Sign Massachusetts Banking IOU Computer

Industry sign banking massachusetts iou computer

so my name is Dominic and Suchi a string laughs Sarah Khouri and we are sort of incubator studio investor in open protocol systems so typically that means decentralized systems one of our main projects is Definity there's another one called fire which we may depending on time introduced to Definity as a project there's been quite a while in them in the making and it's aims to create a sort of limitless to centralized cloud which can do really amazing things so let's get into that in many ways you know it's a kind of world compute platform and for various reasons you can think about this as one of the ultimate business disruptors especially in Silicon Valley so this talk actually was originally just to go into some of the mechanisms that we use in particular applications of cryptographic randomness but before we get into that let's just recap what is that Definity aims to do so Definity is what you might term an open cloud so we're all familiar with you know for praia terry cloud like Amazon Web Services or Google Apps engine or is there something like that the open cloud will actually be run by lots of different people you know semi-professionally oftenly often but you know there may be computers in basements connected with consumer fibre that kind of thing and this cloud will have some very different properties to a proprietary cloud like Amazon Web Services so what is it what is it exactly so you can think about this as a virtual computer right and this virtual computer though is produced as a result of a decentralized network of clients mining clients a bit like mining in Bitcoin that run a common protocol right so in this diagram here we can see the virtual computer right and the virtual computer is produced as a result of these mining clients running a common protocol and we use the term stateful decentralized network protocol right so bit torrents are decentralized protocol but if I send you a file it doesn't really affect anybody else on the network whereas in a sort of simple example if I put a freely spendable Bitcoin on to the Bitcoin ledger right and let's say Sue spends it right then none of you can spend it right it's gone the network remembers that that Bitcoin has been moved from one address to another address right but that's achieved without there being a central record of the ledger right there's no server computer anywhere that keeps a record of what's on the Bitcoin ledger so in that sense it's a stateful decentralized Network protocol as a consequence of this network protocol being run between these clients there are no servers the network remembers the state of this virtual ledger so to finish is part of a new class of systems that use the same kind of stateful decentralized protocol to create a virtual computer and you know the virtual computer has storage has sort of memory right and has a CPU you can execute code right it's a machine you know has its own opcodes a bit like assembly language and you can run you can compile down high-level languages into these op codes and run these op codes on the virtual machine so you know you can read all kinds of software that runs on this thing and yet there is no single server recording what its state is the virtual machine is purely a manifestation of the execution of this decentralized stateful decentralized protocol between these clients and of course you know these clients are doing it for money even though there's no central controller and this mining client here is delivering some kind of cryptocurrency reward for the miner and here we can see as well that you know some of these participants because it's an open network a byzantium you know they're become faulty and they can behave in arbitrary ways and yet you know the the thing has to keep on running so you get this unstoppable network that produces a tamper-proof virtual computer and it's because then it's been created by this decentralized network but it is tamper proof right because if you wanted to tamper with this computer what could you do it doesn't exist it's virtual there's no central server you can go to and you know play with the contents of the hard drive or install the virus in the software right that's why it's tamper proof because it doesn't exist it's virtual right so a particular interest is creating a scalable virtual computer so I don't you've heard this expression before but we want it to scale out which is different from scale-up so we want the capacity of the virtual computer to increase as more mining clients join the network so that means more state storage more computation etc right and the reason this decentralized cloud and virtual computer is exciting is that it enables all kinds of very profound reengineering of business okay so you can re-engineer existing businesses using this technology this tamper proof technology you know that could be supply chain insurance global identity right and these things are enabled because the thing is tamper proof because it's transparent people can agree to cooperate Virant becomes very inexpensive to create financial exchanges and you can also create open versions of mass-market services right so a lot of people are thinking about how to create a decentralized uber right why do we need uber either technologies very simple right most of Hoover's money all those billions of dollars began to but don't go into developing software I mean it'd be very sad if it did because you know probably one talented engineer can code the basic app on his own in six months right so we may be asking well why does uber need these you know these billions of dollars and the reason is it has to go around the world negotiating with monopolies in different countries it has to pay the Paris taxi drivers Union it has to do a deal with DD and China and things like that right and it costs billions and billions of dollars but you know in the future we're going to see open source versions of these mass-market services right where the software is autonomous and runs on the open cloud and we're going to replace you know Zuber Twitter Facebook things even like web search right I know how long it will take but you know the future is open just in the same way the world is moving from proprietary close source to open source software the world will also move from proprietary businesses to these kind of decentralized alternatives which are kind of like open source businesses right so that's that's what we're we're doing and you know in the process of that you know you come up a can some very interesting computer science problems so one of them is you know how to organize one million-plus mining clients right you know if you think about the Google cloud platform reserve both of those services are backed by more than a million servers right so you know if you're going to provide a decentralized cloud with similar capacity then there needs to be a way of organizing that many mining clients which of course is more challenging right quite easy really I mean you know if you have seven million servers and you can get chubby or whatever it is and you know coordinate them centrally not so difficult right but you know if you have a million servers that aren't controlled by any central point right anybody can come into the network and start providing you know participating and mining the computation that you have no control whatsoever nobody has any control because there isn't an organization behind it it's just a protocol right how do you have you coordinate these million processes how do you prevent you know new entrants to the network you know breaking the you know subverting the system and making it break and it's not not so straightforward