The following is a rough transcript which has not been revised by The Jim Rutt Show or by Trent McConaghy. Please check with us before using any quotations from this transcript. Thank you.
Jim Rutt: Howdy! This is Jim Rutt, and this is The Jim Rutt Show.
Jim Rutt: Today’s guest is Trent McConaghy.
Trent McConaghy: Hi, Jim, great to be here.
Jim Rutt: I’ve been fortunate enough to know and I have worked with Trent since when was it? 2001?
Trent McConaghy: Something like that, yeah.
Jim Rutt: Trent, along with a group of his friends, started a computer chip design software company called Analog Design Automation. Somehow I got connected up with them and liked what I saw both the guys and the idea. And I became an investor and the non-executive chairman of the company.
Jim Rutt: The company was successful and was sold to Synopsys, 2004, 2005, something like that? Trent’s next venture was Solido Design Automation, another computer chip design software company. This time working on a different problem in the chip design space and in this case, a manufacturer ability.
Jim Rutt: Again, Trent was the CTO and again the company was successfully sold to an acquirer. Siemens this time. Two very hardcore startups. Two successful exits. Trent you could have done anything you wanted. You decided to do your next venture in the domains of blockchains, public ledgers, smart contracts, decentralized autonomous organizations and such. Why?
Trent McConaghy: That’s a great question. I think overall I’ve always been attracted to technology where I see that they’re going to be a very big leverage point. And with my first two startups, it was certainly in the domain of AI which is a very big lever. Even back in the 90s, some people saw that I was fortunate enough to see it and started applying it.
Trent McConaghy: Even with the second startup it was AI to really help drive more as long which of course can influence the world in a really positive way. So in about 2013, 2014, I really started to see the potentials for blockchain. And not just the Bitcoin side, the money side which I’ve been tracking for a while but what else you can do.
Trent McConaghy: And that led me down a path of playing with the technology and talking with more and more people. Eventually it captured my imagination so much that myself and a team of people around me, we started building something that we thought could make a big difference. And never looked back.
Trent McConaghy: Interestingly in the last two or three years, I managed to close the loop and pull AI back into the loop. So it’s AI and blockchain now. But yeah, it’s overall it’s really about working with technologies that can have a big lever to help improve society and humanity.
Jim Rutt: Exactly. Because I know you’ve always been interested not just in the technical problem, but what kind of impact it could have on the world. What impact do you see from this collection of technologies for literally the future of humanity opening up big picture?
Trent McConaghy: For sure. My pleasure. So in a nutshell, I see that blockchains are incentive machines. So you can use them to help to guide people’s behavior towards things that can be beneficial [inaudible 00:02:57]. And it may be just as a bit background, you unpack that a bit.
Trent McConaghy: Let’s think about what? Bitcoin for a second. Bitcoin, there’s a whole bunch of people running miners, miners, miners, and every 10 minutes someone gets some Bitcoin tokens, 12.5 Bitcoin tokens. And their chance of getting it is proportional to how much security they’re adding to the Bitcoin network.
Trent McConaghy: So in a sense, Bitcoin itself has an objective function. It’s trying to maximize its security, which is defined as hash rate. And so upon realizing that that basically at the heart of Bitcoin, it’s essentially an optimizer that’s getting humans to contribute to help it achieve its objective function.
Trent McConaghy: After realizing that, that’s what led me to asking what else can you do? What other objective functions can you come up with and codify into blockchain systems that are really acting as public utility networks. Like the internet and the World Wide Web and all these, there’s civilization infrastructure.
Trent McConaghy: And so that’s really, the most powerful way to think about blockchain technologies. And maybe just as another complimentary framing, traditionally to get incentives, there’s a wonderful book called How Nations Fail. And it talks about, it starts with the values of the politicians and the leaders, which they manifest in economic institutions.
Trent McConaghy: And those economic institutions are then used to infuse objective functions at the level of individuals and so on. So these institutions are laws and governments and otherwise. And then the final level, it’s really about making it possible for people to start companies and build and so on.
Trent McConaghy: So traditionally this has been manifested via governments and I’m all for governments that are well run and are not corrupt and all that. But what’s cool is there’s now another way to do it, where you can manifest economic institutions, i.e., blockchain technologies as public infrastructure.
Trent McConaghy: And then these will have their own incentives baked in that can help guide the society forward in a positive way.
Jim Rutt: Of course we’re seeing that nation state type governments also do want to have a say at least to some degree about what happens in this space. How do you see that working out?
Trent McConaghy: Well, I think if you think of what the idea of government itself it’s really supposed to be, by the people for the people. And if you think about even the founding of the USA where you have, Ben Franklin and Jefferson and Adams sitting around table designing the constitution and the other documents.
Trent McConaghy: They were really thinking about, how do we come up with a mechanism that can help to create a society where things are positive and can be building forward and so on. Like in some of your other podcasts so I’m describing I think it was with Jordan. Some democracies in there, some liberalism, some capitalism and all that stitched together in a new way.
Trent McConaghy: So governments can provide a set of rules by which once everyone understands the rules they can interact with according to that. So it doesn’t need to get reinvented, reinvented. Another rule everyone’s play especially after income tax was introduced in a big way, is to take in a lot of income and then use this to give a social safety net.
Trent McConaghy: So things like Medicare in Canada or even road infrastructure and so on. So we can think of these as a common shared resource pools, i.e., commons. So the governments have played a role for a long, long time and helping to give commons to the public and organizing all of that.
Trent McConaghy: So all the different institutions are in a sense the commons and that’s a role that the government has played. I think that’s still useful going forward. But I see that what blockchains can do, bit by bit is provide new ways for providing commons, sometimes taking in resources and providing that to the people.
Trent McConaghy: And in some places where traditionally the only way to do it was with an institution that might’ve had a 100 or a 1000 or 10,000 individuals to organize these commons. Now you can have it where it’s essentially this autonomous robot, this network, that has stayed to do it instead. And so you can actually make society a lot more efficient.
Trent McConaghy: So what blockchains are in my view, it’s not against the government it’s actually beneficial in achieving complete renderings just like government have traditionally in for. But doing it in a way that is much lower friction, much lower cost. And can help humanity just to run in a more, not only more efficient way but can help steer us towards self actualization in the end.
Jim Rutt: Yeah. We may also be able to have our nation state governments learn to be good users of these tools that strikes me.
Trent McConaghy: For sure. Yeah. And some have come along farther than others. In a sense right now, some are learning faster than others and understanding the benefits sooner and seeing how they can help their citizens. There’s a lot of used cases out there, not just related to value of transfer, remittance so that sort of thing.
Trent McConaghy: But also, where we started out was in the domain of IP and helping artists, creators of digital art and so on helping them to get compensated. And a lot of these things are too expensive for the government to do a good job just because it costs too much. But if you actually have these blockchain systems, then it can help rule out things like that, that are beneficial to all.
Trent McConaghy: So more transparent government if governments truly want that, then it’s very straight forward to do. It’s pretty rare for that to happen though, unless you have a government with essentially zero corruption or negligible corruption. Even if you have a tiny fraction of corruption, it’s very, very hard to instill transparency into a government.
Jim Rutt: Yeah. The government’s made out one transaction but I’ll be, tell you the citizens sure the shit do, right? And so if we can educate our citizenry that it’s possible to wire up many parts of government to be radically and unhackably transparent. While the government people themselves may not want that, it may well be the basis for a political movement at some point.
Trent McConaghy: For sure and the way that I see, technologies these general purpose technologies like AI or blockchain, you can’t help but have … they almost by definition have an ethical element. That of course then relates to how governments have to think about things and so on too.
Trent McConaghy: So every line of code almost that you write in AI or in blockchain might have an ethical judgment tenant, right? Even with your days with DNS, there was ethical judgment calls that got made that led to these days how I can work, et cetera, right? But back in the day there was just judgment calls.
Jim Rutt: Yep. Go look at Lawrence Lessex famous book Code, right? That in some sense code is the law of the internet. And your very good point that every time anyone creates, especially fundamental code for networks, AI or this new public ledger platform technology, we are essentially making lawful distinctions which are important. And it would certainly be good it seems to me if technologist became better educated about ethics.
Trent McConaghy: Absolutely. And I often hear that there’s rhetoric that they aren’t but, I studied engineering in Canada and there was a class that was part of the engineering curriculum, which is Engineering Ethics. Every engineer had to take it. If you graduated as an engineer in Canada, you have, you get an iron ring.
