feel highly aligned.

SPEAKER_01 15:40

Love at first sight, yes.

SPEAKER_00 15:42

I feel highly aligned with with um Ben's mission. And look, it's this isn't a an airy-fairy, idealistic um motivation uh to try and what broadly said democratize AI. It's been done before, um, and it was done by Linux. When Linux um came on the scene, it was an intervention against a monopolization of the server operating system space. And Linux was an altruistic intervention by an individual, Linus Torvelds, but he um he brought an open source, non-profit, volunteer-driven movement to the market, and now Linux is the dominant operating system. We really felt we were the rebel alliance trying to oppose the sort of a group of oligarchic monopolistic um entities that were controlling the server operating system space, and and now Linux is the dominant operating system, and the industry is healthier for it. And I thought AI needs the same sort of intervention. Well, I'm gonna take a slug of tea.

SPEAKER_01 16:58

Yes, and and actually, uh, you know, this is uh so inspiring for us, and it's kind of obvious that you had this uh mind mateship and soulmateship with Ben because you are so aligned, uh, both as rebels, uh, he calls himself an anarchist, and and in your mission to actually democratize uh this powerful technology. So you both have the hearts in the right place, and I cannot wait to see you together in San Francisco very soon. Uh, what I'd like to you know um ask you now in continuing our conversation, and thank you so much for you know for sharing with us uh uh your life uh here and um and learning, you know, and I'm sure our young audience uh about luck, but you know, translated like timing is everything and right time and right place, but also thinking ahead and patenting your innovations and inventions so you can later on capitalize on that if they are ahead of their time. So a lot of lessons here, and of course, the continuity with the family and all that. Uh, they all these lessons transpire from your story. But coming back to uh your uh QAI uh mission here, uh I think as we move closer to artificial general intelligence, the question may no longer uh be simply how intelligence machines become, but who will own that intelligence? So if uh QAI succeeds beyond your boldest expectations, what becomes uh the equivalent of Linux impact for AI for AGI, if you can spell out that analogy in terms of uh your success.

SPEAKER_00 18:49

So what yeah, I think it you know I ponder on what does success look like? Um, and so again, when we look at Linux, how did Linux um basically extract the the control of the server operating operating system space from a small group of companies like Microsoft and Sun Microsystems and IBM and and so on, and and spread that so that everyone could participate in this operating system and drive a class of technology forward. Um currently the entities, the players that can build uh AI services uh are only a few because it's really expensive to train a frontier model. So and and their models are so large and and unwieldy you can't run it locally on your machine. And that well, that is the current um thinking. And so we thought, well, let's let's test that thinking. Can we can we create an alternate? A digital public infrastructure where we bring AI down from the clouds to run at the edge, a distributed fabric at the edge that's stitched together with all of our computers, where we are all then participants in the infrastructure. We're not um condemned to become renters forever. And this is what um one of our keynote speakers, Rafi Krikorian, who's the was then the incoming CTO of Mozilla, he was the speaker um just before Ben spoke. And his talk was owners not renters. And it it echoed uh a book that had just come out by uh Yanis Varophakis, who was the um the the chief financial officer or the the uh the minister of finance for for Greece.

SPEAKER_01 21:08

Uh I love his boldness. Yes, Yannis is amazing.

SPEAKER_00 21:12

He'd um coined this phrase techno-feudalism.

SPEAKER_01 21:16

He is definitely one uh uh rebel.

SPEAKER_00 21:20

Exactly, and and um so it it appears that the oligarchs who control AI want to um amass all of the wisdom of humanity into their models, free of charge, they're vacuuming it up free of charge. We we are inadvertently donating that to them, and not just in training data, but in every single prompt that we send them and every document we send them. So um there's this massive wealth transfer that's happening. Right under our noses. It's it's going on. So if we look at what success would be for um Singularity net or QAI or any of the movements that are trying to resist this trend, it would be let's say in a few years from now, a decade from now, we have the Linux of AI. We have an alternate where we've got a fabric or an infrastructure, a digital public infrastructure that's stitched together with all of our computers, where each one of our machines, when we're not using it, when it's idle, in its idle cycle and still burning electricity, it's now available to participate in um in AI jobs or any compute job. And then each each of these nodes would have its own wallet and it would earn a passive income that it could spend on that same network. And so we're not burning tokens forever and and continuously becoming subscribers and renters of our own um intelligence. We are now part owners of a digital public infrastructure built by the people for the people. Gosh, I'm on a soapbox, I need to step down. Yeah. But does that make sense?

