AI Panel: decentralized AI from dream to reality by Danielle Tichner

Transcript

yes all right hi everyone this is a decentralized AI panel for you and I'm going to introduce my pretty famous panelist to you so I'll start with Ben GTO from um Singularity that everybody knows jial from Lo and Tom is doing so many things I don't know which is Network zero Network zero is like my funing that's what you want to you is there anything we can do with the mics organizers it's too strong that's cuz we have the Pas here I

think it does it does it sound good hello cool wait yeah it's okay yeah we have a PA so it's they did something they did they adjusted okay fine excellent so yeah um I want to start with a little story for you about the centralized a so I did a panel in Hong Kong I don't know if it was s 2017 or 2018 I did it for Rise you know web Summit and they asked me uh to cover actually cyber security and I decided that I wanted to look at the angle of how cyber security interact inters with AI because basically you

know there's a lot of AI used you know by by hackers and there's a lot of AI also used by cyber security firms to uh offer all kind of cyber Security Solutions so I I invited cyber security not famous cyber security firms and I also wanted to invite Ben as a ey expert to have this debate so we had a very good panel and it was a nice discussion going on and at a certain point Ben was making an argument about what could be done you know to protect sinks and how sinks could be carried and I I'm not sure if we did it in 2018 or 2019 but at this stage uh web Summit was still not crypto-friendly in the middle of the panel then goes and and and uh takes

over the whole thing and he says well you know you want to be safe you have to decentralize AI the other panelists were look were looking at him like what and he gave the whole pitch of Singularity back then then you know so here we're sitting now in a totally different environment after open AI came coming out and after so many uh things that happened in in in in this uh I don't know how do we call it geni on overlay domain so I want to start this panel by asking Ben to give us a little overview how it was in 2017 you know he's a p basically the Pioneer not basically he is the Pioneer

of decentralized Ai and crypto communities so take us back a little bit tell us what you had what you didn't have yeah sure sure and I think what the Hong Kong event you mentioned I think was 2018 and that was a it was a crypto bear Market starting then so B basically Bitcoin was dead and the the idea that crypto could be used to to make AI better for everyone was considered sort of nuts by the mainstream of the of the of the tech world but yeah if we I mean if we go back further so I mean AI has been around since the middle of the last century I've been working on AI I guess since I was a kid late 70s early 80s I remember my my wife dug up

recently a web page I put online in 1995 claiming I was going to run for president on the decentralization Party candidate no now I wasn't I wasn't 35 yet then so I couldn't run for US president now now that I'm I'm guess I'm still not old enough because you're supposed to be 90 to run for US president but but I'm they'll make an exception I'm I'm technically old enough now but I've realized that'll be a terrible job and I don't want to do it but the the point is back then the internet is decentralized right so when the internet came out anyone can put a server on the internet it can talk using tcpip it's wul insecure but at least anyone can do it and there's no Central controller so at that point it seemed clear to myself and and many others

really that you know you want AI to be decentralized in the same sense that the internet is the whole the whole Global AI Network should be controlled by everyone and no one anyone can put a machine online running some AI it talks to the other machines online running other bits of the AI and the overall AI of the global network is is emerging from different AIS that different people have put up there so I I tried to build some stuff like that in the early Arts using early versions of java it was just so slow and so expensive we even talked about doing decentralized money back then but it was also very slow and very expensive and didn't I mean given how

fast like visa and MasterCard cleared a transaction it didn't didn't seem to make sense so yeah you fast forward you to 2015 16 17 you have ethereum you have Smart contracts still too slow and expensive at at at at that point I mean in a way it still is now but way too slow and expensive at that point but you can start to do something right so we built Singularity net which is an agent agent system for uh allowing multiple AI agents owned to control by different people running on different machines written by different people to communicate and coordinate together without any Central controller and singular unet I think was the first to do this in a serious way but there's a

bunch of others that popped up around around the same time and you know actually we thought at that point being naive and and star like in in late 2017 we're like all right crypto is going to the Moon forever so we we will ride this bull market to the singularity and then you you use this to build smarter and smarter Ai and move toward current narrow AI toward artificial general intelligence then toward super intelligence of course Krypto has been more of a roller coaster coaster ride it's it's doing nicely now centralized AI has advanced very quickly using Venture