Trent McConaghy: This ring that you wear around your pinkie, it used to be iron now it’s made out of stainless steel probably cause it’s less damaging to your skin on your finger. But, there’s actually a tradition in engineering which is a big chunk of technology professionals for ethics.
Trent McConaghy: And I know that that’s had a profound influence on me. Going back to my days of engineering and then also studying a lot of philosophy and psychology and whatnot. So it can just be in the curriculum and of course in the fields where things are profoundly influential like AI and blockchain.
Trent McConaghy: In my experience, a good chunk of those people do think about these a lot, right? So it’s not, in any either’s, all issues that are popping up and ones that have people had been aware of for a long time and people are trying to do something about it. Same thing with blockchain, right?
Trent McConaghy: Sometimes I spend time just trying to raise awareness of new things that we hadn’t thought of before too, that we should try to account for if we can.
Jim Rutt: Yeah, I think you’re right that many practitioners are very sensitive to ethics. Of course the problem is it doesn’t take very many who aren’t to cause a problem. So we shall see. Let’s move on to another topic. Trent knows I’m always a zealot about performance and throughput in software.
Jim Rutt: How many times did I pound the table at ADA and said, “God damn it, this isn’t fast enough.” Right? And then I constantly have bitched about the ridiculously low transaction rates of most of today’s blockchain systems. I hope some of that nudging, it may have pushed you towards one of your first big scientific discoveries in the blockchain space.
Jim Rutt: Which was the DCS Triangle. Where you showed that blockchains can’t have all three of decentralization, consensus and scale. Tell us a little bit about what drove you to that thinking and, try to explain to our smart but not necessarily blockchain literate, audience what the DCS triangle means.
Trent McConaghy: In general it points to an engineering trade-off. I should note a fundamental trade off is simply an engineering trade-off among these three characteristics that you want. And how I arrived at it was looking around at the time, it was about 2015 maybe early 2016.
Trent McConaghy: And we hadn’t built our initial system this, IP and the blockchain system on Bitcoin under the hood. And we ran into major issues of throughput in scale. I think I gave a talk on it in fall of 2014 long before, most others noticed because we were just so early.
Trent McConaghy: And so that led us to be building another blockchain system, which essentially is decentralizing MongoDB we call it the GDB. And it leveraged to really be worked at in be done to make MongoDB scale. And we actually initially had built on RethinkDB which also has a lot of scale.
Trent McConaghy: RethinkDB and MongoDB these are all distributed database systems, which means that they get their performance their throughput all this, by spreading the compute load and the storage load across many, many, many cheap servers. They still have a single point of control so they are centralized but their resources are spread.
Trent McConaghy: And that’s really the difference between distributed and decentralized. Distributed means it’s spreading the resources on many machines. Decentralized means no single point of failure. So we will pick GDB, improved and improved over a number of years and it was targeted as a scalable blockchain database.
Trent McConaghy: And to this day, it’s running in a very nice piece of technology I’m very proud of it. And what my team and I had rolled out. And some people would complain though because they said, “Okay, well that’s scalable we see that, great.” It’s consistent, which means that the transactions that go through are reliably there.
Trent McConaghy: You don’t have an issue with the double spend problem. Double spend basically is the idea, if I have a Bitcoin and I send that Bitcoin to you and I send it also to my wife, then the system should say, “No, no, no, you can’t do that.” Or at least the Bitcoin only ends up out of your wallet or my wife’s wallet.
Trent McConaghy: And that’s the double spend problem. And that’s a very nice heuristic to see whether or not you have enough consistency in a pragmatic way. So we’ve talked about scale, basically can you have something that can serve the capabilities of the planet, and that can go to a scale of Google or Facebook?
Trent McConaghy: Can you be consistent? How do you prevent double spend? And the final thing is, can you be decentralized? And in the sense of decentralized, the definition I gave before, no single point of failure. But you can put this on, you can have more extreme versions and less extreme versions.
Trent McConaghy: The more extreme version and the ideal version is where it is permissionless. And what that means is anyone running a node that could potentially say that the transactions are going through or not, they don’t need to have any information to run the node. They just took into the network and they’re running.
Trent McConaghy: And that’s a permissionless system. You also want it to be censorship resistant because if you’re not censorship resistant, then along comes sadly acting, governments, et cetera. That will try to block things. So that’s the ideal is that you’re permissionless and censorship resistant.
Trent McConaghy: But I stepped along the way on this access you can start off with centralized control. And a step along the way is where you have a federated set up where there’s maybe 10 or 20, essentially authority nodes. So together it’s a group of factors that together are decentralized there’s no single point of failure.
Trent McConaghy: But there’s only 10 or 20 of them. So to compromise them you need, if one or two or a handful are compromised then you’re still okay. But if you’ve got more than half compromised, you’re in rough shape. And depending on the algorithm maybe it’s one third, maybe it’s two thirds and so on.
Trent McConaghy: So that’s the idea of [inaudible 00:14:56]. In Big GDB, we actually compromised on the decentralized side. We just had a federation of 20 nodes, 50 nodes. So we had scale and we had consistency preventing double spend but we were not permissionless.
Trent McConaghy: We would have loved to be permissionless, the technology just did not exist at the time. And there was other systems though Bitcoin and Ethereum, they were fully permissionless. So you didn’t need permission to be running a validator node. And they were consistent. They solved the double spend problem, but they were not scalable, right?
Trent McConaghy: Bitcoin was topping out at 10 transactions per second and the each transaction was costing 10 cents. So that, was not putting it to global scale Ethereum similar issues a little bit better but not a lot better. And then finally on a third example is, inside IPFS, which is this modern file system meets the World Wide Web.
Trent McConaghy: The World Wide Web protocols are really protocols for the giant document in the sky. Or you can also think of it as, layer on top of the core TCP IP protocols that allow you to download documents, see documents and so on. But if a link is removed, then you don’t know and you just end up with dead links over time.
Trent McConaghy: And if the data and the link has changed, then you don’t know if it’s been changed or not. So someone could be changing the information under your feet and you wouldn’t know, there’s no integrity checks built in. And so there’s the system called IPFS, interplanetary file system. That has basically taken these ideas from the World Wide Web Protocol, linked in new ideas with the dual signatures and integrity checking.
Trent McConaghy: And basically create a new protocols also for serving up data in the same way as the web, but where you just have much better integrity guarantees. So it turns out that under the hood it’s very, very scalable. Way more scalable than Bitcoin or Ethereum to the level of scale of the whole internet, which is very nice.
Trent McConaghy: It also is permissionless. You can be running IPFS technology without asking permission as part of the overall IPFS network. But when it doesn’t solve is, it does not solve the double spend problem. It has some degree of consistency using, pretty new cool technology called CRDTs. Conflict-free resolution data types, or the CRDTs.
Trent McConaghy: And overall though, it doesn’t prevent double spend. So here we have three different technologies, Bitcoin and Ethereum on one side, Bin GDB to be in another side, IPFS is on another side and each one of those technologies only solves two of the three.
Trent McConaghy: And that’s the DCS triangle that I observed. And what’s cool is in the, one year after that, one and a half years, there’s at least two other people that independently discovered it. And then realized that I had written about it before and then started referencing my work, which is pretty cool.
Trent McConaghy: It is not a fundamental law though. So it turns out that it was an engineering observation. Trade-offs exist all the time, but sometimes if you’re sneaky [inaudible 00:17:35] you can bypass all of the issues. And so what we’ve seen is in the last couple of years there’s been a lot of proposals to scale in various ways, and now these proposals are starting to get deployed as technologies.
Trent McConaghy: And the networks are just beginning start to get out there and will be maturing in coming years. So [inaudible 00:17:52] Ethereum as a technology that has consistency. And fully permissionless decentralization with scale as well as many others. [inaudible 00:18:01], Definity, Algorand all these.
Trent McConaghy: So that’s quite exciting. And I even wrote him with that in my initial article that it was just an engineering observation. And in the end it’s getting solved which is great.
Jim Rutt: Are we sure it’s going to get solved?
Trent McConaghy: No.
Jim Rutt: I’ve been hearing about these alleged speedups of the radically trustless networks for years now and nothing ever happens, they maybe get a factor or two.
Trent McConaghy: Well, actually I’m hopeful and here’s why. So there’s a few different tricks you can do. One of them is simply if you want to have consistency with traditional approaches to technology, use algorithms in the vein of BFT Byzantine fault tolerance. Which means you’re can have, a good chunk of bad actors, Byzantine actors attacking things in your silo K.