SPEAKER_01 23:31

No, absolutely. And um, you know, um you are so right that today's uh chatbots feel very personal because they learn everything about us. Yet the intelligence, memory, and incentives ultimately belong to someone else. And this is indeed the outrageous thing which you are underlining here, and Yannis Barovakis as well in his book and so on and so forth. And we are very well aware of that as well, at singularity net. So the question is, and you uh alluded a bit to the technologies which you are implying, and we will dive uh deeper into that. But you know, the question is when does an AI truly become my AI? Like um, you made some differences also. I um heard you say there's a difference between personal AI and personalized AI, you know, but I mean, when does it really, really become my AI? In case you want to talk about all that in the grand scheme of things and also from the perspective of uh of your technology.

SPEAKER_00 24:38

On my desk here, I have I have a calculator. Okay, so if I'm doing my tax return, I've got my personal finances in a in a separate file here. I've got you know scraps of receipts and so on. And let's say I were to use this um and and that I'm using this device, this calculator, in order to get a job done and to file my taxes. Um compare that to um an AI service that says, oh, just upload the spreadsheet of all your um of all your data and um we will we will prepare your tax return for you. You feel at some point that hey, I've gone and divulged a lot of personal information about my finances, and we we're doing that. So the the disambiguation of what is the tool and what is your personal data, that that line has been blurred. And so we we are we're continuously seeing the blurring of of those lines. It's at one of the conferences I was um asked a question, and uh um I I things became really clear to me um just in the moments when I was asked the question. And I said it's very important for us to disambiguate language from knowledge, and um smart AI, uh chat chat GPT, or even Claude um has amassed up all of human wisdom. Well, is it really wisdom? It's it's Reddit, it's TikTok, it's um you know everything they could find on the internet. It's um it's language, but it's not necessarily knowledge. Um, it could be counterfactual knowledge. And so a lot of the research that's going on within Kwai is trying to figure out how we disambiguate um knowledge from language, and we see knowledge as a way of um storing facts, um, and language is a way of transmitting those. When we think about it, how did it evolve in humans? The very um power of humans as a species that made us the dominant species on the planet was our ability to transmit knowledge that's in one mind to another person's mind, and so language becomes the codec of telepathy. Now, I saw that because of my history of working on video and audio codecs. I it bec it came to me in a flash that language is the codec of telepathy, and it is codecs are by their very nature lossy, some some can be lossless, but our human language is very lossy. So as I'm describing the thoughts and the knowledge that's in my head, and I'm transmitting that to you over a serialized string of words that then go through another codec before it gets decoded on your screen and your speakers and your headphone, and now it you form in your brain another knowledge graph. And so we're exploring the representations of knowledge in we start off with humans and and and biological systems as a great metaphor. Of course, we can construct any way of doing that in a computer, you know, with databases and and spreadsheets and so on, but I think um the as as with AI, the inspiration of looking at um biological systems and neural systems as the example of something that was successful and robust and evolved over a long period of time is a great is a great guiding principle. So we're getting back to basics at Qai and looking at knowledge representation, and then how can you transmit that in language, in natural language, such that the recipient forms the same knowledge graph uh in its in its mind uh later on. Um so I'm I'm ranting here, but I'm trying to answer your question about um personal and personalized.

SPEAKER_01 29:50

Yeah, you know what I wanted to actually um ask also, which is I think very related to that, because uh Quali is building what you call a personal AI operating system, and that's a fascinating concept. I mean, when people hear operating systems, of course they think Windows, Linux, Android, blah blah. You're absolutely right. Well, yeah, so yeah, so tell us in that context as well.

SPEAKER_00 30:17

Yeah, so if you if you um squint at it and say, well, the operating system is the layer of software that controls the resources of the computer, and operating systems have have many layers to them. If you look at the Android operating system, underneath that there's Linux, and then there are layers of operating system. So, what is the set of processes and protocols that would control a compute fabric like QI is building? Qi is building QINet, which is a distributed fabric of nodes, and how do we control the resources on that node? But do it in a decentralized way, where no one um manager, uh no centralized um system is managing access, is managing compute, is managing storage. It is all in a set, in a in a decentralized distributed um set of protocols. Each node is a peer. They're equal, um, they have equal control, they have control over their own system and how they um talk to other systems. And um so when you look at it like that, that layer of software, that set of protocols, um, in concert constitute an operating system. And so um I stand by it, even though a lot of people you know pick me up on it and said, Are you gonna reinvent Linux? And even even Linux got a bit put out when I started saying that. Um but I think they've come round because just this month um Kwai was invited to form a work group under the Linux Foundation. So we um we started at the beginning of the year contributing to the Linux Trust Over IP Foundation because they have thinking there that we didn't want to duplicate on how to put together a decentralized trust graph uh of peer nodes. How do we manage identity and authentication and proof of personhood since we want each node to be attributable to a single person, um, not anonymous bots that can run on this network? So we got to do that. Yeah, totally.

