J1 SAY 06 Trent McConaghy Ocean Protocol Towards a Practice of Token Engineering

all right good afternoon everyone hi well I'll get started so today by the way I'm Trent nice to meet you all it's really a pleasure to be here at HCC to join in and have a conversation with with all of you folks about about what we're up to but also improving the overall ecosystem and my talk today is actually about that based on some experiences that I've had and my team has had in the last year with respect to token design and so let's get started my title is towards a practice of token engineering you might ask why there are jellyfish all over my slides and this one I just think jellyfish are cool frankly if an AI could be an animal it would probably be a jellyfish that's really the main reason so let's get started maybe's there we go so this key is networking all right technology so let's get started it's a bit of motivating slides around data and incentives and of course more jellyfish back in the late 90s I had my very first job doing professional AI research and I was overjoyed right people were paying money to basically play around with their own networks and evolutionary algorithms all these sorts of things and I was given a data set of 20,000 data points and I was told go and classify between these 13 different classes and it was audio radar data so there was um it was data from birds flying around and tanks driving around and so on and the AI model had to classify between these and I I iterated and iterated and initially my initial models had 55 percent accuracy that is 45 percent error it was a summer job so at the end of the summer after four months of toil working 12 hour days I was happy to do that because it was AI and I loved it I got it up to 65 percent accuracy so within one summer of work I improved the accuracy by 10 percent that was still 35 percent error which is unusable and that was actually kind of the practice of how AI research was done in the 90s and even the early 2000s fixed size data sets and do what you can to try to come up with fancy algorithms to improve the accuracy right now something happened though there was a paper that came out of Microsoft in 2001 and this image is from the paper and what it shows is something that is frankly super embarrassing to anyone who likes to design AI algorithms what it shows is on the x axis is the number of data points basically number of words the Y accurate the y axis is accuracy and basically for years and years and years we a AI researchers were on the left axis we had a fixed size data set we were right up there we had a fixed size data set and we were competing over 5% error or 2% error and all of our algorithms were not performing so there was very little AI that was actually shipping but what these researchers said was you know what instead of changing the algorithms let's just add more data and not just 10% more data 10x more data one hundred a thousand X more data guess what as more data was added the models of course got better and better and better so as you see we went from 75% accuracy all the way up to 95 97% accuracy and the interesting thing was it didn't really matter which algorithm you had in fact the simpler algorithms perform slightly better this is super embarrassing because people who have PhDs in AI algorithms suddenly it's like it doesn't matter Google picked up on this in the mid-2000s they wrote a paper called the unreasonable effectiveness of data where it referenced this work and a few other ones and their modus operandi after that was go around and gather as much data as you can what do they do they went and bought satellite companies they went and bought mapping companies all of this to gather more data more data more data they call themselves an AI company that's a head-fake they are a data company through and through now why do they care about that well if you have more data then you have better accuracy as we saw and with that if you're a Google otherwise then you can sell more ads right higher click-through rate all of that right so and you get this data network effect of more data more data more accuracy more money and it's hard for people anyone else to catch up you know Microsoft has been chasing Google with Bing for more than a decade unsuccessfully so there's sort of this default incentive to hoard the data you don't want to share with other people because then their models will get accurate too right that's kind of the core problem and so we have all these data silos now everywhere and the planet and data itself is considered the world's most valuable resource this is from an economists article a york so ago and it all comes down to incentives right Charlie Munger who worked with Warren Buffett for decades with one of the world's most successful investment funds says show me the incentive and I will show you the outcome right so the question is what can we do about the incentives here what can we do about moving away from this incentive of siloing data well block chains are a funny thing they are incentive machines what you can do is you can get people to do stuff by rewarding them with tokens magic internet money right this is a superpower you know there's lots of other cool benefits of blockchain providence and some more contracts and all this but to me the most powerful benefit of all is getting people to do stuff so then going back to the problem of data let's change the incentives using an incentive machine a blockchain and let's change the incentives such that people want to pool the data rather than simply siloing it they get benefit from that and that is the underlying goal of what we're up to with Ocean protocol so we started working on ocean protocol and I'm gonna describe a few early iterations that we had and some pain that we had and then what we did about it so like all of you I'm sure we love to whiteboard so we were drawing out this