Trent McConaghy: And algorithms that follow these BFT protocols, they scale very badly with a number of actors. They can only go up to about a 100, 125 actors because the bandwidth balloons, it’s the bandwidth usage is a square of the number of actors. So then that makes you limited to 100 actors in a federation or whatever.
Trent McConaghy: But one of the tricks that people are using is basically running beacons. So maybe you have 10,000 possible validators, but every say 24 hours it randomly picks a 100 of them. And that’s basically enough time that, they won’t have time to collude form a cartel or something.
Trent McConaghy: But you still have, you overcome this engineering limitation of the bandwidth exploding. So that’s one very nice example. Another example of basically is to shard and have lots and lots of blockchains that can talk to each other. And probably the most two prominent projects that direction would be, Cosmos and Polkadot.
Trent McConaghy: And you can have basically various change. Each chain is dedicated to a particular niche or space. So these are two example technologies that are coming along nicely and they do have the benchmarks to back them up. They’re starting to roll out and I’m hopeful for it.
Jim Rutt: That’s good. Though I must say, I still wonder whether trustlessness is worth the cost, right? It’s almost a religious issue with a lot of people in the blockchain world that we have to have this radical trustlessness, uncensor ability, et cetera. And in the current systems results in crazy costs.
Jim Rutt: Bitcoin network cost something on the order of 5% of its market cap to operate each year. By the time you count, mining and transactions that’s nuts. Even something as grossly inefficient as the US banking system gets by on something less than 1%.
Trent McConaghy: Well I actually think, so Bitcoin was the first and it actually addressed the problem of solving the double spend problem in a completely new way. It was really truly an advanced in computer science sense, but it was grossly inefficient. It remains grossly inefficient.
Trent McConaghy: But network effects have taken over and people are very hesitant to change it in any real way because it opens up a whole can of worms. However, these other new technologies I’m talking about, they don’t use massive electricity costs. They don’t use massive resources.
Trent McConaghy: There were getting to within the same order of magnitude of cost of running as traditional distributed systems. So that is very hopeful. You don’t have to do massive work bringing electricity to do that, you have, instead you do things like proof of stake where you’re staking that you’re going to do X, Y or Z.
Trent McConaghy: And if you don’t follow through with that, then people can take your stake slash your stake. So it is actually quite promising and hopeful, that way and I think all of this will come to bear in the next two or three years. And these technologies, some of these networks are rolling on as we speak.
Jim Rutt: That would be a good thing. I have remained questioning about where the world has gone with this relentless, a libido for trustlessness. But if we can break the DCS Triangle, then that’d be great to have. Trustlessness would be great to have if it weren’t too expensive.
Trent McConaghy: I think there’s room for shades of gray. There are places where it makes sense to have a federated chain. Internal for example, internal data sharing within an enterprise. And maybe it’s a multinational enterprise and they just want to have, they didn’t want to lose all the data if there’s just one hack in one place, right?
Trent McConaghy: So if they’re, each office is running a couple of nodes, then maybe you have to actually take two or three offices to really get to it. Whereas right now all you need to do is compromise one place, right? I like to think of it as a security model as the following.
Trent McConaghy: The traditional security model is Eminem. So you’ve got this, a yummy chocolate on the inside and a hard shell. But if you managed to pierce any bit of that shell, then you get to get in and eat all the yummy chocolate. And blockchain flips that on its head, where it now you’ve got basically got this super rock hard nut at the center, which is simply the list of transactions.
Trent McConaghy: And then around it you’ve got, lots of young chocolate that’s maybe hard or maybe not things like the exchanges and different things. And they can get hacked and now and then, but that doesn’t compromise the integrity of the core system. But it’s really, really hard to compromise the integrity of the core system because there can be, there’s 10,000 copies of the ledger all around the world, right?
Trent McConaghy: So that’s very useful. I also see the going back I mentioned these federations internally, you can have shades of gray, right? Some chains make sense to have KYC hanging around. Other chains make sense to be fully trustless. And the trustless is actually in a unique example where this, one wonderful essay about social scalability, getting past Dunbar’s number, which I know you would like to talk about. Have 150 people.
Trent McConaghy: So how do you coordinate humans on a much larger scale without resorting to massive hierarchies and whatnot. And if you think of what, Bitcoin has managed to convince tens of thousands of people to work together towards a common objective function. Simply by virtue of aligning the interests of Bitcoin, this objective function.
Trent McConaghy: And the same thing with the work we’re doing in Ocean, right? Ocean is trying to maximize the supply of data that is relevant. And there’s many, many more like this. So it’s all shades of gray.
Jim Rutt: Yeah, that’s good. That’s really good. I love those thoughts actually, the idea that in some sense blockchain reverses the security model, and leaves you at the core with as far as we and unhackable actual database. While the stuff above may be soft, the reverse of the traditional corporate security model.
Jim Rutt: I’m going to move on to the next topic and again, try to keep the exposition as non technical jargon free and acronym free as you can. Something that people are interested in but don’t understand in the slightest is, what are smart contracts? To my mind, it’s perhaps the next most powerful concept beyond the blockchain itself, that has happened in the blockchain epoch. Could you tell us about uh, what smart contracts are and why they are important?
Trent McConaghy: Sure. So it’s simply a code that is running on decentralized substrate. So traditionally, if you go to a website, you’re going to be interacting with it. Say amazon.com and that code is running on Amazon servers. And if Jeff Bezos pulls the plug, then you can’t talk to amazon.com anymore, right?
Trent McConaghy: So there’s a single point of control or with banking too, right? There’s code that runs for your local bank and that particular code even knows what your bank account balance is. And if the bank goes broke, well maybe you’ve got some trouble or if there’s a hack, there’s trouble and so on.
Trent McConaghy: So this is just traditional code running on traditional centralized systems. What smart contracts are, is code that is running on this decentralized substrate. And once again, think of it like a public utility that’s just out there. Just like we have the public utility of electricity and roads.
Trent McConaghy: And even, going backwards more air as a public utility. The soil we stand on, all of these things are public utilities. And now though we have a public utility where you can run code and you don’t really care where it’s running on, and it’s just there. So you can put some code out there it’s running and you can, rely on it running to execute things.
Trent McConaghy: So and that’s basically smart contracts. So you can write this code, call it a smart contract, deploy to Ethereum or some other network and it can be doing things for you. So what are some of the things? Well as a baseline, it can store a balance or some tokens you have, maybe a token that you have that represents some Bitcoin or Ether tokens or Ocean tokens.
Trent McConaghy: You can have inside the smart contracts you control it. You can be having access to datasets or IP or otherwise. But then you can start building on top of it too. So there’s this whole, and something that’s near term that’s emerging as decentralized finance.
Trent McConaghy: So now you can have peer-to-peer loans where some people are putting up some of their holdings as collateral other people are borrowing that. So there’s basically borrowers and lenders. And the borrowers are getting interest to the lenders are, sorry the borrowers are paying interest the lenders are getting interest.
Trent McConaghy: And that actually takes over what traditionally has been a function inside banks. But now it’s running on this decentralized substrate. That is way more secure and has way lower costs of running compared to, all the humans and the electricity and all the other code and these traditional banks. So it’s much more socially scalable that way.
Trent McConaghy: And it started with this very simple baseline in the decentralized finance space. The decentralized baseline of loans and all this, but now they’re starting to be more and more extensions that are going to get more and more interesting and exotic over time. With things like insurance, things like collateralized debt in fancier ways and, we’ll see the emergence of mortgages around this and everything else, right?
Trent McConaghy: So from a consumer perspective, it can be as simple as, okay, now instead of having 0% or 1% interest in my bank account, you could have 5% or 15% depending on what the collateral is that you’re loaning out. Or whatever it is you’re storing, right? But it’s going to get a lot more interesting, continue to get interesting in coming months and years. And that’s just one example.
Trent McConaghy: But beyond, I really see that these smart contracts, just there’s the thing of smart using codes software is eating the world. Well, now the software that is eating the world is software that is running on this public utility infrastructure. And you can get it to do amazing stuff.
Trent McConaghy: So, what excites me is things like AI software running on this decentralized substrate. It completely changes the rules of interacting with an AI. It’s AIs that, they can be their own agents, they can have their own resources, they can own stuff. They can accumulate resources. And you can really run with this.
Jim Rutt: Yeah. As you know my friend Ben Gursel was on The Jim Rutt Show six or seven weeks ago. We talked quite a bit about his Singularity net project, which tries to do just that. So the next part of educating our audience a little bit before we jump into your Ocean Protocol.
Jim Rutt: It’d be real helpful if you could describe it about a similar level of abstraction. The concept of a decentralized autonomous organization, sometimes known as the DAO.