SPEAKER_01 33:03

You know what I wanted to say is like we've spent years discussing self-sovereign identity, and we are in the blockchain community here also in SingularityNet. So, you know, I'd like to take this uh now that you got here one step further. I mean, can we imagine self-sovereign intelligence? Like an intelligence that represents me, works in my interest, uh, you know, on my behalf, uh, but also remains entirely under my control, which I think is what your technology is uh offering.

SPEAKER_00 33:33

Yeah. So when when we started describing this peer-to-peer fabric, and we said, look, you don't have enough compute power on your desk um within within your home, you don't have enough storage capacity, you're going to need to leverage um external capacity. And today people are doing that in the cloud, they're renting storage from Google or from Apple, and they get their monthly bills from them. And um we said, well, you could leverage the spare capacity on other people's machines, and so everyone could be participating in this in this sort of multi-tenant storage fabric. Okay, and then the uh the critics with even within our organization said, I don't want my personal finances sitting on my neighbor's computer, even if it's just a shard of it, even if it's like a 100th shard, I don't want it just sitting there because it's it's my personal data, and if we're making a personal AI. So the first problem we had to solve was how do we have something stored remotely uh in an encrypted form, but make it still searchable in its encrypted form. And that body of technology is called homomorphic encryption. Now, you know that you know I worked on at White Vine Technologies, we worked on all sorts of cryptography, and homomorphic encryption was a a pipe dream at that time, but I've been tracking the progress in the literature over the years, and we kicked off a research project two years ago within Kwai, and uh we collaborated with the Society of Industrial Applied Mathematics, we collaborated with um uh Pomona College as well, and ended up with some partially homomorphic encryption algorithms, and these demonstrate that it's possible to have a vector database that has your personal information on it, but you scramble it and you can host it on an untrusted host, but it's still searchable in its scrambled form. You can send a query, but you've scrambled the query. So now the untrusted host, you're just renting their high-speed vector database. You're renting their infrastructure, you're not giving them the data. With Sam Altman today, you're giving him your data, you're uploading your data in clear. You can't say, Oh, well, can I upload my document zipped with a password? They'll say, No, you have to give us them the password. They have no mechanism to be able to receive your personal data scrambled such that it's still searchable in its scrambled form. And so we set ourselves that task, and in fact, um we we are so we've had success in creating algorithms, but we didn't think that we should be in the business of creating cryptographic algorithms. Instead, this should be a standards body. Uh, we should we should standardize a protocol and issue a call um for um uh a call uh an RFP uh in the same way that uh NIST did for AES. Um when I started off 30 years ago in the cryptographic space, it was just as DES was coming to the end of its life, and we started using triple des. And at the same time, AES had um the AES call for proposals had gone out, and there was a competition, and eventually the Rhindal algorithm won, and that's now our AES algorithm that the industry's been using. So we thought a similar approach needs to happen in order to secure your data, and that's the first step in personalizing, and not personalizing, but making it personal. Making your um AI personal is securing your personal data.

SPEAKER_01 38:08

Totally, totally. And you know, uh now that you mentioned Sam Altman and the approach of the centralized uh Behemoth labs, one uh criticism often raised is that these frontier AI labs um are requiring tens of billions of dollars for their game. Can an open source ecosystem realistically compete with organizations spending unprecedented amounts on compute? Uh what what what do you think your odds of success are in this uh race, if we call it like that?