so one of the a good starting point when you're thinking about this is to make the observation that randomness is is the fundamental engine used to drive stateful decentralized networks okay and the first stateful decentralized network of course was Bitcoin so if we look at proof-of-work which was created by Satoshi Nakamoto we can look at how they did it that they do it and in this case you have all of the miners race to solve a puzzle right which in fact is just trying to trying different nonce values until a hash of this block of transactions falls beneath the target value right and you can set the target as low as you like to make it as difficult as you like and the solutions are found randomly in a place on distribution there's no you know clever strategy for solving this it's a brute-force task and when one of those miners wins the race right by finding a solution to the current puzzle he has made a temporary leader okay he's made a temporary leader and can broadcast a block of transactions right now the next leader of course is unknown no one knows who's going to solve this puzzle first because you can only solve the puzzle by application of brute force computing power so there's no way of knowing who's going to solve the current puzzle next and this is actually very powerful because it means you can't dos the leader because you don't know who's going to produce the next block you can't dos that leader and prevent them producing the block right you can't manipulate it you can be sure that you know everyone gets the fair crack of the whip and proportion to their computing power right or at least in the theory and you know so long as there's an honest majority of participants and as measured by their computing power the chain will function correctly because of this random selection of leaders and the adversary can't control the chain so in fact you know Bitcoin is this extraordinarily has this extraordinarily simple mechanism behind it and yet it achieves what would otherwise be very difficult you know it can bring a huge network of you know tens of thousands of participants to agreement and this would have been impossible using traditional distributed computing protocols which have arranged quite quadratic message complexity right so you need like 100 million messages something just to round you know running one of these traditional protocols so so you know it's a simple mechanism but nonetheless you know the insight and there were various other dimensions to it and made it possible to create this this first Network so anyway I mean you know so our thinking is well if if you know randomness is the fundamental engine of these kinds of networks then the obvious tack is to try and produce the best source of randomness that you can write so in our case we use this thing called fresh or relay and it's a sort of unpredictable deterministic source of randomness that can be generated by decentralized networks and all of our other protocols both within Definity and in higher-level applications that depend on it so we have like a decentralized commercial banking system which gives out loans algorithmically using randomness all of it depends upon this incorruptible unpredictable randomness the Thresher relay produces so let's just quickly think about the model that we are using here so we have a network of processes maybe talking about processes when I say process what I really mean is you know a piece of mining software okay in Definity each mining process is a fundamental fundamental unit of computational resource all right we expect each mining process to have some amount of computational capacity and we have ways of measuring it they're connected a p2p broadcast network so it's a bit different a Bitcoin you can imagine something like academia right and each process has a mining identity right so if you want a mine Definity you know you can of course credit process but to join it to the network it has to have a mining identity that the network recognizes yeah I'll just like you know like a public key kind of identity and these identities are what mediates participation so if you had a private network created but private decentralized network created by a bank you know they'd have a trusted dealer that would define a list and before does move you all I mentioned that you know in the future we're going to see these technologies being used as a replacement for databases because although you know if you take this technology right which is designed for you know in this case millions of servers right and you take it and you put it in an organization as an alternative kind of database or application platform it'll be you know a hundred times less efficient a hundred times less efficient than the database of course right but that's not where the costs lie in a database like its chief you can go and buy some server blades and you can go and you know buy some database software right and you can get some exchange software and configure it and so on but that's not your real cost if you're a business right the real cost is employing the administrators right who'd you know dude disaster continuity the security guys who make sure that some hacker doesn't get in and get hold of the SQL interface and you know go select star delete or something right that's the real cost all of the human beings that you need to run the database right in in a mission critical application whereas with these decentralized technologies you know you may have many many more servers and you know the computations way more expensive but the systems don't need nearly so much administration right because they're designed to be much much more robust and they can withstand faults and things like that anyway so you know you know in a private network you know the IDS mediate participation and all that works but in a private network you just have a trusted dealer which means you know systems administrator you know in a public network you typically you know somebody all credit cryptocurrency deposit right to Crane identity so for example let's I want to create a I want to add a process a mining process to the network I might make a deposit of you know $10,000 in cryptocurrency so the identity and that that's like a strode into the identity so if my process misbehaves maybe I'll lose that and there are some other things too so we assumed any wind affinity massive network size because it's our main priority is to create you know and it is there's this big open cloud so this network is organized into random groups and just we won't we can come to how they're organized into random groups later but just assume for the moment that these processes are all assigned to groups and each process is a member of multiple groups the groups are intersecting imagine that the group has 400 members each group has 400 members right so you know there are thousands and thousands of these groups a group in order to participate in the network has to set up threshold crypto scheme right for example they can run VSS secret sharing protocol right and you know they probably just like 51 percent threshold so you know in in order to create a signature maybe 201 out of 400 have to agree right and we use Dan Bonnie's BLS signature scheme and has some very interesting properties and they're just the adjectives I suppose you'd use to describe BLS but it's unique and deterministic and essentially that means that within in this case with our threshold whichever 51 percent of the group signs a message the signature is always the same right and this is actually a very very powerful property so you get a group of 400 processes right and you take the threshold 200 and one of them and whichever 201 you take if they sign the message this message you know the same