Trent McConaghy: I guess some of your crowd or maybe a fair bit is people from the complexity science space. After your introduction, I had a call with David Krakauer who runs a Santa Fe Institute. And he asked me, tell me what is the hook for DAOs or blockchain for me? And he comes from the artificial life space.
Trent McConaghy: So I said, it’s a life form and it’s true, right? DAOs are a new form of life. And it wasn’t me to first conceive of that. To my knowledge, the first time that I saw it was with Ralph Merkle talking about it in a paper he wrote in 2016. Ralph Merkle, the famous for the nano tech and before that he actually did some really great cryptography in the 70s. Famously created the idea of the Merkle tree.
Trent McConaghy: As a data structure with some nice integrity guarantees. But so DAOs are a new form of life. So Bitcoin itself as a DAO. Basically it’s this organism that’s out there that it’s really hard to kill. It has many, many copies. It’s somehow convinced humans to let it keep running on machines that humans have given it.
Trent McConaghy: And it has properties of resilience. It’s also anti-fragile. So it’s been hacked or people who have tried to hack it several times. But every time people try to hack it, other than people come in and fix any put into bugs. So the last major bug goes in Bitcoin to my knowledge, was a transaction malleability bug in 2013.
Trent McConaghy: And other, that got fixed. And to my knowledge, there haven’t been any major hacks to Bitcoin at all. And 2013 and before it was so young and so under the radar that no one really, nothing really bad happened to Bitcoin. So that’s an example. Another, any of these permissionless blockchains out there are DAOs, sort of thing are DAO et cetera.
Trent McConaghy: But what’s cool you don’t need to run your own chain to have a DAO you could simply run a smart contract on top where the control is more autonomous. So you can put other smart contract where you can’t change it once it’s put out there. And if it’s doing things in an an autonomous fashion, then it’s a DAO. A decentralized autonomous organization.
Trent McConaghy: Another way to think about it is it’s maybe perhaps the next corporation. So corporations, they’ve been around a few hundred years, first came out of [inaudible 00:30:40] and then a little bit England. And they kind of, in the accumulated personhood over the years so that they have legal rights and so on.
Trent McConaghy: So they’re these autonomous organizations powered by humans that have their own legal rights. And are used to help, pool capital together to accomplish things that maybe a single human can’t do, right? Well, DAOs are that too, except they don’t necessarily need any corporate or legal structure. They’re just simply running on Ethereum or whatever.
Trent McConaghy: But they can be used to pool together resources. There’s this project of three years ago called BADAO. Which was basically targeting literally people pool together $150 million worth of resources and it was going to be this. A fund that was going to invest in other projects.
Trent McConaghy: It happened to have a smart contracts law so that the project got wound down, not so gentle way. But that was a really cool budgetary idea and there’s a lot of other DAOs. So there’s even projects that make it easy to construct DAOs for organizing humans in whatever way.
Trent McConaghy: So maybe whereas before you might’ve started a company to do something or maybe a co-op, now you can create a DAOs. And these DAOs, they can reshape the companies hierarchically if you want with people or co-ops. But you have way a much broader design space of organizing people.
Trent McConaghy: So there’s a lot more flexibility in how you implement governance in, how decisions are made, how resources are allocated and all of that. So I see it as a rash it up and technology for organizing humans and bots.
Jim Rutt: You mentioned that corporations have legal rights of quadsipersonhood probably too damn much in my opinion. But some of it it’s necessary. Is it fair to say that a DAO has analogous rights but the rights are in terms of rights to enter transactions into a smart contract, and or public ledger system and that that is actually their leverage?
Trent McConaghy: Yeah. So basically for example, if you create a DAO a smart contract running on Ethereum, itself can have its own resources out of the box, its own essentially private keys, all of this thing. To hold a hold its own tokens, have its own wallet and so on. And with that then it can transact with, other agents running the system, other DAOs, other blockchains, even external things in the real world. So absolutely yes.
Jim Rutt: That provides a great base I think for our audience. Now let’s turn to your most recent project, which is still going strong, which is the Ocean Protocol. It uses many ideas from the blockchain space that we’ve talked about. Tell us about the Ocean Protocol.
Trent McConaghy: What we’re trying to do is catalyze a new data economy one that is open and permissionless. And I guess by analogy in the world of blockchain, blockchain started out where are there on the heels of the financial crisis in 2008. Where there was this economy, the money economy that was closed and opaque.
Trent McConaghy: And the decisions were made by a small handful of people. And Bitcoins was sparked by that and was released and has been unlocking this open, permissionless, more transparent, money economy. Now if you look at world bank fingers and stuff, something like 25% of the GDP of the future will be based on data.
Trent McConaghy: And that’s very soon, like by 2025 or something. And even now, a huge part of the world is running on data in bigger and bigger ways every day. And we see influences around all the time, but influences that are negative. So things like, large corporations like Facebook and Google, they’ve accumulated a vast amount of wealth because they figured out how to leverage data.
Trent McConaghy: And convert that data into value using AI and then accumulate wealth that way. And so what we have then now is essentially we have a data economy, but it is closed and opaque. And controlled by just a small handful of players for the vast majority of data. And this is very dangerous for society.
Trent McConaghy: We shouldn’t have to rely on Facebook improving its algorithms just so that we, that America has a democratic election for example. Or Britain has a democratic referendum. Instead, what we really should have is a substrate for data flowing around that is a public utility.
Trent McConaghy: That makes it really straightforward for data to flow from A to B to C to D, without these middlemen that can be compromised that definitely have the wrong incentives. Facebook is a for profit company. At the end of the day, they have their fiduciary responsibility to their shareholders.
Trent McConaghy: What’s your objective function? Make more money. How do they make more money? Send more ads. How do they send more ads? By basically data mining you, to target you as best as you can. So Facebook is incentivize to get all of your data and to basically sell it back to you in the form of ads.
Trent McConaghy: And Google is no different, although they managed to veil it a little bit more. So those are probably, the best example is Google or Facebook these sorts of guys. And there’s actually not that many companies like that. Most of the Fortune 500 companies out there, they have this idea that they know they have lots of data, but they haven’t been able to convert that to value with AI and so on.
Trent McConaghy: And on the flip side there’s a lot of AI folks out there, who have a lot of AI expertise but don’t have access to a lot of data or the resources to really turn it into value. And this is why, it’s only these organizations that have huge amounts of data and huge amounts of AI expertise that really compute.
Trent McConaghy: That they are extracting the value. This is super dangerous for society and not the future world that I want to live in. So as I was realizing this, I guess I should mention the reason AI and data are related. Starting in the mid 2000s AI researchers started to realize if you wanted to have a more accurate AI model, then you add more data.
Trent McConaghy: And the traditional way was, get a PhD to spend four years of improving the algorithm. But the new discovery was basically, okay, just add 10 X more data, at a 100 X or a 1000 X more data. And your error can go from 20% to 10% at 5% to 1% to 0.1%. And so it’s actually embarrassing to AI researchers because you no longer need to justify a PhD.
Trent McConaghy: You just need to spend a bunch of money and go and buy data. And this is why Google for example, by satellite imaging companies and so on, so they can get more data. So the name of the game is data for AI. Anyway, so this is the landscape that we’re in, where we have this opaque, closed data economy.
Trent McConaghy: And I started asking the question, how do we have an open permissionless city economy where you can, where there’s in incentives for commons public data. As well as means for people to leverage their private data to transact it if they want. But giving shades of gray or privacy because it needs to reconcile privacy.
Trent McConaghy: So that’s the goal of Ocean. And what we’ve designed is something that manifest that. So Ocean we see as the substrate for a data economy. And how it’s designed, it’s designed to unlock data in three ways, all in ways that basically overall then equalize access to data.
Trent McConaghy: Which in turn equalizes opportunities for AI, spreading the potential benefits out to everyone. But it’s really equalizing opportunities, benefits we can’t control, right? It’s really up to the people themselves to use it. And we unlocked data in three quick ways, by making it easy to build marketplaces and have permission on data with fine control.
Trent McConaghy: By bringing compute to the data itself, which allows people to get value from private data without losing control of it or without compromising privacy. And finally, just like Bitcoin has this objective function to maximize its security. Where it pays Bitcoin tokens if people contribute to the security.
Trent McConaghy: Ocean has an objective function to maximize the supply of relevant data. And by doing that, basically it pays tokens from the Ocean network to people who supply data on relevant data. So that’s what Ocean is in a nutshell. It’s super opinionated, in a way where we think it’s really important.