is an example of an early algorithm that we had trying to address some concerns around you know if you have 10 people contributing 10 data sets how does each data set contributor get paid can you defer the payment till the end royalty style all of that so this is an example algorithm that uses Emily I to extract impacts and there's other things here too we've started to list a homework encryption and a whole bunch of other things right so lots and lots of whiteboarding bring in friends from all around the world to iterate with and so on also very early on towards the first talk of the day we wrote down goals and two of our main goals were a healthy ecosystem not just initially but long-term right and then also all throughout this maintained ethics and values right that it was always absolutely critical to us and we kept iterating iterating so this is that whiteboard expanded a whole bunch of other things you don't have to try to read my writing here it's terrible and this is also other people's writing so iterating iterating right looking at multi-party compute and and all this sort of thing - and also looking at various pricing schemes with auctions asking ourselves you know what if we had data that looked like Steam it what if we approached data the way that numr I doesn't accept cetera so just lots and lots of iterations here and there right and various sorts of designs and as time went on we started to actually distill a set of questions right so here on the top part you see how do we ensure that the data supplier gets paid what about the problem of free writing and so on and on the bottom of this whiteboard you'll see on the Left we're starting to list different signals I'm an electrical engineer by trade and in the world of electoral engineering you have signals in terms of voltage or and current and then there's other ways to view it and then we started to write down different tools and patterns that we might use this is the bottom right so this was we were starting to get there but the challenge was it still felt like we were flailing right what is it that we were trying to achieve when did it when did we know when we were done our token design right when should we move on from that sort of that first way paper to actually writing code and so we asked yourself is there a way to structure this better to look at this and improve the process and around the same time I started to realize tokenize ecosystems are a lot like evolutionary algorithms and I had spent a lot of time in the world of AI including evolutionary algorithms roughly speaking and evolutionary algorithm or EA is survival of the fittest right you can have a bunch of individuals running around running around doing whatever they want and you every generation maybe you kill off five percent or 50 percent the ones doing the worst and you let the other ones make babies have some mutations and cross over some changes and you keep repeating repeating and repeating and over time you converge towards an answer so that's the general idea of evolutionary algorithms they go back to the 60s independently invented in three different places and they're actually wildly powerful and more as long as their friend to by the way so I realized that tokenize ecosystems were a lot like evolutionary algorithms it wasn't just myself either someone earlier VA sort of to notice this too and some others so this this table here shows you very explicit examples so when you've got a tokenize ecosystem you've got this block reward function guess what that's just like the fitness function or objective function of an evolutionary algorithm so just like you're trying to say minimize the error of training a neural network in the case of Bitcoin you're trying to maximize the hash rate for security and then when you go to measure this thing how well is it doing and an evolutionary algorithm you're measuring the fitness right you know what's the error in the case of of tokenize ecosystem it's really the proof proof that something was done some sort of proof of work for example right you've got system agents right so in evolutionary algorithms you've got these individuals they run around there's 50 of them or 5,000 of them and they do their things and you keep the good ones to get rid of the bad well with the world Cheng's this is actually humans and actors the miners the token holders and so on right they're running around they're doing their thing and there's even a system clock right and evolutionary algorithms you have a generational update you know every generation you have you make babies you evaluate etc and actually in watchings we generally have block reward intervals right every 10 minutes for Bitcoin and so on and finally and this is the key thing or another thing and the evolutionary algorithm for incentives we can't control these individuals right they can mutate they can crossover they can do whatever they want right but what we can do is reward them for acting well so let them make babies reproduce and so on or the ones that are doing badly you just kill them off right and it's amazing with an evolutionary algorithm even if you only kill off the worst 5% every generation you get incredibly good performance right well we can actually have similar things in the world of tokenized ecosystems what we do have some new things we can reward by giving tokens and we can punish by things like slashing steak or kicking out of a registry there's other ways so rewards and incentives and disincentives so when we realize this we realize then if these things are so similar let's approach token design as optimization design but that of course begs the question what does optimization design look like so luckily enough I'd been doing this for a long time designing evolutionary algorithms and CAD tools that had to embed these things to just work right and there's three steps if you want to design an optimizer that