message then the signature is always the same and they can't subvert this so what we do is we have the current group here it is in pink sign the signature of the previous group right you can probably see what's called threshold relay and the signature is a random number obviously right if the signature wasn't around a number then you'd be able to predict the signature of the group and it wouldn't be secure right so it's funny kind of thing I mean III look at signature schemes or at least this one you know you need to turn mystic ones not really a signature schemes at all but more as you know random generators and trivially once you've got this random number right are in this little silly formula here you can just modular it against the length of the list of groups to select the next group right so you just basica ly have the group's sign the signature of the previous group creates a new random number it's completely deterministic right the group can't manipulate it and only requires 51% of the group to participate in signing to create it and then this signature is stuck into this super simple formula to select the next group which creates a new signature on the last one and so on so you're just creating a new random number using the previous random number as a seed and you get this completely deterministic see random sequence that can't be manipulated will always be produced as long as you know 51% of the group are online and functioning correctly which turns out to be very very useful however the proper term for this is verifiable random function vrf I don't know it's that familiar term I actually got a damp bond I thought I was calling an authenticated random number scheme and then when your guys at Dan Boneh said oh no it's not an authenticated random number scheme it's a verifiable random number all right okay I'll call it a verifiable random number then if you say so and the reason it's verified a verifiable of course is the the numbers that are produced in this deterministic sequence through relaying across the threshold schemes are verifiable by the group public keys right so if you see one of these random numbers broadcast across the network you know the block height you know the group public key and you can just use your verification you know signature verification function right so it's not possible for anybody to fool anybody else that the random number is something other than it should be so we've got this thing you know the new values are produced on Scholz agreement and i'm manipulable and unpredictable so in a kind of diagrammatic way and it's kind of straightforward right here we have this group at block height h minus one and the message is of course the threshold signature from h - - you can catenate it with the actual actually that should be right kept kicking concatenate with the block height put the signature shares into the pls special signature signing function use the output threshold signature to select the next group which uses the previous threshold signature as a message and you know create creates a new on a new random number and so it continues ad infinitum right and you can just keep on relaying through all the blocks producing this deterministic I'm an imputable unstoppable effectively unstoppable sequence of randomness so what does this cost and it's kind of mind-blowing ly inexpensive right so the message that each process in the group broadcasts because you know in p2p networks are broadcast networks for various reasons and it cloaks the identities of individual servers and things like that the message that they broadcast is only 84 bytes in size you know so you've got 20 bytes for the process ID which is the hash of the you know the hash of the processes public key does the signature share like a BL s signature shares 32 bytes and then they're gonna sign the actual message that for various reasons and that's 84 bytes so with the group size of 400 threshold and I by 200 201 you know the maximum data that's actually broadcast in the network right for this to happen it's only 34 kilobytes in practice you only need the first half of those messages anyway 17 kilobytes and then you can produce the threshold signature which as soon that's broadcast people stop relaying you know stop relaying these messages so in in practice you know you're probably looking at something like you need 20 22 kilobytes of data right on each depth right it's nothing right and we had a very very larger network you know a million processes distributed across you know Europe and the u.s. you probably got a delay diameter across the network for about two seconds I'd say so this cannot happens pretty quickly so the resilience as interesting as well because it seems like this can't be very resilient right but the thing is that the as long as you make sure the process processes and numerous as the number of processes in the block increases the kind of variance goes down pretty rapidly so if we imagine a network with only 10,000 processes of which 3,000 are faulty and of course actually this is a really kind of catastrophic situation I mean if you have a decentralized network of professional compute miners and almost a third of them are faulty you know that that's a pretty bad situation but nonetheless we'll take that 10,000 processes 3,000 faulty which obviously means that 7,000 are correct group size 400 threshold 201 you can just use hyper geometric probability to calculate the probability that 200 or more processes are faulty right which would stop the thresholds signature being produced it's like 10 to the minus 17 right I mean it's not gonna happen well that's different that just where you came in those and the the mediate the densities of the processes mediate participation so whereas with Bitcoin you couple civil resistance with the actual consensus mechanism with this you always imagine that it's the identities to do it so you know that could be you know cryptocurrency deposits that are made to the density's it could be and we use a kind of hardware scheme that's a bit different where think all unique State copy put it in a private network you wouldn't even need it right you just would be a trusted dealer we give out the IDS so but their handle so there's interesting thing actually with generally you know Bitcoin is a very funny thing I think it kind of holds up people's perceptions but he ain't good computer science design always involves separating outcomes concerns right you don't want a kind of mechanism which has multiple dependences so if you look at Bitcoin the cost of creating that random number right because the brute force piece is brute force computation is the civil resistance so you know the consensus mechanisms of the system by which a random number is generated to randomly select a leader is also that the mechanism that stops people cheaply entering the network but ideally you want to separate concerns especially when you're trying to create a more scalable network so that you can address each different concern more optimally so in the finna see we you know just have it membership is mediated by the identities and then you can imagine you know any number of different schemes civil resistance schemes of potentially applicable to the dentist's but I mean you know most commonly now people talk about proof of stake so yeah well you can actually believes proof of work is that yeah exactly as a membership yeah yeah yeah I mean in in this system you know we imagine that the miners will be professional miners so typically it'd be like you want to credit process okay deposit some thousands of dollars encrypts of currency to that to create the identity says if it's a little bit different to some of the models but yeah I mean it's it's kind of funny how