SPEAKER_00 38:47

Right. So so so we handle the storage part and the and you know search, and then we we started tackling the compute part. Okay, so how can we how can we leverage our small our laptops and our Mac minis and uh you know and our PCs and so on? We might have a gaming PC with a GPU card, and how can we string all those together and compete with the big um data centers? It's really difficult, and I don't think a head-on um uh competition can you you'll ever beat them on their metrics. But here's where we can do better. First of all, a distributed fabric spreads the load across the electrical grid, it's not just drawing power from one the you know the power sources in one zip code, it's um it's distributed, and so that makes it more robust. Currently, we have a very fragile um infrastructure for AI. We've seen some outages last year that took down the entire um industry for a day or two, you know, when um this wasn't data center outages. Well, there was an AWS outage um last year, and then there was also uh an outage in the security stack, one of the components in the security stack that that's now become um uh you know commonplace in most enterprise systems that failed and uh it took down many systems. So um what do we do? It's our thesis that a distributed peer-to-peer fabric becomes more robust. But your question was about training. So we we we looked for what open source projects are working on distributed training, and there was this project called Petals. Um, it's now a ghost town project, uh, you know, like many uh GitHub repos, nobody's contributed to it for two years. But we took it two years ago and uh we started using that as a sandbox to see can we um run a uh run inference in a distributed fabric before we run you know before we do do any training, can we can we pivot that whole thing to just run inference where you've got a big model uh that can't fit in any of the computers, but you break the model up such that it can run um across multiple machines, and we've proved that that can happen, and that's part of Quineet. Um I think the the answer is not in building uh another competing frontier model. I think the answer is a mixture of smaller language models and knowledge bases. Currently there's a conflation in um in the language models like um Mythos and Opus and uh GPT, all the GPTs, a conflation of knowledge and language. And I think it's highly important for us to break the two apart. Um and once we do that, we can have the knowledge. Um now you've got you've got the opportunity to own your own knowledge and decide whether you're going to contribute your knowledge to the community fabric, so that the each of our nodes is a a repository of knowledge as well as a processor of language. And um understanding the role of knowledge and language is is super important. Um, it's not in the interests of Sam Altman for those two to be broken apart, because um as long as they're conflated and they large, uh they can only that the it's only feasible to run it in uh their data centers. But once we break the two apart, we can then have the opportunity of having the knowledge reside with the owner of that knowledge and give them the opportunity whether they're going to make that available to the computer to the community for free or for a fee. Um you know you might have people with certain expertise, and it's it's right for them to charge a fee to you to for you to be able to query their knowledge base. Okay, so now we've disambiguated knowledge and language. Um it turns out that the language models we can use can be a lot smaller. Llama 8B works just fine. Once you've done created that disambiguation, um I'm I'm I'm I'm making it sound a lot easier than than it actually is. There's a lot more um in we we're doing research in the area of uh uh um of uh you know subquadratic neural networks to make um uh the neural networks run faster. Uh we we've got um uh research efforts in the area of um symbolic uh reasoning uh networks that are not sort of the traditional uh um uh transformer-based neural networks. And so uh this I think these are the important things that are going to help reverse the trend, the monopolistic trend that is is going on at the moment.

SPEAKER_01 44:54

Absolutely, and you know, uh thank you so much for for giving us all these details. It's a completely different paradigm uh altogether, it's clear. I'd like to move uh slightly from the technical to the societal dimension and the implication of what you are bringing uh to the fore here. Uh today's, you know, alignment largely reflects, and I mean AGI or AI alignment, largely reflects the worldview of whoever trains the model. And uh what you are doing, I think, is opening also different perspective to alignment because open source personal AI can enable something much richer where values emerge from the individuals and communities rather than from a centralized authority and how they want to train the models or to restrict them for that matter. So uh the question is do you see alignment also? I mean, can it become decentralized from the perspective of uh representing uh diversities of communities? Speaking of also of your origins, right, and the conflicts that existed there and all that?