Trent McConaghy: It’s super opinionated basically trying to create this public utility network for the benefit of society.
Jim Rutt: Because you get with a tangible example of what may be two different examples. One for a data owner who may want to put their data out on the Ocean network. And the second for say a pool of AI talent that might want to use the Ocean network to get access to data.
Trent McConaghy: Absolutely. So one of the examples we started with and we are continuing down this path is, Toyota Research Institute came to us saying, “Okay, we’re working on autonomous vehicles. And we’ve run the numbers and we see that we need 500 million miles driven, in order to get accurate enough cars that won’t crash.”
Trent McConaghy: And we’ve ran the numbers and it’s going to take us towards a couple of decades to get that on our own. And that’s accounting for exponentials and everything. And they said, “We would love to find a way where we can, share and buy and sell autonomous driving gear with the other auto makers. So that together we can get accurate enough cars and then it’s a win win for everybody.”
Trent McConaghy: So we actually built a prototype for them it worked nicely, but they saw that it, Toyota itself wouldn’t be able to roll that out further on its own. Simply because it really needs to be as an industry consortium. So the person behind the project in Toyota, Chris Bellenger, he then left Toyota and created a foundation called MOBI.
Trent McConaghy: Which is a mobility open blockchain initiative. And within MOBI, one of the key initiatives there is this autonomous vehicles data exchange. And basically that is a win win across the board for the automakers as well as for your grandma or whoever who gets autonomous vehicles sooner because then they get better mobility options.
Trent McConaghy: And finally, for anyone who rides a car, because as soon as autonomous vehicles get accurate enough to be as accurate as humans, as safe as humans, I mean later, they’re more accurate. Right? And that’s going to be beneficial. So if we fast forward 20, 30 years, people will be shocked that humans ever drove cars at all, given how dangerous it can be.
Trent McConaghy: So that’s one example. And I see it as quite a promising thing. It’s continuing down the path.
Jim Rutt: Let me drill into that one a little back I like that example. So I’ll try to restate it. Tell me if I get the restatement wrong for our audience. The idea is this MOBI foundation would encourage all the, or most of the major players trying to build autonomous vehicles, to pool their experiential data into a common database.
Jim Rutt: Or at least into a federated database where all of them could have access to each other’s data for doing analysis, replays, simulation, et cetera. Is that approximately what’s proposed here?
Trent McConaghy: So the data actually can stay on premise inside Toyota, Daimler, et cetera. And Ocean Axis is this access control layer. Where basically if I’m an engineer in BMW, I can access the data and all these other automotive suppliers of the data. Without actually having to download that data to my own silo, if you will, right?
Trent McConaghy: And then I can even be training models across these different silos eventually, right? So you can do, business intelligence across silos. You can train AI models across silos. But the that benefit is, as if you had pooled all the data into one big pool.
Jim Rutt: From Toyota’s perspective now, it sounds you have a mechanism to allow computation to run at Toyota on Toyota’s data. Without that computation having access to the lowest levels of the data. At least in a way that the data could be say stolen by GM or Mercedes.
Trent McConaghy: Yeah. And this is a key thing that Ocean is building. When we started out the project, we just said, “Okay, if you want to buy the data, if you want to use it, then you have to download it.” And of course we ran into this challenge very quickly. Well, what about data that is personally identifiable or really, really valuable? Right?
Trent McConaghy: And the most valuable data is private data. What about that data? And we explored various schemes, right? And there’s a bunch of schemes emerging in sort of crypto space. Things like homomorphic encryption, which is computation on encrypted data. Or multiparty compute, which is having many different parties each computing just a tiny chunk of the computer and then trying to emerge the results together.
Trent McConaghy: And those are all really great techniques, but they’re not quite ready for prime time. They’re getting there, but it’s still two, three, four years away. And even if you do have those, if you’re a big enterprise, these techniques are asking you to still put your data out there in the wild and hope that the cryptography keeps you safe.
Trent McConaghy: It’s much more secure feeling if you know that your data isn’t leaving beyond your firewall. So how do you address that? And the answer is bring the compute to the data itself, right? And let the algorithms get to train these models behind the firewall. And then the final model can either stay behind the firewall or it could be released as well.
Trent McConaghy: Although if you release it, you have to be very careful about people reverse engineering it. So let’s say if this way is to have it stay behind the firewall or stay in a decentralized fashion where you’ve got some limits on how many queries people can have. Otherwise they get reverse engineered.
Trent McConaghy: So you basically have to think about all the levels of the stack, right? Like where does the data go to, what are all the different attack factors of how people might try to grab the private data. And then come up with a system that addresses that. And at the same time accounts for the pragmatics of technology that’s available today or what’s just that thing that’s one step away.
Trent McConaghy: So in Ocean we’ve used as much off the shelf technologies we can, invented what we needed and where we needed. But in general, maybe I’m a creative entrepreneur and all this, but I really try to minimize the amount that I innovate. And that my team does and instead try to leverage everything we can off the shelf and only innovate where needed.
Jim Rutt: And that’s always the smart way to go. When I did my own startups, I always tried to only have one hard problem that I personally had to solve or my team had to solve. The rest was basically systems integration, but in an intelligent fashion but always liked to have one ugly problem that gave us a sustainable advantage.
Jim Rutt: Let me go back to drill into your example though. One more time to make it clear to myself and hopefully to our audience. So let’s say Toyota has a big body of data from operating a fleet of prototype, early self driving cars. And through the Ocean Protocol they’re able to cut a deal with BMW to have some to that data, and presumably in return for having access to BMW’s data.
Jim Rutt: Now some engineers at BMW want to train a, or continue to train a model on Toyota’s data. How does that actually work?
Trent McConaghy: There’s a few ways where the main way of thinking about this is a new emerging technique called federated learning. And generally speaking, how it works, so this idea emerged a few years ago in this centralized computing space. Where you initialize a neural network model with just random weights.
Trent McConaghy: Like you initialize a neural model with just random parameters. And then you say, “Okay, now we’re going to update this model from the data of silo A,” so maybe all the data from BMW. And then so you don’t send the model there, you just updated it, you get an update delta record if you will.
Trent McConaghy: That then comes to the model and then while it gets updated. And then you go along to silo B say that Daimler has, and then you update the model from that. And then you go to the silo C and you update the model with that. And you keep going down the path until you’ve gone through all of the silos.
Trent McConaghy: And now you’ve got this model that, has been training all the silos, but none of the silos have seen the model in detail itself. And the model itself then is trained. Now this idea of federated learning has been around a few years. It’s taken off in some ways in a cool way.
Trent McConaghy: And Google to its credit has pioneered it, but it has pioneered in a centralized way. So there’s three, levels of the stack that need to be decentralized. There’s the data itself, that needs to stay in the silos or stay on the consumer mobile phone or whatever. There is the orchestration of the training and there is where the final model resides.
Trent McConaghy: So the Google version of this has a data itself sitting in the silos, but Google itself with the orchestration in a centralized way. So we can see all the things that are going on and on basically or has the opportunity to. And the final model as well is controlled material. So you can reverse engineer that and yank out the PII potentially.
Trent McConaghy: What you need to do is you need to have all three decentralized. You need to have the data decentralized, you need to have the orchestration decentralized and you need to have the final model decentralized. And by doing that, then you can retain the private data. So Ocean at its heart actually in order to make this happen, in unlocking data to unlock private data, it needs to do this decentralized orchestration.
Trent McConaghy: Which is basically think like a computer pipeline that, grabs data and then cleans the data and then trains the model and source the model, et cetera. So step it, one, two, three, four, five, but it oversees those steps in a decentralized substrate. Instead of having a single entity controlling that and that’s critical.
Jim Rutt: Let me drill in one step further, see if I can make it clear to myself and the audience. Let’s imagine the BMW guy has created a TensorFlow based model and is running it against his data. He’s done. Now he wants to run that, he wants to improve that model by running it against Toyota’s data. How does he actually do that under the Ocean Protocol?
Trent McConaghy: He will kick off an overall job that says, “Learn on this dataset, learn on this dataset, learn on this dataset.” And the Ocean Protocol itself has middleware software that knows how to connect to each of those datasets and basically run a local Docker container. That is running locally side-by-side and the BMW data right in this container where it’s, doing the basically the model update.
Trent McConaghy: The delta weights if you will. That then once they’re computed, they are sent to the model that is sitting there. So using Ocean Protocol basically. And so you’ve got this model sitting in this decentralized way. It just received the weights from this first silo. And the weights were computed orchestrated by Ocean running this local Docker container side-by-side with the data.