the how robust it is right you know Jen intuitively imagine it would be that robust that if you had that many processes 3,000 of them are faulty and you take a group of 400 that it's 10 to the minus 17 that you're actually going to have you know sufficient number of faulty processes to stop the signature being produced threshold relay block shake so okay so we've got this kind of random beacon thing that just produces this deterministic uh manipulable incorruptible orbs unstoppable sequence of random values right how can we use them so here's an example a very simple example what I call a probabilistic slot protocol so at each block height right we've got a new random number and here we have a bunch of block Heights HHS learn H plus 2 and so on and for each block height the randomness the random number defines an ordering a random order and obviously of all the processes in the network and they're starting to slots there's a slot 0 slot 1 slot serious love 3 and so on all the way to n minus 1 slots and that's why you've got this p with some made-up process identify err in each cell so what we do is give each slot a kind of window time we're going to we're going to get these processes to we're gonna ask this process to produce a block in preference order and so we're going to say you know 1 we're at when we arrive at a new block height the process in slot 0 let's say has 5 seconds to produce a block right and if it doesn't produce a block in 5 seconds then the process in slot 1 will have 10 seconds right from 5 to 15 seconds and if it doesn't produce a block then the process in slot 2 you know 20 seconds and so on right so it's a bit like a kind of leader failover right because we don't to run a consensus between lots and lots of different processes this is probabilistic it's messy but it kind of converges and if obviously you know it's within five seconds and slot one has been naughty and tried to broadcast its block after four seconds other processes will just ignore that block and not relaying and in addition to giving these slots a different time window we also assign a some number of points to any block they produce so if a process in slot zero produces a block we assign that block one point if a process in slot 1 produces a block we assign at half a point if a process in slot two producers have locked we assign it a quarter of a point and so on the decay and the points actually is you know you couldn't right then you can change it and get different trade-offs so here that we've got some Q Q equals more point five so every time you go down a slot the the value of a block from that slot is is halved right so what might this look like how does it help us converge so here we have two two chains okay so we start off both chains are routed in slot zero at block height H and then there's a you know there's a fork block out h plus one you know we're going to slot zero block and a slot one block and then a page plus two you know the chain that was previously in in in slot 1 goes to slot zero and the proof that the chain that was previously in slot 0 goes to slot 1 and so on so the interesting thing of course is that we score these slots so we can score the chains right so if you were in you know you urine slot 0 @ h plus 5 you would have to decide which chain head to reference right as your parent okay so what you do is you just run the scoring if you know how to choose between them and you would see that the orange chain is plus four points from this common ancestor and the green chain is plus two point seven five points right so you would choose the orange chain and you would create a new block that reference the orange chain yeah you can't so sure or just the network could become a synchronous and so what happens is the is if the score of the chain goes down then the windows expand and if the school went down it's like difficulty in between you'd expand the window size so for example if it if it was continuously lower you could have any in theory any window size like slot zero could have you know a 30-minute window could be naught to thirty minutes right obviously would never get that far right but yeah in principle so I mean this this thing I should I should actually caution right doesn't there's a lot of attacks and problems with this that it's more complex I'm just trying to show you that the the basic idea so you probably know about this as the key attacks are nothing at stake and if you know anything about like like the business that you know that an attacker doesn't cost your tack or anything you can just create loads of competing chains and try to the equivocation you know you can a bad a bad a faulty process might create two different versions of a block right to try to confuse the network I think will selfish mining going with ending guns third so it's a very attack so anyway I mean the actual full solution is more complex and of course the threshold groups can do more than just produce a random number they can also sign the blocks right they can actually sign the blocks themselves to kind of timestamp them and increase the consistency so I don't know this might be a bit technical if you're not familiar with with Bitcoin but there's a thing called an SPV proof like a simple payment verification proof it means that a client who's not aware of the full chain can validate a transaction that somebody else gives provides them with if they only have a 20 by root hash Merkle root hash of the current set of threshold groups right which is quite a remarkable thing so you can take a piece of software that hasn't you know been party to the full chain and if it has this root hash it's possible for another computer on the Internet to convince them that a transactions actually been executed by the chain which is a probably similar in some ways to a Bitcoin or something similar yeah yeah exactly exactly that's right so the client will just have the the hash and then you'd have a you know merkel path to that root so well know the client only has to have the 20 so the client has to get 20 byte updates and it might get that from some you know trusted party or whatever I said but the actual proof obviously is a much bigger yeah but it does mean it doesn't have to stay up to date with the chain which is useful the security properties of confirmations right you know if your transactions bear it you can calculate that from the chain score which is actually more predictable in some respects than a proof of works with proof or work chain so we've got high performance you can get fast blocks and the consistency feels very fast so just quickly how does I don't know I mean you know proof of work as Bitcoin user and aetherium has a variation on this how does it how does it compare to proof of work from a proof of work is one of the problems proof work you know proof of work really does need to perform very well to justify its existence one of the problems is just the environmental cost okay so you know motherboard just few months ago predicted that Bitcoin could consume as much electricity as the whole of Denmark by 2020 right so in three years time Bitcoin will be consuming the same amount of electricity as the whole of Denmark and this is clearly absurd okay it's clearly a ridiculous thing I've got a friends who mine Bitcoin and it you know they've got they're using dedicated power stations in Inner Mongolia and you know vast amounts of electricity are being consumed so this is genuinely a big problem with proof of work a real-world problem the current of theorem experienced with proof of work is there currently 50% of their blocks mined are empty okay and this is because proof of work produces random numbers in a platform distribution right and the consequence of that is that the