SPEAKER_00 46:08

You know, I I wish I had the answer to it. I I within Kwai, we've got a lot of deep thinkers in this space, and one of the um gratifying uh results of just forming a community of critical thinkers is just sometimes sitting back and just observing the conversations that go on. Uh, we've got one deep thinker, he's named Steve Vitke, and he talks about um what we need for personal AI is to have a digital twin. So we have ourselves, and then we have this trusted executive assistant that um understands our aspirational self. Which which I okay, so so I'm gonna paraphrase the the way he says it. And then it is like your you know your executive assistant, but it's also your uh chief of staff, and it manages all the other departments or it delegates to other departments, which are agents, agentic um systems, and um he says that so alignment is not a monolithic uh is this model aligned with human values? That is just an intractable problem to solve because whose human values you're talking about? All you can do is focus on the individual, and the individual saying, I want to make sure that this computer system is loyal to me as a loyal agent, it's not a double agent, in fact, working on behalf of a vendor, um, which you sometimes get the impression that you know Google is doing that. When you ask Google a question, you know, the first the first results that come up are sponsored results. So, hey, so you weren't actually acting on my best interests, you were you were giving me the advice of the the the highest bidders in your in your marketplace. Um so we want a loyal agent, one that's truly acting on our behalf and representing our best interest. I think that's the exciting part is actually engaging, having a digital twin that engages with you to figure out well what are your interests, what what is your what are your better aspirations, which is which is I don't know, computer systems don't don't do that, they they behave like you know inanimate tools like my calculator, and um it's it's not asking what you know what what does success look like for you. Imagine if a system did that, um your personal assistant, what does success look like for you? Um and it might be freeing up time, it might be it's and it's different for every person. And so now once you are able to engage with this ongoing conversation about what is success for you and what are your aspirations, you've now got a an agent, a digital twin that is aligned with that personal mission for you. And now if everyone is armed with that personal assistant that can give them a 10x improvement, rather than corporations you know always seeking their 10x improvement, if if we could each of us get a 10x improvement, imagine what that does for society. Imagine that what that does for the economy. Um, I will become a better employee. Well, I'm I'm not in the market for being an employee, but you know, if if people are freed from the the drudgery of their life of interacting with um health insurance companies, with all the services that they interact with, if you had a personal assistant, a digital twin that was able to engage with those things for you, it basically frees your mind to be able to do other stuff, innovation, um, wealth creation, um making your boss richer, uh making yourself maybe more successful, we hope.

SPEAKER_01 50:52

Yeah, so so you know, uh, this brings, first of all, I want to appreciate the fact that you are giving credit to your community, uh, and you know, it seems like it's very exciting there to belong to your community, a lot of ideas and exchanges and so on. So, but this one in particular, yes, with um better selves, our better selves, it's fascinating. And it leads me to this question. I mean, at what point do these uh personal AIs uh stop being assistants and become autonomous participants in the economy along aside ourselves? Uh, because being our better selves, I mean they work for us, but they can also, of course, do other things on our behalf. So uh would we need new social contracts uh in that case? Like when they will start to negotiate contracts, buy homes, free our time, you know, coordinate uh health care. In fact, you developed an AI CTO early on, as far as I as I know, maybe I'm wrong, but I think you are a pioneer in in automating uh this kind of yeah.

SPEAKER_00 52:03

So um again, in our community, there's this discussion going on, and some of the in partner organizations. Um, so this notion of um uh autonomous agents uh being trustworthy and remaining aligned with the their owner's mission. First of all, we don't want unattributable um anonymous bots to operate on our network. That's that's just a guiding principle. I I don't know uh if it's based on any it's it's just it's just a gut. So take that as a starting principle. And so what what we've said was that every agent that's running on our decentralized fabric is going to be attributable to a human that is ultimately legally liable for the behavior of that system. And um so are there is there is there a social contract? Well, there are there are uh like a myriad small small contracts that happen, smart contracts that get struck up. So my system says, okay, I've got an AI, I've got a job to run, I've got uh I and so it sends out um a request to the network, and that that process is called intent casting, and it comes from one of the gurus of our organization, a character called Doc Searles, who wrote this book called The Intention Economy, and he describes this thing called intent casting. So you put out um uh a call to say, Hey, I need um some processing of uh a hundred tokens a second, and I need um this amount of storage. Okay, so you've now got machines that respond, and very quickly, within sort of milliseconds, microseconds, um, you end up with um a negotiation and a contract is formed, and there is the job is done, and then there's a reconciliation and and wallets are credited and debited. So that's how it's intended to work. Okay, so there are lots of contracts that happen. There's not one massive uh contract that goes on, um it is lots of bots acting in their own self-interest or in the self-interest of their owner. Um, and we feel that that sort of network is the most robust, and uh you know, one hopes that you don't have to intervene and stop it from being gamed, but there are going to be forces that are gonna try and game the network and get undue um influence and power and access.

SPEAKER_01 55:10

Um what is your experience with your own network in that regard? I mean, it does the balance uh shift towards good or towards those who try to game the system?