Capital money and money from Sovereign wealth funds and and all that we've had Chet GPT the whole generative AI Revolution but you know crypto is still doing decently AI is still fueled by open- source code so it's not a big mystery how to do the most Cutting Edge a AI stuff so I think we're in a position now where we can put put Cutting Edge AI algorithms both deep neural Nets and other AI techniques like logic systems evolutionary learning hyper vectors can take multiple AI techniques put them together in a decentralized network and I think we're in a position in the next couple years to make smarter stuff in the decentralized AI ecosystem than big Tech

has done and to roll it out at at scale and and and rapidly right and this is uh this is so looking back into the historical period you mentioned we had fetch. a and ocean protocol launching a little bit after Singularity net with with their own versions of decentralized AI agent system so one thing we've been working on in the last couple months is creating a new entity called the artificial super intelligence Alliance so the singularity Nets ax token Ocean's ocean token and and fetch.

I's F token they're all going to merge into a single ASI token and we we may end up merging other decentralized AI projects into the alliance as well yeah the goal is to pull pull together the decentralized AI Community you're merging the decentralized intelligence into one super intelligence but we we coming back to the Tok guys leave it a little bit we're coming back we're doing another round of to is that an offer already for us so actually very good primer so uh looking back at what you had and what you didn't have what you were missing in my humble opinion were tools and community so speaking about tools you guys floi are

doing Federated learning I mean we would like the audience would like to know a little bit more because I don't I don't think everybody heard already about your product so we would like to know a bit more and also like how you helping these guys move forward yeah Federate learning is a super super important uh ingredient of system yeah there's a lot lot of I mean I'm a researcher for the past past 10 years working um AS Global head of AI in Royal Bank of Canada and also a research fellow in Imperial College London where my main focus is on the AI research and especially Knowledge Graph natural ention processing and it's it was in 2022 when when I see F federative learning as an emerging technology in AI

where Google Apple has already been using it but uh the way they use it in traditional way is they still have to control the Central node right Google has still has to stand in the middle say okay I am the guy who who is going to distribute the resources into different nodes even though that yes it's privacy preserving in a way that A and B don't see each other but yeah I see everything I'm I'm a master so um that's why I want to bring blockchain in so there's no Central node we don't need a central node to actually facilitate everything every training leave all or way like we all the trainers actually forming a consensus to join together to to govern the training and if anyone's playing

evil that's slash mode right if if he's playing good then eventually he'll get rewards and then it's this incentivization mechan mechanism is also a way to ensure that okay even if we don't have like even if we have a headless type of training cohort we can still ensure like everyone is gener enough so that's basically the idea behind flock that's why we call ourself flock actually it's upon federa learning block flock so I uh yeah I casually do some stand up comedian myself thank you so I was looking you know a couple of weeks ago we had an argument about the whole subject because that's the subject that we discuss every day decentralized Ai and I went back to look up your white paper and I was like oh my God it's all about governance

there's so much about governance so that takes us to you Tom I wanted to ask you to share a little bit because we were together in zalo and there's a lot going on on in the EI Community could you share a little bit what is going on how it all came to exist and what do we expect from this yeah so um the ex effective acceleration movement was kind of uh birthed out of uh like this backlash of the collapse of EA and uh the rise of AI doomers so um I think it was what year is this now uh was it 20 early 2022 or 2023 I forget um owski went on the bank list podcast

and basically scared the out of everyone with this uh AI Doomer thesis and when I first saw that I was like well he's obviously just like Incorrect and everyone will dismiss him as a fool but then I started getting all these like emails from my different crypto groups and mailing lists and stuff like that and crypto people who are generally saying hey we need to decentralize everything and like you know no government we're saying oh my God I'm really scared about you know AGI now and you know should we get governments involved and should we like be petitioning our congressmen to like ban Ai and like so I was like oh like this is mimetically spreading this is dangerous right and so I wrote a whole article that was like you know the title

was um you know AGI isn't going to Doom much all it's the doomers that will Doom us uh or something like that um and like the the the thesis there is like if you follow that line of thinking and you want the government to get involved like what actually happens like is that the the government says uh hey you know only a few companies get a license to create AGI right and then you have a problem of uh alignments where it's not that uh agis are aligned to humans or not it's like which humans are the agis aligned to right because it's the ones who actually create these agis that are going to be controlling them at least in the early days right so I think that's a that's a massive problem get government involved and uh giving a few license to