sooner you publish a block in Bitcoin or etherium the greater the chance that somebody else will build on top of it right look at the greater the chance it will get a confirmation so as a consequence of this miners have discovered that it's more property to profitable to mine an empty block right rather than delaying to validate transactions in a block they're just broadcasting empty blocks to get the block reward right and they're also finding that mining on an empty block is more profitable than mining on a block with transactions again because you don't have to delay doing validation so you know theorem has this very very low what's known as a gas gas limit on its block so they can't do much computation at all and so by contrast if you look at Thresh or real a probabilistic slot protocols we've got a you know fixed slot publishing time so blocks you know release sino you you you you've got five seconds you know you can you can spend the first two or three seconds validating computation before you release the block right without risking that somebody's not going to build on top of you that that means that you know you can spend a lot of time validating transactions and corporate more transactions into blocks there are no empty blocks you can have a massive increase in the per block gas limit and consequently a massive increase in unchained throughput obviously the environmental cost of calculating the LS threshold signatures is negligible I'm like rush forward but you know this is without the group's signing blocks which makes makes the s stem incredibly consistent you get effects of finality very very quickly but I think it's just interesting to show even without that and even with you know these five-second blocks as compared to bitcoins to 10-minute blocks we actually need far less confrontational confirmations than Bitcoin so to achieve the same level of irreversibility right in for these transactions as they get buried in the in the system so just so you can see the bigger picture so we've seen how we can create a kind of chain a threshold real a chain from threshold relay itself but the system is actually crucial crucial component of you know fully scalable decentralized cloud this is this is what it looks like okay so you've got three different layers you've got a consensus layer which has some kind of threshold relay chain you have a validation there which validates the transitions to state and then you've got a storage layer which just involves lots of storage shards now the challenge with you know a stateful decentralized network I'd affinity is that even though you may have a million processes you know and tens of thousands of shards right which managed state storage actually you have to have a global system for validating you have to have a global system for validating the transitions to the state so you know transaction comes in to a shard and there needs to be a global mechanism for validating the state transition the results so anyway without get into the details you've got this top level pressure relay chain that records state route from this thing called a validation tree and the validation tree does for validation what a Merkel tree does for data like a Merkel tree makes it possible to notarize you know a vast amount unlimited amount of data with a single hash root hash right and a validation tree does something similar using single validation towers at each node the validation towers are driven by randomness they can't work without incorruptible randomness so this top-level threshold relay chain is both recording the state routes a single state route for the entire database database having exabytes of data and it's also producing the randomness that drives this global valid validation layer this validation tree and the validation Taos inside of it and then down the bottom you've got these state shards and of course processes are assigned to these state shards based on the randomness again the randomness drives the organization of this network and it also powers all of its mechanisms and this is the curious thing with a centralized networks that you can credit have a network potentially of millions of processes and you know no central point of control and it's actually randomness that you need to drive this organization assign roles and also make various mechanisms work so it's cook you just some dependence of randomness and always just skip ahead to super quick scaling we've got the single the validation tower I won't go into this now but you basically randomly assign processes to build successive levels on a tower and they validate state transitions and also state transitions further down these two easier to understand and particular interesting so one of the things we want to do is run a search engine on this infinite decentralized cloud right because clearly that's one of the most important mass-market services that we all depend on right and it's kind of fun too because you know all the old-school Bitcoin guys you know just deny anything like this could ever be possible and our objective is to create this infinite decentralized cloud where we can host these mass-market services as well as just a normal business infrastructure so you know clearly searches is a very different thing as compared with even something like uber right people expect very quick results right when they type in their search results they want to get results back very very quickly right and if you think about you know some of the systems we saw on that previous diagram where you've got a global validation there right and you've got you know NER upon their validation in a hierarchy and things like that the obvious question is on how on earth are we going to be able to provide that kind of performance right because there's so much computation going on right to make this thing hang together and so and you know the answer is well you you make the you know validation you know this is like this is a virtual computer and it's configurable right just in the same way as when you write software and see you can configure you can use different api's and so on so you make the validation configurable and you allow the validation firstly to be lazy right so when somebody performs a search result sorry when somebody performs a web search all right it does actually matter that the results that these miners are producing are validated otherwise all that's going to happen right is people are going to go cup along type in web searches right the web search is going to grants the decentralized Network and the miners are going to return the results with their own advertising inserted into the search results because the miners are untrusted right we don't here they are right they're all controlled using sort of game theoretic mechanisms so they have to the results they provide have to be validated but there's no reason for them to be validated in advance of the results came back so you know what you do have in effect is you know the search query comes in obviously we've got lots of redundant data and a centralized network so it's kind of very suited for production of search results the queries you know the query results are returned to the user immediately without any delay and then afterwards if and only if the random beacon says so the query results are validated right and if they fail the validation then the miner that produce those search results has his deposit taken away he's expelled from the network it's very easy to create absent you know proof of byzantium behavior right so you know that's an example of how the random randomness can improve the user experience it makes it possible to create this lazy validation mechanism where search results are validated after after they're returned right