Trent McConaghy: Once those arrive with the model, you have another Docker container running in the Daimler, in Docker container running next to the Daimler data. And it computes weights and sends the weight update to the neural network model itself too. And then it keeps doing that and one silo at a time.
Trent McConaghy: BMW, Daimler, in GM, Ford, et cetera. And in the end it’s got a model that has all the data from all the participating automakers.
Jim Rutt: And does that model then get shared as well or is that just BMWs model at this point?
Trent McConaghy: It depends on the use case. So in the case of autonomous driving data, autonomous vehicles, if there aren’t privacy issues for that model, then it would be okay for BMW to control that. But in other cases, if you have a model that is sufficiently high fidelity, then a smart AI engineer will be able to yank out personally identifiable information.
Trent McConaghy: And you don’t want that. It’s super dangerous, right? And so how you account for that is the model itself needs to be basically running as its own, essentially a DAO inside this decentralized substrate. Where any one entity can only make a few queries from it per day for example, right? Or whatever’s needed.
Trent McConaghy: So you want to make sure that no one can query it enough in order to yank out personally identifiable information. So basically that that’s the way, so just I talked about these three layer three things you need to decentralize before the data itself, the orchestration for the training and the final model.
Trent McConaghy: And ideally you have all three, in some cases you can get away with having that neural network model held by BMW for example. But only if it doesn’t have the privacy issues.
Jim Rutt: So again, just to drill a little bit more into what’s actually happening. And say the BMW TensorFlow based model actually gets copied over to the Docker instance at Daimler or Toyota and run their against the data.
Trent McConaghy: No, no one can because if that was happening then inside Daimler, if you have a sneaky AI engineering inside Daimler, they can grab the model reverse engineering and yank out at a bunch of PII that might’ve come from the BMW data. You see.
Jim Rutt: So the data flows across the Ocean Protocol from Daimler to BMW?
Trent McConaghy: No. So you basically got this model that is getting built that’s in the decentralized cloud, right? In Ocean as its DAO. And its getting update an update from BMW, its getting an update from Daimler. Its getting an update from other automakers.
Jim Rutt: But where does it actually run as it’s updating its weight? Because as we know that’s computationally intensive and bandwidth intensive to be scanning vast pools of data and updating weights on our potentially gigantic, giga size in a neural net.
Trent McConaghy: Yeah. So the model itself, I mean most of the compute is actually in computing the update part itself at right next to the data. So that’s integrating of those weights into the model itself is much less compute intensive. So if you want it you, there’s a few options. One of them is where you have some basically running directly on chain which is not really very feasible, for any real size to be honest.
Trent McConaghy: Another one is where you have some shared substrate that all BMW, Daimler et cetera are running together that, they run as their own DAO each of them has one key. And then each of them has set up the rules where, this model that is being queried, they can query it and no one else. And each of them has rate-limiting their query.
Trent McConaghy: And in that case, you’ve got keys around this, it would be running on one of their data centers where there’s BMW data center or otherwise. But in that case it would be multisig up, each of them would have keys to only do certain things.
Jim Rutt: I still don’t quite see how the computation touches the data in a way that’s sufficiently tightly coupled that it can be efficient.
Trent McConaghy: Right. So are you talking about with the training or are you talking about with a prediction given a model?
Jim Rutt: So let’s say Toyota has 10 terabytes of very low level data from a whole bunch of prototype, self driving car. And you need to run that very low level of data through a trading protocol presumably. And that is updating weights continuously, including doing expensive things like calculating gradient descents, et cetera. And try to do that over a wire seems highly unfeasible.
Trent McConaghy: Well, yeah, I know you’re not doing it over a wire so maybe I wasn’t clear on that.
Jim Rutt: Where does the data and the compute, reside within the same, at worse data center but better still the same a tightly coupled cluster?
Trent McConaghy: That’s exactly it. Inside the data center, inside of the cluster, right next to each other, right? So they’re talking to each other, not even over the network at all, right? Then there’s memory that they’re sharing on the same machines and so on, right? And this is actually happening today already, right?
Trent McConaghy: So there’s a version of TensorFlow that is the federated version and TensorFlow threat federated. And it already offers these options. And the versions of Ocean with federated learning that we’re moving towards is leveraging the TensorFlow technology. But then pulling it in the other levels of decentralization at the level of orchestration.
Trent McConaghy: So that you don’t have man in the middle attacks there and the level of the final model itself. So that you can’t have PII, personal private information leaked out of that final model.
Jim Rutt: Does it also provide controls so that if all the parties agree, BMW’s model is owned by BMW? So they can, and again you say the model has to go over to Daimler to run against the Daimler data. So presumably you need some very powerful security to keep Daimler from grabbing the model while it’s in their data center.
Trent McConaghy: No, the model doesn’t need to go to Daimler. So the model setting, basically to compute these updates with grading descendants stuff, the heavy compute, each of those is done they compute next to the data itself. So have you compute next to BMW data and the result of that is a weight update that goes to this model city in the decentralized cloud, similar for Daimler.
Trent McConaghy: But what’s not happening? The model is not coming from the cloud down to Daimler at all. So there’s no risk of that way of the BMW data, getting in the hands of the Daimler on that way.
Jim Rutt: But you have to have the network though to calculate, the deltas, right? The neural network has to exist close to the data.
Trent McConaghy: The deltas are basically the way they federated learning is, you don’t need to take the previous, you can do all the weight updates. You can repeat them all simultaneously if you really want. I just described it where it’s serially because it’s easier to understand conceptually as humans.
Jim Rutt: Okay. I guess I’m going to have to go look into this. This is a, I was using TensorFlow about three years ago, but haven’t looked at it since. I’ve got to go back and learn about federated TensorFlow it looks like.
Trent McConaghy: Sure. And to be clear, there’s the federated TensorFlow but then on top of that you have to decentralize not just the level of the data, but also the level of the orchestration in the final model, right? So that’s the key distinction.
Jim Rutt: Yeah, I understand that. But I know at the end of the day you still have to get two things close to each other or it ain’t going to scale, right? You got to get the, the delta calculation from the training set into a updated model, close, very close to each other. In technology time space.
Jim Rutt: You and I know exactly what we’re talking about it’s hard to give me immunicate this to everybody else. But this is a which I am the most extreme zealot that I always make, have to make sure that people have, it made a fundamental non-scalable architectural decision.
Trent McConaghy: Absolutely. I think there’s, I mean you and I have really had this conversation, but one thing we are very proud of from Solido days, at the heart of Solido we were doing Riverbend estimation as one of the products. Calculating is this circuit, one in a billion chance of failure versus 1.1 in a billion.
Trent McConaghy: And of course then to do that you have to draw potentially millions or billions or tens of billions of Monte Carlo samples. And we had got things so close to the hardware level that, we were drawing 1 million Monte Carlo samples per second on one core. These are the things that get me very excited too. Hence many conversations and same thing in the modeled training, right?
Jim Rutt: Let me ask another question. This is one that comes from my long career of dealing with database products in my own companies and that Thomson Reuters, et cetera. Which was what we called back in the day, the ontology problem, right? Let’s stick with our example of self driving cars.
Jim Rutt: The way Toyota may have structured the description of the low level data about the operation of their self driving car. Maybe incommensurable with the way BMW did. They make call things, not only just call the same thing different names, but divide the space of measurements up in ways that aren’t commensurable. How do you deal with that problem?
Trent McConaghy: Yeah, that’s a great question. There’s a few ways depending on the particular market if you will. If you are a vendor selling data and your data format is incompatible with everyone else knew by your stuff. So you’re incentivized to get your data in a format that everyone agrees with.
Trent McConaghy: For sort of emerging problems like the AB training, this is where things like the MOBI working groups are very useful, MOBI is this, mobility consortium. So there’s a working group on autonomous vehicles driving data and they work together around. And come up with the different variables that they want to point to, what order they’re in and so on, right?
Trent McConaghy: And in general, this is the question of standards, how do standards get set? And there’s other, traditionally there’s a few ways. There’s the top down way where our company says this is the way it is. Like Apple say, here is FireWire use it or not. But you can also go bottom up where a group of engineers were representing various companies come up with a protocol and then it gets deployed.
Trent McConaghy: And this is things like Bluetooth and WiFi for example where it’s, there are [inaudible 00:57:42] standards that could develop by working groups over a number of years. So I see the same thing here. At the end of the day, the standard they could use a lot are the ones that went in the market. Maybe that’s a truism, but yeah that’s how it is.