control the only agis that exist right so I wrote a whole article about that it did it was rather popular in the in the eak community but you know really the central thesis there is like hey it's probably impossible to align agis to All Humans because humans can't even be aligned with all humans like we haven't figured that part out like we haven't aligned humans right but uh if every like set of humans has their own AI then like it kind kind of like levels the playing field for everyone right I mean few people are aligned with themselves exactly yeah you can yeah we're we're we're uh multifaceted uh beings like we were're in conflict with ourselves constantly so it's an impossible problem right um but like uh you know we have a

few rules which is like you know the golden rules like do unto others as you would have others do unto you right it's just which is basically a a non-aggression principle right and like that's really the the thing that we need as as beings right um so yeah I think it's super important that we get the technology to everyone so everyone can use it and it's not uh you know nerfed or censored by particular organizations and so I think this is why the decentralized AI movement is very important because hey if you're only plugged into chat GPT and you have to do like you can only query the things that they say that you you know you're allowed to query then that's not very that's not very good and it only gets worse from there right so

various there's lots of various techniques to make sure that the rest of us have uh access to the Raw on tap potential of AI and I think that's where the decentralized AI movement is incredibly important for the future give me one word on the community because the governance is important but you need to have a community to govern it so what can you tell me in the community I actually don't think that um I don't agree with governance um like I think everybody should just have their own Ai and then they just duke it duke it out with each other I don't think we should govern anything like that well on on that sense I think there there's a beautiful thing about AI because um um the the the metrics is actually the the is is deterministic like like for example the paper rewards all you need

right so you yes it can be a very super ASI or AGI right but the the we can just set up a very deterministic reward that's a number 99% that you need to reach which can be easily then used as a mechanism for example when it comes to governance so that's a beauty of AI which I think yeah it works in a way that you don't really have to have a very complex mechanism to do the governance if it's you know or oriented I mean if you're dealing with a very narrowly construed AI which is working toward a specified reward function one can perhaps will a governance mechanism around that but I mean the the nature of open-ended intelligence is is is such

that there there is no fixed reward function for general intelligence human intelligence or or super intelligence so I guess me mechanisms like that can be valuable but yet they're they're existing within a larger open-ended intelligent framework so this is this is the threat of centralized AI if you have a uh non-general super intelligence right it has a reward function that we don't like if it's the only player out there it can go and and cause Havoc right but if everyone has their own similarly power AI they can then those AIS can battle that AI right if one goes Rogue then you have tons of other AIS to battle it right that's that's the that's

why we need a lot of let's hope these battles are not what happens there's exactly but like like more likely they detect that one has gone Rogue like very early and they're like uh oh like let's you know let's turn that one off let's let me swap the word governance for coordination yeah so I think uh we're really terrible at that uh everything that I've seen in crypto with Dows and democ Democratic voting and all of that has turned into a total show and destroyed a ton of value so um like I think that I'm very bearish on that uh like I'm I'm more of like you know the individual should like run their own thing and make their own Cho crypto networks tend to try to do governance via vote buying bya one token one vote and the the pathologies of this

are clear de Democratic governance doesn't necessarily mean voting and I would say there are many open- Source communities that evolve in a rough consensus in a democratic way without trying to count votes and make make it precise in that manner yeah I I I like that that's that you're describing rough consensus which I think like like Fork ability and rough consensus is I think the like more free market way to do that right because like if you're in open source community and like half of the people who want to contribute to that Community don't agree with where it's going it gets forked and then there's another one that's like hey this this repo is this but in this particular way and this repo is that in

that particular way and and the rough consensus ends up in two different places right and if it's such a contentious thing that it's so important that you have to Fork the community that will happen and if it's not it will get resolved yeah thats crypto well that well that's that's the real decentralized governance networks right I mean the fact that you can vote what this network does or not and not saying it's of no value but it it's these V are easily gained in they complex I say is of negative value I've seen everything be destroyed by it it can be negative it can be positive but I I I think that the fact that the machines are owned by different people the software is open and if a bunch of people don't like what the network is doing they can just run

different software on their machines and form a different network like that's exactly that's really fundamental I mean that's that's like voting with your feet and that's that clearly is what works and that that's the power of of decentralized networks for sof the right to exit now I want to take it back to you Ben in the sense that having this discussion here what's next for you what do you need to basically continue this well we're in a fortunate position now because crypto is doing well AI is doing well we have a lot of cool software we have a great community of people who want to help us keep building this this software so I really think we're in a position now now where within the