but miners can be kept honest because they're not earning much from these search results and they certainly weren't err won't earn much by inserting some advertising to the search results and if the randomness says that their results should be validated right and it turns out they were you know surreptitiously inserting search results they're going to lose a lot more they can lose their their process identities which probably had many thousands of dollars worth of cryptocurrency invest in the other one is is it's a more of a difficult problem that the randomness also solves is lottery is charging so you know people with these decentralized in from the clouds they have to pay something for every piece of computation albeit not very much but you have to pay if you didn't have to pay for computation obviously you could just dot the entire cloud and that also goes for search something like yeah web searches now it's a very very low value operation right however wherever money is involved wherever payments involved you need to perform the full depth of validation because although an individual operation might only involve a small amount of money if you do lots and lots of operations you know that can add up to a big fraud and a big loss so you still need to Vout invalid there anything that to the full depth anything that involves money but you know with with the randomness we can solve this again it's quite straightforward right you basically increase the for example increase the standard charge on this operation by a thousand times right and then determine that the billing is only done one in a thousand times on the direction of the randomness so 999 times the operation is free there's no there's no you know that the mining clients performed the operation and they don't get paid there's no financial transfer well the the miners can't do that so the miners in order to create a mining process you have to get a mining identity and to create that mining identity you need to deposit some quite significant amount of cryptocurrency to the computer itself to the to the to the system yeah exactly yeah so you might get away with well I mean in actual facts of course the the the mining reward wouldn't go directly to you it would just be charged by the system as a whole and distributed in different ways but you know fundamentally you know let's say for example you know you were trying to make money by inserting you know search results into a web query right yeah you could get away with it a few times but you're not gonna earn much money from that and you know if the random beacon directs that actually those results that you returned now needs to be validated right you know you can create a proof that a proof can be created that you returned invalid results right and that would be filed to the system as a transact a special kind of transaction which would basically expel you from the network and would mean that you're mining identity was thereafter useless basically so now that's kind of it I mean there's another thing with fire I'm the one one thing I'll mention to just quickly is without gains of the we have another system which I think illustrates you know quite how powerful randomness can be we've got a system that can run on Definity or a system like aetherium that produces a stable crits stable cryptocurrency in the same way the fiat currency is created by giving out loans right so it creates this thing called Phi ok and fire is basically backed by loan collateral so you know if you have and it's you have a different Phi for every currency area so if you have you know if you had a 10,000 fire USD that 10,000 fire USD would in fact be backed by leans over people's assets like cars business cash flows personal guarantees that kind of thing so anyway it's kind of exciting thing to do and in order to make it work you have these validates that come in make deposits you know in some ways you see the actual high level system works rather like the low level network deposit right this high level application works using some of the same principles as the actual low level network protocol right you've got these validators who come in and make a security deposit to make sure they behave themselves and then these game theoretic mechanisms that control their incentives and behaviors but the computer if I can call it the computer this is actually a theorem logo uses these validators as proxies to give out loans algorithmically right if you think it's impossible how can you have right you know you're the world computer and you take some like smart contract software okay decentralize commercial bank can you stick it up there how can you have a bank that runs completely autonomously sounds impossible right how can i how can this be possible and the answer is you know game theoretic design mechanisms the same kind of things that you see in Bitcoin and the theorem and affinity and in this case though it basically uses a random sequence of validators so you know someone will come along one of these validators will come along sorry one of these validators will be approached by somebody wants a loan if the validator agrees to where they become a proposer they you know instruct the computer hey I've got a new loan proposal and I agree to it or I think you should grant it the computer instructs them to encrypt the loan proposals in a standard format to the public key of the the next validator who becomes a checker this guy you know decides whether or not that load proposal should be granted and you know it's repeated and another validations chosen up to some number of validators according to the algorithm and you know through this process anyway the the computer is able to use human validators who constrained in their economics by their deposit and various on their reputations and various other things and it's able to use just just a lot of a software you know an algorithm is able to use humans as proxies to give out loans and we've had various economists looking at this and the interesting thing is that you know the results were getting back are that this thing should be able to give out loans more judiciously than banks right more judiciously than banks and you can do all kinds of incredible things by the way with random sequence of validators that you can even have decided peals like one of them this guy says propose this I says yeah approved the loan this says this guy says no these guys are upset because the reputation goes down they can even decide the who is wrong and who is right using another random sequence right validators you can't collude because you don't know who's going to be chosen next so I think that's probably it for me in questions Keem can be realized with controversy all right now yeah it could accept the randomness isn't the randomness on aetherium is currently not as I'm manipulable and unpredictable as you'd like so I think the current you couldn't you could do it you could do it but it would it would be suboptimal but I think the you know lot of italics got this mechanism where he raps hashes and Gore and out if you're aware of it but yes I think that that would be better yeah I think so I mean you know once you've got stable currency why not you know enable people to transfer it and privacy you using as you case not proofs and yeah there's a lot of once we have stable currency there's a lot of applications like financial there's a lot of things that would have happened on the theorem had it had a stable you know good to currency on the chain that haven't you know so I think so finishes point of view is just to try and address these kind of foundational pieces of the puzzle you know creating a decentralized cloud that has much more capacity and better performance properties