Jim Rutt: So in this case, again, I want to try to keep it on the same example. MOBI may get agreement from a number of car companies to, structure the low level data that’s being spot out from their self driving cars into a standard format that they all agree to use.
Trent McConaghy: And sometimes they’re going to be competing formats, right? Like a PhD student and a professor out of UCLA came up with a protocol to connect a network of networks. And the governments ignored that and decided to come up with their own. And the government one was many European nations and so on they called it OSI.
Trent McConaghy: But the PhD student, and the supervisor [inaudible 00:58:28] Bob Kahn, they’ve invented TCP IP and it was running and it was working. So in this case you have two competing standards and basically the one that was working one, right? It was also a few years earlier.
Trent McConaghy: But overall, that’s the NS example of two completely different approaches to development or of a standard or a protocol. And one of them emerging is the winner.
Jim Rutt: That’s funny. I have actually a talk about that exact issue. How TCIP beat OSI. So not just it can be tangible in this automotive, automatic driving space. Has MOBI been able to get the various manufacturers to agree on a shared ontology of how the low level events are described?
Trent McConaghy: Now they’re moving towards that and they have clout in the sense that they represent 80% of the world’s auto production. And they have regular meetings, each of the working groups has regular meetings and so on. So to my knowledge, there has been the meetings continue. In the last couple months, I’ve seen things happening that way.
Trent McConaghy: So even just for the autonomous vehicles. But who knows, maybe there’ll be some company X that comes along that says, “Here’s our autonomous vehicles. Here’s the standard for everyone follow or you’re left in the dust.” So at the end of the day, it’s what’s successful in the market.
Jim Rutt: I do know for sure that these ontology problems are tricky and subtle. And if you don’t solve them, that shit don’t work.
Trent McConaghy: Actually, one that we encountered was, when we were doing IP in the blockchain with this project called Describe. We hadn’t developed your own protocol at first of how do you license art? How does someone claim copyright? How does someone transfer rights of a limited edition? All of this.
Trent McConaghy: And we did a first cut but then people said, okay, I want to slice and dice this to many different owners or many different jurisdictions or time-bone and so on. So we went back and we looked at what does the music industry do? What does the photography industry do? Et cetera.
Trent McConaghy: And in the end, there had been some protocols developed out of London about seven years before. That weren’t quite ready for blockchain, but we tweaked them and made them ready. And that was very nice. So sometimes there, you just have to dust off some great ideas that people came up with five or 15 years ago.
Jim Rutt: Absolutely. We’d talked to pretty good example about people who have lots of data and want to do something with it, perhaps cooperatively with the automotive company example. Let’s talk the other side of the big picture that you drew earlier. Which is some folks say who have some wonderful AI technology, that they want to be able to essentially sell as a service across data to various customers. How would that look on the Ocean Protocol?
Trent McConaghy: So Ocean at the end of the day, it’s designed its heart is decentralized orchestration which is these AI complete pipelines. And then services get fed into these AI complete pipelines. And these services are services that provide data or services that provide compute.
Trent McConaghy: Those services that provide compute can be manifesting AI algorithms. So that would be outweigh where someone is providing a service where they’ve got their own even black box AI algorithm. Another way where is where, let’s say you’re a PhD student or you’re a startup and you have developed some fancy AI algorithm. You can actually use some supply that as data in sort of this way with Providence as security and so on.
Trent McConaghy: And then other people could be using that to be running with their own compute. We will see, I think there’s quite a few possible ways that this can emerge, but at the heart of the Ocean Protocol is for these pipelines. And there’s quite a few directions that the ecosystem can emerge this way.
Trent McConaghy: And to give a feel, right? The service is feeding, the data side they can be from behind silos. They can be from the centralized web things like Amazon, Esri. Or they can be from the decentralized web things like storage or file coin. And same thing for the compute, right? It can be behind the firewalls, Docker containers running the BMW data centers.
Trent McConaghy: It can be centralized cloud like Amazon and C2, or it can be decentralized cloud like Golem or specializations. So Singularity now is wonderful example of decentralized compute with a real focus on AI compute. And the supplying the AI algorithms themselves. And that’s where there’s a very nice interplay between Ocean and Singularity now.
Jim Rutt: Yeah. So that was my next question. These spaces are partially adjacent, partially overlapping, partially hierarchical. How would you compare and contrast Ocean to Singularity net?
Trent McConaghy: Maybe as a pre-phase, Ben and I have known each other for years. We’ve been friends for years and he’s officially an advisor to Ocean, I’m official an advisor to Singularity, so that lays the groundwork that there’s, we’re just bound to be cooperating. And in fact we are, both Ocean and Singularity now are platforms in a sense.
Trent McConaghy: But also with aims to have applications running on top of them. Singularity net has more of a bias towards AI algorithms running, agents running, AGI, this sort of thing. Ocean has more of a bias towards unlocking data and supporting these compute pipelines. So Singularity net will be plugging into Ocean and supplying compute and algorithms.
Trent McConaghy: And similarly Ocean will be plugging into Singularity net to provide data. And so you can have either one on top, another one below. But probably the healthiest way and most useful way is simply to think of them as side by side feeding each other.
Jim Rutt: Seems like that makes a lot of sense to me and should help both. And since I’ve been close with both Ben and with you, I hope to see you guys work very cooperatively. I’m glad to see that you’re both working together in a nice way. That’s all good.
Jim Rutt: Now let’s switched into a little bit different level, we talked earlier about governance and how these new public ledger and smart contracts, DAOs allow new models of governance. How’s Ocean governed?
Trent McConaghy: Eventually Ocean will be a public utility network that is fully permissionless and the code will have stabilized such that it won’t need any change or anything. And in that case it will just be this autonomous thing. And asking how Ocean is governed it would be asking how the air around us is governed or the wind or the soil or anything.
Trent McConaghy: And there can be a degree of governance. So that’s the ideal, right? Where we want to get to. And of course that means also it’s permissionless and since you’re persistent and all this. And that’s where we’re headed, but it’s going to take a long time to get there. And in the words of one of your previous colleagues David Holtzman who had helped architect the modern DNS bake slowly.
Trent McConaghy: And this is a maxim that David has repeated to me many times, bake slowly. And in the context of rolling out Ocean or decentralized networks, what this means is you start with something that has centralized control. And bit by bit you decentralize as the system stabilizes and so on.
Trent McConaghy: And you try it while you’re doing this, you try to make it so that people don’t have to trust you very much either. So it doesn’t have to be where like at the final final step I remove all my keys put up to them I got full control it’s not that. So it’s about giving up control bit by bit by bit as well.
Trent McConaghy: And so Ocean right now, we are running it as an authority based network of federated network. Mostly with [inaudible 01:05:17] and within hopefully a year, maybe sooner we will be running on a fully permissionless substrate. Such that anyone can be a validator on that.
Trent McConaghy: And then, so that’s the lower level of the substrate that runs the smart contracts, the Ocean’s smart contracts. We would be running, we’d be happily running on say Ethereum right now. If Ethereum was fully scalable, but it’s not fully scalable and we need the scale.
Trent McConaghy: And we’ve written about that too. So that’s the lower level. The higher level of what governance comes down to at the end of the day is how does the software itself get upgraded? Right? And every other governance discussion, basically falls back into that. How does the software itself get upgraded?
Trent McConaghy: And so how it is right now, there’s two parts of the Ocean system. One part is the smart contracts of the code of the number of Ocean tokens. And that has no, that’s basically a hardwired and in fact it’s sitting on Ethereum and we have a bridge. So it’s very, very hard for us to change or touch you’d have to do a hard fork, which is a whole other story.
Trent McConaghy: And then the rest of the Ocean system though, we actually have it set up where there’s basically a version control software. Such that, if we find a bug then the team can go in and update the code. Using basically M of M signatures, this so called multisig. So three of five people need to agree that the code doesn’t need the upgrade, so then they do the upgrade and suddenly the code is upgraded.
Trent McConaghy: But as time goes on, towards this final idea where it’s this public utility that no one can own or control at all, then we will remove all the keys, all ability to upgrade it all. And the only way that you can change it after that is with a hard fork. And a hard fork as a form of governance too, by example for maybe for your audience.
Trent McConaghy: When Bitcoin its the only form of governance is hard forks where if you want to upgrade the Bitcoin protocol, you basically create new software you start running your nodes. Where those nodes copy over this date from the existing Bitcoin. And you try to convince everyone else that’s running Bitcoin nodes to run your new nodes to run your software.