optimizes successfully so the first thing is you write down the problem you formulate the problem this is the the goals which is objectives and constraints and it's the design space I'll talk about each then you try and existing selva you don't want to reinvent the wheel right so you try it and if needed you try different problem problem formulations or solvers I'll talk about that and then only if you needed do you design a new solver right because otherwise you're just wasting time so this was the steps so formulating an objective function and constraints this is the canonical formulation up here the top thing it says minimize fi I equals 1 to NF these are the objectives the things you want to minimize or maximize if you're designing a car you might want to maximize speed you might want to minimize fuel consumption and then subject to constraints inequality or equality constraints so any inequality constraint might say power consumption has to be less than 10 milliwatts or the Equality constraint might be if you've got a car maybe no matter what the tire diameter is 17 inches and then you also design a design you have a design space sort of where the optimizer can run around right so this is formulating the problem there is an arch to doing this it's not easy and actually interestingly and the next step as you're trying the different solvers if the solver doesn't converge if it doesn't do a good job you have some options you can try different solvers but you can also reformulate the optimization problem so in the world of optimization for example in the late 90s people figured out how to solve convex problems in polynomial time this was a big breakthrough but because before that we could only solve linear problems in polynomial time convex means one big hill but before that they didn't have a way to do it so that was really great because there was this big push by people to try to figure out how do you formulate a whole bunch of problems in convex form so you could solve them very quickly so or maybe your silver doesn't work very well so you try to use another one right maybe instead of an evolutionary algorithm they use two a cast of gradient descent or something and if it does well in the end it converges so this is an example here of a vanilla evolutionary algorithm running over five hundred generations trying to minimize its it's objective function and if it turns out that you can't solve it it just doesn't work then you need to design a new solver right so this is an example of a solver that is I designed in a paper from about 10 years ago and it just happened to be addressing the problem that I needed to solve there at the at the time there was no really good way to have something that could search through a grammatically constrained space in a way that wouldn't get stuck easily so this the solver basically did that as an example and then if you're successful um at the end of your optimization run or runs you get really cool results so this is an example where you have two objectives maximize the gain bandwidth and minimize power so the ideal thing is down on the bottom right the ideal point but there's always trade-offs so in this case every single dot you see is a different result that worked but this is an example of an optimization or three optimization runs that worked really really well right so if you are successful in the design of these optimizers you get good results so back to token design remember we realized token design looks like optimization design I've laid out what does optimization design look like so let's apply the same steps roughly speaking - token design what are they again formulate the problem try an existing pattern not solver and design new patterns and a pattern is a building block and they can be various levels of the hierarchy I'll talk about that so these are the three steps what does this look like for token land let's go through them so formulating the problem right in this case we have to first we'll ask you're trying to come up with the objectives and constraints right ask who are your stakeholders right what do each of them want and also what are the different attack vectors because every single attack vector is going to turn into a constraint do I prevent this yes or no right and then from that you can translate these things into objectives and constraints I will give examples in this later for ocean by the way and then trying existing patterns right so we actually were enumerated our own lists of patterns that are out there and there's actually quite a few emerging some people are calling these crypto economic primitives I like the root-patterns because it goes back to design patterns in software or architecture so some examples there's a bunch of patterns for curation there's patterns for for proofs of compute work etc there's also patterns for identity for reputation for governance for third-party arbitration and more right and within each of these you have to ask you know is this appropriate for me sometimes they might form the core of your system or they might be Bolton's that you have on the side towards addressing a particular attack director for example let's drill in on a couple so curation we've seen in the last year even the last six months that toking created registries just took off right eath denver something like a third of the the projects used TC RS great you know that's very very cool because if it works that's all you need right but it doesn't just stop at TC ARS right you can actually say what if we want to promote people just like people going from early Wikipedia contributors to higher and higher levels of editors right you can go from you can get promoted and there's something out there called steak machines that does that so it's it's a generalization of TC RS for that or you can also have continuous valued membership right this is curation markets right and they're