and so on we've introduced we've got this thing called a blockchain nervous system which is kind of decentralized governance which is driven by human controlled neurons but you know in this algorithm called a fake liquid democracy in it can it has a kind of super user access to the to the to various op codes like us instructions and the virtual machine and it can kind of do various interesting things so which one are you know address the platform layer and then we've got various things we want to do but then the next thing is you know how do you create stable currency as it turns out though you know there's there's a lot of interest around the world for systems that can compete with banks because no one's ever been able to do it before right and you know if you talk to regulators as well it's very doing if you're a Bitcoin guy and you talk to a regulator like oh you're scary right you just want to enable people to buy and sell drugs or something right and you or invest in your speculative currency and the price goes up and down like crazy but when you talk to regulators about creating systems that give out loans more judiciously than banks and can give out loans even in the down cycle right in the down part of the credit cycle you know they're they're very interested actually so I think this is the way forward and you know it's just a matter of timing julen @t may just some technical stuff yeah yeah sure sure so and so this is one of the stranger things about school and economics they're they don't teach anybody what money is right and and you know especially people coming from the technology industry don't have much of a clue and actually many economists don't much for clue and that's I think why people call Bitcoin a crypto current currency right and you often hear this is often repeated thing that you know bitcoins backed by this 21 million limit and fiat currencies aren't backed by anything this is a myth that's the first myth fiat currencies are backed by the assets of the economy right they it's basically an IOU right and it's also not created by central bank's that's the other part of the myth right so cryptocurrency is sorry what cryptocurrency fiat currency is actually created by commercial banks so you know when you go to a bank and you want to loan r ght the bank rates the money out of thin air it's not coming from savers it's not coming from the central bank if you want it like if you go to the bankers that want a million dollars to buy a house in Palo Alto right they literally just go create the million dollars give it to you and on their balance sheet right the loan is an asset that one mortgage is an asset right and the money is a liability so it's the same the car I want you know ten thousand dollars for a car they just flip some bits in the database they create the ten thousand dollars from thin air give you the $10,000 and on their balance sheet the loan is an asset right because it's secured against your personal guarantee right or your property or the car or lots of things right and the money that $10,000 they credit from fin as a liability so natural fact the that the loan will probably be on the balance sheet is something like you know eleven thousand dollars because you've got to pay back the money with interest right and that's why of course the bank's get terribly carried away so if you look at like the you know 2008 disaster you know what's happening where the banks are saying right you know a subprime you want to buy a house of course here's a million dollars on their balance sheet it's recorded as a million dollar liability but the loan is is perhaps recorded as a 1.1 million dollar asset right and so if you just give out loads of loans on your balance sheet it appears like you make it made tons and tons of money right you can pay yourself huge bonuses until the Tillet problems occur and then you have to mark all the assets down maybe the asset isn't worth 1.1 million dollars is worth a hundred thousand whoops were insolvent the government has to step in right but anyway I mean the mechanism is cells pretty good and you know so what you have is that the money is backed by loan collateral every you know unit of fiat currency is essentially backed by this aggregate of you know personal guarantees business cash flows assets like houses and cars right and so that's why it's that's why it's stable right and so based on yeah but I'm even an unsecured line in a sense you know you've got some kind of force right you can send that you know you can send them at the men with guns the bailiffs after whoever is the money right so you know money is money has lent out with varying degrees of security then you can often grade loans A to G and if it's a high-risk loan you know it's going to be like a G or something and consequently you're gonna charge a higher rate of interest right so on a basket of these loans you're expecting that yes that will default but the interest returns and on those other ones what will make up for it so yeah so I mean to the basic idea fires just to give out loans algorithmically right and create stable currency backed by loan collateral and so it's exactly the same mechanism right and forgetting how it's done for a second this is actually Tina pointed out the other day there's a better way of doing this I mean imagine the so it's give the basic balance because you want to have this is stable right so imagine that you I'm the computer right I'm the computer and you come to me and you wanna loan for $10,000 to buy a car right and you know I use my network of validators to decide whether to give you the loan so I'm gonna give the loan and the validators have a leak cannon because they're on the hook to me they they have a legal relationship with you too that they can invoke if you don't pay the computer back right and so now you know I've got to procreate I don't do the same thing as a commercial bank I want to create the money out of thin air so I do the same thing right so you're gonna bore at $10,000 Phi right ten thousand five USD right so I create eleven thousand fire USD I'm gonna give you ten thousand and I'm gonna give the other thousand just say to the validators that's the interest that you're expected to pay right and now you've got to pay back eleven thousand dollars right you bought at ten thousand we've got to pay back eleven so you know you are on the hook for $11,000 you gonna pay it back and I've created $11,000 or Phi right and you know whenever you pay that fire back I'm just gonna burn it it's just going to disappear because the fires like an IR you it's always backed by NASA in this case it's your personal guarantee or some property you'd have or whatever right so you can see that you know you know you're a lone receiver right so you the computer gave you a loan right in Phi $10,000 but also some other guys the validators got another thousand dollars of Phi right and they can take that fine they can put it for sale on the exchange right in order to get dollars because you want to go and buy a car and the car dealership doesn't accept five dollars they only accepts real dollars right so you basically get the five put the file in exchange you get the real dollars so you know in total eleven thousand dollars has been or ten thousand but the point is you've got to pay back eleven so over time over the next ten months at eleven hundred a month or something you've got to send money to the crypto currency exchange receive Phi and then pay that for computer now all that's done for you of course in practice that would all be done for you right behind a web interface and you know you you wouldn't necessarily be doing this yourself but that's the mechanism but the point is that there's a balance right that and then the