SPEAKER_00 55:23

I've been uh so we we don't have the answer. We flirted with um with the whole crypto community. Um, last year we were deeply in con deep in conversations with um uh an Australian uh crypto company that says, hey, we we're a for-profit uh this crypto company, they're a for-profit entity running a distributed multi um tenant storage fabric. And I said, Oh, we want to build one of those. And they said, Well, we've got it already. We want to donate the code to you, and we want to donate our entire treasury of crypto coin to you worth about a million dollars, and that should become your utility currency. Um here's here's why we pulled out of that agreement. It's because the community's values, their community's values and our values were not aligned. And as we tried to um broker conversations between the communities, we just saw that that misalignment was um very distracting and and disruptive to the thing that we were trying to do. So um you can see that this decentralized movement is not just within that it didn't come as a rebellion against AI, it's it's been around for a while, and uh many decentralized communities in the crypto space have tried to wrestle with um uh people trying to gain undue influence. So it's not something that we need to go and re reinvent. We don't it's uh ourselves, we can learn from how other communities have have solved the problem or are trying to solve the problem and and stop um monopolistic trends. Um so I yeah.

SPEAKER_01 57:29

I I You know, as a as a cryptography pioneer, of course, uh you've been there during the cyberpunk movement and the honest uh movement, which actually we were hoping we will achieve when uh the blockchain appeared. And yet the crypto communities uh have captured that movement, and and unfortunately, yes, it's kind of uh I mean it's fainted now.

SPEAKER_00 57:55

Yeah. Well, do you remember even before um when crypto when cryptography was cool, it when cryptic cryptography was a revolutionary tool, um, in the days of PGP, I remember when um uh the the sort of critical thinkers within the industry were saying that's gonna be the big liberator of um of societies is is cryptography. And then it got subverted, it got taken over by big um wealthy interests to basically secure their revenue streams. Um Hollywood embraced it and and and frankly, you know, I uh what one one of the startups that I worked at, Wide Vine, uh, was one of the pioneers in that space. Um but um yeah, so so these tools that they they they could be used for good or for for evil, and um AI s certainly can be used to uplift humanity or further enslave it.

SPEAKER_01 59:10

Like any tools, of course, you know, and uh yes, I totally totally agree with you. Uh I just wonder if there is anything else like uh, you know, now ahead of the conference, uh, you want to share with us some final thoughts, some message, or maybe tell us what conversation aren't we having enough uh that we absolutely should. Uh uh now that you're speaking also about this, yeah, the uh more uh uh negative aspect of the decentralization is not all uh bed of roses, if I may. So human nature.

SPEAKER_00 59:48

Yes, yes. I'm gonna be a bit of a naysayer on the whole um consciousness debate. I think it's a bit clickbaity and distractive. Um the real the real clear and present danger is um is the massive wealth transfer that's going on under our noses, this tacit assumption that the oligarchs um can own everything they see. And it's a it's it's not anything new. This sort of land grab has happened every time a conqueror comes onto a new continent. You know, there's the the image of the um some uh folks landing on the island of Manhattan. You you know, there's this painting and the Native American um and and they're they're basically handing over some pots and pans and some trinkets, and the Native Americans don't know that what they've actually entering in here's a contract to buy the whole island of Manhattan. And there's a similar thing of this guy Jan van Rybick landing in at the tip of Africa and having that same conversation with the the local San and Khoi people there, and that land grab is going on, and the land grab is not of dirt and soil and geography, the land grab is of our own wisdom, our own personal data, our own intellectual property, and it's not just of individuals, it's of corporations as well, small to medium businesses. Their intellectual property, their data has been sucked up into these um large language models that are going to condemn us to be renters forever. And if we don't act, there's a window of opportunity to act, and we can't rely on regulation or government action to save us from the overreach of big tech. There's no cavalry coming, you are the cavalry, and so that would be my call to action to everybody. You've got to get up and do something, join an organization like Qai, doesn't have to be ours, but there's a window of opportunity to bend the arc of history and and and do the same thing that Linux did 30 years ago. It's been done before, we can do it again.

SPEAKER_01 1:02:27

Amazing, amazing. Indeed, you know, like history repeats itself with the conquerors and uh the oppressors, and you were born uh in an environment which is was reflective of that. And uh yes, let's hope we can change it when it comes to to AI, AGI, and our future. Because as Linux changed computing, uh shifting power from centralized control to a global community of builders, perhaps uh we can do this also uh for the world of AI, and quite, you know, I commend you for the project, for putting your heart in it uh at retirement. I mean, of course, you could just sit on a beach somewhere in the Bahamas, but you are doing this, and it's amazing. And I really we are so grateful to have you with us, uh Reza. Um and uh we look forward to continue the conversation and also to collaborate with you uh at our, you know, we will we will have some announcements at the conference, and I'm sure that we will be working closely together because we have the same mission.

SPEAKER_00 1:03:32

Michaela, thank you so much for being such a generous host.

SPEAKER_01 1:03:37

Thank you. See you soon.