singularity net and the and the superintelligence alliance ecosystem I think we're in the position now or during the next few years we're going to be able to build AI which is much smarter than what big Tech is is putting out and if you have so you have GPT 4 5 six if if we come up with something much smarter than GPT 6 running on the decentralized infrastructure because we're not just trying to jack up Transformer neural Nets but we're integrating them with logic engines evolutionary learning systems and other forms of of AI systems on our own substantial but decentralized server facilities I think you'll then have a situation where the the balance of power in the AI World shifts from a

few big tech companies to decentralized networks and certainly techniques for Federated learning will be part of this because you need Federate learning to train for example sizable neurom models on on on decentralized networks right and so this this will create a situation which is very Troublesome for the powers to be because if a government wants to stomp down on AI well okay they sort of can in their borders but the network doesn't live in anybody's anybody's borders right and they can't lock up the guy who owns the servers because it's not just on one guy's servers right so it become it doesn't mean governments and people can't regulate and guide what's going on but it's it's it's a bit more

elusive than something that's controlled by a single company and there there's a a quote that my friend the psychologist Alan comes gave at the beneficial general intelligence conference we had in Panama a couple months ago he said don't worry nothing is under control right yeah I think and that's let let's let's hope it stays that way then it can self-organized into into a beneficial confi configuration so really what we need is just more and more Community more more developers to jump on board and help build the decentralized AI more people to design applications using decentralized AGI and AI tools and make noise about these applications that's

called for Community guys G can you support on your on your yeah yeah of course thanks as B and Tom say right that Moto has to be Community owned that's exactly what we are trying to build the product that we are launching soon about decentralized training so basically everyone who hold either data or graphic computes or or or a task right they they can just join the training and then bunch of them if they don't agree with any other people they can have their own motor wrong um they don't have I mean yeah they don't have to always you know have a a consens as ideology about AGI because it's not going to happen like there are different thinkings different School of philosophies in different areas like we want them to all control their own uh idea right that's exactly the thing and

also because of f learning which is very important that helps us to preserve user privacy because I also know another saying that hum human beings real authentic data died in 2016 because that's the time when gbd1 came out and there are so many synthetic data generated and now like everyone's using gbt for for example when they send the emails to do the proof reading right that's not your created data that's that's synthetic like like and then models TR on those synthetic data it's like amplifying any bias that machines might have so it's it's dangerous that's why we want to that last mile of a human real Dignity of your real uh uh uh private data right which can be securely run on your own control under your own

ownership that that's the real like AGI that we are looking forward in the future thank you now Tom I need you to put your we have five minutes left and I need you to put your investor hat and give us a little bit uh your sentiment or whatever you see for the AI tokens well then opened with the EV there the merger I mean what do you see in AI token world we see a lot of volatility we see Eric wall and everybody else complaining about yeah token Behavior what's your take yeah so I think it's very early days and I think we're in that stage of like uh 2017 kind of icos where everyone was creating these like so-called useless utility tokens uh not

to say they're all useless right now but I think I think we're in a very early phase where we're not not exactly sure uh how tokens will work within uh AI ecosystems in an ideal way um and so uh some projects may have it more figured out than others uh some of them are just considered to be like AI meme tokens right um and what I've seen in crypto is that uh you know the best case for that is like you're like like a meme token doing well for a particular technology like ends up like funding public goods um and that doesn't always happen to be the case but if it you know high integrity people are the ones that are working on

the project then that uh tends to happen which I think is a net benefit um and then you also get like folks who are just kind of uh riding the wave and they kind of just like you know do an exit scam or something like that but um yeah so I think there's um uh you know a lot of uh as we figure out what systems work it's it's really difficult to design a token without knowing how the the system is going to work in the end and we're still figuring out how the systems are going to work and once we figure out how the systems will work then we'll understand how the token um like actually uh helps the system to work right in the case of Bitcoin the Bitcoin asset actually is the uh is the security

of the entire Bitcoin chain it's the reason that it works right um and I think I think we'll we'll get there and you know the ASI team uh uh teams have made some good strides in that direction um so I think we'll I think we'll get there and like on the way we'll we'll fund a bunch of good stuff and you know we'll fund a bunch of scammy stuff um that seems to be like the way of crypto I mean evolution is uh messy wasteful and often absurd right exactly that's what we're seeing here so our take don't worry n is under control exactly yeah well thank you guys very much great perspectives thank you audience for having us thank you cheers organizers