computer does various other things too it does market making and it'll always sell a fire one dollar one cent for the certain kinds of asset like digits and they will always buy buy back from you for 99 cents so crazy like an out of spread but but fundamentally the flows of capital are balanced right there might be spikes in demand supply but fundamentally they're balanced because you've you know you've wanted to borrow ten thousand you've got to pay back eleven thousand eleven thousand dollars worth of fires been created right that eleven thousand dollars worth of I can go to the crypts of currency exchange and over time you'd have to buy it all back kind of thing there's a bit more complex than that but that's the fundamental fundamentals of it so it's fundamentally it's the idea that there will always have to be after you've sold yeah yeah yeah wouldn't essentially say that so in the short term you have that the computer itself does market making for certain assets so then the computer will always sell you five for one dollar one set for example and buy a fight from me from from $0.99 right in exchange for something you can just so it creates an outer spread so the actual market makers on this exchange isn't really like a normal exchange it's not speculative exchange you know there's gonna be like a fixed spread and you know the money's gonna pile up and be sold at a fixed price basically and be bought at a fixed price so exchanges perhaps the wrong word it's more like a and interchange service because there's a you know with what ever you always know the money's backed by assets and cash flows and things like that frame and you know in fact if you if you bought ten thousand right in effect there's gonna be you're going to need great assets to do that right you know that ten thousand thousand you gonna pay back but that's secured against probably more than eleven thousand dollars worth right maybe it's secured on the car or something like that and in addition to your personal guarantee and something else so you know that's basically the same as how Fiat works with the difference that the the randomness ensures that the loan origination project process is purely rational so the loans are given out in a far more rational basis than a bank you know makes them and so you know the advantage of a system like this is you know you look around Palo Alto there's a hundred plus bank branches right it's really expensive you know these bank buildings right and you know you all those staff compliance people and then you know this system just deletes all the bank branches deletes all this Dolf deletes all that stuff right and yet it can give out loans more judiciously and you can imagine in a future world people might go and meet validators in a virtual environment you might have it on your phone or you have a VR headset or something right you wouldn't go into a bank branch you just had like some software that were connected to a validator in a virtual environment that's not even difficult to do don't need all this stuff right but the big thing is that you know even with all this stuff even with all these branches and people banks just aren't very good at giving out loans you know and they the consequences of this are that um the consequences of this or that you know you get the credit cycle is exactly exacerbated which causes a you know huge harm to individuals in the economy yeah so banks are a very problematic thing they cost a lot of money because at the end of the day obviously all those bank branches and people in those banks right I mean someone's paying for them we're paying for them right right you know at the end of the day even though they have ways of hiding the expense banks are extremely expensive I heard some estimates almost 7% of GDP goes on financial services right it's extraordinary so much money has just been wasted on this stuff so the first thing we do is also cut that cost out of the system secondly you need to protect our economies because you know the credit cycle is massively exacerbated by injudicious irrational loan making in commercial banks and this causes all kinds of individuals lots of harm and thirdly banks are becoming a kind of civil liberties issue right because they've seen do all kinds of things including lots of people in the cryptocurrency industry who've been you know had bank accounts closed down right because they don't like I think the fair is right perhaps they'll want to close down banks I don't know but when you think about the whole world being digitized right you know money's becoming digital and a bank closes your bank account right it's a very serious situation actually you know and you know you can't necessarily just can't buy a flight right try and book a flight without a bank account right possible so banks you know ravenously ruthlessly profit driven and yet they hold this special role in society which they're now beginning to views by attacking competitors and various other people while and goeth whole picture but you know they're totally incompetent in the UK we've just had Tesco Bank are just in the UK recently and I couldn't believe my eyes a 20,000 customers that have money stolen from their accounts overnight and often they'd have their accounts emptied out 20,000 people and have their account center down in some hack of Tesco Bank and so other people couldn't pay for that fuel to get to work and their rent and things like that so they contacted Tesco and they said look at us cos we're very sorry we're addressing the hack as quickly as we can it will take several weeks probably to sort it out and but but as a courtesy to our customers we're offering 25 pounds free in your account to help you get through right which would you know probably pay the the the gas you know you to doing a couple of journeys to work or something right you know and that's that's what you get 25 pounds to last few weeks then of course in in DLS we've had Wells Fargo committing a perpetrating an absolutely horrendous fraud against you know American citizens five thousand three hundred five thousand three hundred Wells Fargo staff are being sacked right because they created five five million plus fake accounts to defraud Wells Fargo customers so Wells Fargo customers weren't noticing they're getting all these fees right they were coming from these shadow accounts that these Wells Fargo Stafford so I mean you know you combine the fact that you know they're closing down the the banks are closing down the accounts of you know basically computer scientists who work and essentially space right which I absolutely horrendous right one of the most horrendous things that probably ever happen it's like McCarthy type stuff they're damaging the economy I forget what it is but it's in the trillions of dollars was lost average people lost trillions of dollars in the 2008 disaster they're generally unsettling the economy on and you know in lowering GDP and things like that but exacerbating the credit cycle they're extremely expensive right I think that's that's where you know this but this industry is very very interesting because you can the technology although it's taking longer than people would like to do it but the technology is potentially very very disruptive and it can you know dress problems with industries that just haven't had any competition there are no startup bags right there are no startup banks right I mean we all know it's incredibly profitable right why are there no startup banks and now finally we can begin to see a way that we can actually completely obliterate them not using technology and just cutting out all the problems that exist for society but also just the costs right so that's one thing one of the reasons this kind of starts exciting