characterized by a blending curve and you can even go further you could actually have hierarchical membership and this is something we realize you can actually introduce labels which are very popular in any I actually something that I'll talk about later is proof - curation markets so with ocean it turned out that none of these other patterns quite worked for what we needed so we actually did use we had to invent a new pattern but it's something that can get deployed elsewhere I'll talk more about that and more recently some folks introduce something called refundable tokens which is taking unfindable tokens and adding fungibility so this is an example of just emerging patterns for just curation right but of course there's other things - such as the patterns for proofs of work and within a subset of that there's proofs of compute work so we went in and actually enumerated these right subsets there's work around data and there's work around computation itself within data we've got for example proof of space-time which is famous within file coin but there's also a proof of replication proof of data availability which is a subset of proof of space-time and more same thing for computation there was talked about in earlier talks on this things like zero knowledge proof and there's subsets of that and seek' Starks is he cased arcs or interactive proofs like homomorphic encryption and so on and so forth so there's a whole bunch of patterns out there and it's really useful to understand these to have these in your toolbox when you're doing token design so with this flow these these three steps I talked about let's start analyzing that's that's use this in a case study to look at a little bit of Bitcoin and then on the heels I'm going to drill into a full line flow for how we went about a systematic approach to designing so Bitcoin Bitcoin actually has an objective function its objective function is to maximize the security of the Bitcoin network right it defines security as hash rate which is compute power which is electricity right and overall it's making it super expensive in terms of electricity spend it's super expensive to roll back changes to the transaction log that's basically what bitcoin is trying to do right you can even actually formulate it right it is an objective function like I have here where the expected value of the block rewards ie your chance of getting Bitcoin in this case it is proportional to the hash power of your actor which is your contribution to security your work function and times the token dispensed at that block in that time interval right so this is actually the Bitcoin reward function expresses an objective function right and I say expected value on the left because it's it is out over time right if you actually want to have it very low variance you join a mining pool at the top but overall you know it's very lumpy otherwise at the lowest level how well did it work bitcoins in incentives well this is a picture of a mining Center right I'm basic farm so people are maximizing security that is people are maximizing electricity and the sheer power of incentives of Bitcoin it's going to use more it's on track to use more power than USA by mid 2019 isn't that incredible it's already using more power than most small countries right but by mid 2019 more than all of USA and this is by the way a for better or for worse thing right there's big debates about this but it shows the sheer power of incentives but also when you're designing these things think about the ethics in the values and what might come out of this right let's now talk about ocean and the going through ocean design as a case study so first of all formulating the problem we actually wrote down the list of the different stakeholders so they are the people with the data the service providers the custodians all of that the people doing the referring slash curating and this by the way it could be data exchanges and so on to those people verifying the data making sure that it's good that it's not failing and so on there's people consuming the data and finally there's the keepers or the miners right and each one of them has gives and gets right so for example the data service providers well they're providing the data what do they get in return well probably some form of tokens ocean tokens etc etc etc so for each of these fake holders they have they're given and get this is just the key stakeholders by the way we live we have another ten but most of them actually aren't as especially in the core of the system right so you have to remember there's things that will be on chain and an off chain and there can be a whole ecosystem around thinking about Bitcoin right at the core of Bitcoin it's super simple but this whole ecosystem that's formed around Bitcoin with wallets and miners and pools and exchanges in all of us right so it's really important to understand that and a key part of this is where do you want to be where should your stakeholders be playing a role in the core network so in terms of formulating the problem once you've got the stakeholders then it's translating into objectives and constraints right so for ocean it actually came down to a phrase that we repeat to ourselves again and again and again maximize the supply of relevant data right and you'll see shortly that generalizes services too so overall we wanted at first reward token rewards if you supply relevant data the thing is how do you know what relevant data is so very quickly we realized we have to leave that to the crowd right so the crowd actually has to play a curation function on the data itself so really it's token rewards if you supply data and curate it and by supply I don't mean just published but I mean also make it available so and that was sort of the high level objective but we had a whole bunch of constraints and at the highest level we distilled into seven constraints so for priced