Trent McConaghy: And if you convince enough of them or if you convince all of them, then you’ve successfully executed the hard fork. Sometimes maybe you can only convince half and then there’s truly a hard fork. You’ve got your system, and you’ve got the previous people who prefer the old way.
Trent McConaghy: And that’s okay, that you’re two communities each has his own philosophy how things should go forward. That’s completely healthy. So fork away. But a hard fork, while it’s a form of governance, it’s very painful to execute because just logistics and so on. So you can’t do it very often, maybe every three months or every six months.
Trent McConaghy: And if you have a very young system that has potentially a lot of bugs and you want to be able to change them quickly, you want a faster way to upgrade that. And that’s really what this on contract software upgrade mechanism is that Ocean has.
Jim Rutt: Interesting. Now are you going to be fully on, in fact, are you now fully open source? Do you intend or if you’re not, do you intend to be?
Trent McConaghy: We are fully open source with an Apache Two License, which means that there are very few restrictions on it.
Jim Rutt: So in other words, if someone did want to take what you have and do a hard fork, they could just, they’re free to go ahead and do that.
Trent McConaghy: Yeah, I mean they can deploy their own system of Ocean tomorrow, right?
Jim Rutt: And then modify it and go along their merry away, right? And as we know that’s happened with Bitcoin, how many forks have there been at Bitcoin? A lot have been attempted what about? Four or five actually have a little bit of validity going on with them.
Trent McConaghy: For sure. And some other systems they do a hard fork, but they modify the code of law and start from scratch. This is things like Litecoin which is a modification of Bitcoin or Zcash also. But they really go in their own direction, right? But then there’s other modifications in the Bitcoin system that retain the state and try to have just very small changes.
Trent McConaghy: So yeah, there’s Bitcoin cash and Bitcoin SV and many, many more. And what’s interesting though, if you rewind two or three years, there is a big fight or big fights in the Bitcoin community. Bitcoin should go in direction X, Bitcoin should go direction Y, direction Z.
Trent McConaghy: But what’s happened over the last two or three years, each of those different philosophies is now invited in its own chain. Say so Bitcoin itself is the biggest, but there’s maybe three or four others that are each worth say $1 billion.
Jim Rutt: Yep, exactly that was an example I was thinking of. The Ocean Protocol has its own token, the Ocean token. What role does the Ocean token serve and how is it linked back to governance, if at all?
Trent McConaghy: Those are great questions. So overall it serves a few roles. First of all, it’s a unit of exchange. So when you’re buying and selling data sets with Ocean, then inside the smart contracts itself, you’re buying and selling with Ocean tokens. There can be in last mile marketplaces where people can buy and sell with say US dollars, and then in just as conversion from US dollars to Ocean. And that’s completely okay too.
Trent McConaghy: So that’s the starting point. But if you had just that, then there would be no point of an Ocean token because, you could just have some coin that is a representation of the US dollar like die or something. So there has to be other reasons to have a token. And the core reason for Ocean to have a token, comes back to this objective function that I mentioned before.
Trent McConaghy: Which is we want the Ocean network to maximize the supply of relevant data. This is open commons data. So we want there to be millions, even potentially billions eventually of datasets inside the Ocean network. And so how we set it up is where it uses the mechanism of inflation.
Trent McConaghy: So whenever people submit datasets to Ocean, and other people download those, then the person who supplied the dataset and has put some stake in as well, they’re going to expect Ocean tokens. They’ll get rewarded Ocean tokens in a probabilistic way, right? Their chance of earning Ocean tokens goes up, the more that, the more useful datasets they supply.
Trent McConaghy: So that’s the keyway and that’s really around this one way of unlocking data, right? Really driving towards the incentivizing this commons data. And the tokens plays two roles there not only in being paid in Ocean tokens, but also you need to stake in Ocean tokens. The more you stake, the better, the more you get paid.
Trent McConaghy: And that in its own drives things. And really for the other parts, those are basically also as [inaudible 01:11:10] data marketplaces come on board, then they will have sticky mechanics too with Ocean tokens. So if I’m buying a dataset, I might get a discount if I hold a certain amount of Ocean tokens. And this is a common mechanic that you see across many other blockchain systems too.
Jim Rutt: So interesting. The amazing amount of creativity that can brought to the fore, by combining those elements that we talked about first. The idea of public ledger and tokens, smart contracts and DAOs, it allows a master builder like yourself to create something that’s never been thought of before.
Trent McConaghy: Absolutely. And actually like as a quickest side, when we started designing Ocean, it felt like we were … After I started iterating on it for a few months I’m like, when do I know when I’m done the design, the token design? And I really realized that I had been failing.
Trent McConaghy: So I started to sit down and I said, I realized that if I cast the problem as one of design of an optimizer then I lay out the objectives, I lay out the constraints, I have a set of design patterns or building blocks that I can work with. And basically once I’ve hit all my objections and constraints, I’ve got a pretty good first cut design.
Trent McConaghy: So I actually followed that approach. And very quickly within a matter of days, I actually arrived at a first cut design and then refined it and iterated more with colleagues and so on. And when I wrote about that, basically I started to call it token engineering. And that has emerged as a new, essentially discipline of engineering. It’s very, very young.
Trent McConaghy: But this pragmatic set of techniques where you’ve got a set of building blocks that you use, to construct your system. And an emerging set of tools to leverage those building blocks, a sense of ethics, a sense of and a duty of theory. That’s really the seeds for a new engineering discipline. So it’s really exciting help to spark that movement.
Jim Rutt: That’s great. And we are at in this brave new world where you can do that. Anyone who has the talent and the commitment and the drive can create something, as amazingly intricate as Ocean by brute force.
Trent McConaghy: Yeah. Brute force. But even better than brute force so is simply leveraging, the building blocks that have emerged in the token engineering. So things like, there’s something called the TCR which is, where you have a curated list of items. And there’s other mechanism building blocks for curation, there’s building blocks for governance, there’s building blocks for staking, et cetera.
Trent McConaghy: And then you can compose these in various ways, without having to reinvent the [inaudible 01:13:33]. And that’s very useful.
Jim Rutt: Yeah, even better. I mean, Brian Arthur, one of our Santa Fe Institute people whose book On Technology is underappreciated. Talks about how, most technology development is combining what’s already there in a new way that somebody hadn’t seen before.
Trent McConaghy: Absolutely.
Jim Rutt: Indeed. Okay. I looked briefly at how the Ocean coin was doing and it seemed to be tracking Bitcoin very, very closely, which I presume is good.
Trent McConaghy: I look at the price of Bitcoin now and then, once a week, once every two weeks. Same thing with Ocean. It’s very tough to model exactly how the dynamics of the crypto market work. There’s some macro trends. Ultimately how I see it as we continue to build the value into the Ocean network and the users then, it will drive the value of the Ocean token and there’ll be this positive snowball effect that happens.
Trent McConaghy: Bringing more people into that ecosystem, having a bigger impact, et cetera. So it’s really, leveraging the exponentials. If it tracks Bitcoin I’m sure, I don’t know why I would but that’s probably a good explanation.
Jim Rutt: Actually my hypothesis when I saw that, take a look at the two do a side by side, you’ll see it’s remarkable how closely it tracks it. And I presume it’s that the world at this point is not made a strong opinion about Ocean, but has concluded that Ocean is probably a good thing. Therefore, it’ll track remarkably closely with the big kahuna BTC. And of course that’s much better than your typical ICO, which starts high and goes low, right?
Trent McConaghy: Yeah, for sure. So I think, there’s pros and cons of having a token that’s out there public but it’s honest in a sense, right? And now there is a proxy for success overall. Maybe it’s not correlated with, the future promise but over time depending on the token mechanics you have, the value will come.
Jim Rutt: Well, I’ve got a few more things on my question list, but I think we’re pretty much out of time. I’d love to have you back on to get an update on how Ocean’s going and maybe jump into some of these other technologies. But I think this is the place to wrap it up. I hope our audience is taking a very rapid ride through the whole stack of blockchain, smart contracts and DAO as an example in the form of the Ocean Protocol. And has learned what a brave new world we’re living in here.
Trent McConaghy: Thank you very much for having me. It was a very fun conversation and I appreciate the depth of the questions.
Jim Rutt: Yeah, I really enjoyed the discussion as I always do when I speak with Trent. There’s very few people I know that I can have a deeper conversation with and this has been one of them.
Trent McConaghy: Well thank you Jim.
Jim Rutt: Production services and audio editing by Staunton Media Lab. Music by Tom Muller at modernspacemusic.com