data is there an incentive for supplying more data is there an incentive for referring is there good spam prevention right because you don't want to pay people every time they submit data even if it's junk data that would just that's garbage right so that's for price data but with ocean we also want to incentivize the data Commons right so for free data the data Commons is their incentive for supplying more is their incentive referring is there good spam prevention right besides that we also want to incentivize for actual actors in the ecosystem rather than just the Hodler right so what this means in the words of Fredersen does the token give higher marginal value to the users of the network versus external investors right or another way of framing or a specific way is does the return on capital increase as stake increases right and then other things are people actually incentivize to run keeper's is it simple or as simple as possible is onboarding low friction so these are the ski constraints or checklist we had but then beyond that we actually just brainstormed and brainstormed and group the ideas so you know what other objectives do we have towards good acting via staking identity reputation and so on what are the list of attacks we actually have three pages of attack that we attacks that we've written down so so we just page after page of attack others have made checklist to willing Maguire made a checklist more sort of token economics related but it turned out to be quite useful too so just long checklist and finally of course the key values so we actually wrote down the key things we wanted there to this really matters to us right so it's really important for the unlocking of data and the human right to personal data privacy and consent but also the spread of power on the spread of value right we've seen some some token lunches out there where one actor got twenty or thirty or forty percent of the initial tokens offered and that does not spread the value right so what can you do about that and then some other things to things like resilience work with the law etc so that was the step one right formulate the the the problem right the objectives two constraints the design space and then try summing sixteen patterns as you guys saw there's a lot of patterns out there a subsets of subsets so here's some ones we looked at right take a TCR token created registry for just actors right the good players vs. the bad so only the good ones are in there and some some systems out there doing a nice job ad chain I really respect them for having just that on chain and everything else is off chain right that's pretty cool so we asked could we keep could it be as simple as that right or what about just for data right or what if we had a registry for actors and for data just that right and maybe bring in some curation market stuff right so this was some initial patterns that we looked at and of course what do you do with these right how do you know whether they're good or not well guess what we have our objectives and constraints right so here we actually these key questions we asked for the price data is there an incentive for supplying or referring right well for that first design the actors registry it actually doesn't there's no incentive it doesn't help they're right for that first design the actors registry it actually does an okay job and span prevention it doesn't help for supplying free data as well so it's failing on some things passing in others right and so each of these design you can see there's some passes but there's some fails there's some X's or so so right so of all the different sort of patterns that we looked at in issue against these constraints they weren't working they weren't working so we started saying what other new patterns can you introduce what minimal changes can we make in order to address our concerns and we we started realizing that if we bound together the actual work of making data available with curation that's a new pattern and it's very powerful because it's combining a prediction of something having relevance with something actually being relevant or popular and we had a couple of variants of this and the variant number six is what we're calling a proof to curation market and so then against the constraints we see variant number six is actually hits the checklist for every single one of these goals once we got through this then we went through the longer checklists and it turns out to do a nice job there too after some tweaks here and there and here and there right so this is actually how we arrived at this this particular design was actually initially flailing planning and finally then formulating the problem very tightly having key questions to ask and then trying different patterns and iterating coming down to I showed you the objective function for Bitcoin this is the one for ocean actually and what this is is we're trying to reward curating of data and making it available right so on the left is the expected reward for the I user on dataset J and then the yellow is the prom a reward proportion the reward is proportional to predicted popularity which is how much stake you have in the curation market are you putting your money where your mouth is but on its own curation markets that can flail if it's not tied to actual work depending on the application so we tied the predicted popularity with the proofed popularity which is actually did you make the data set available when asked right so and that's the yellow the green here sorry by the way the log stuff is in there in order to avoid stake whales you don't want the rich get richer in a strong way so log actually turns out it folks use it all the time for diversity and it actually has relevance with something called a Kelly criterion in economics so that's the core of it and then we've got the tokens during the integral and then the far right there's this R thing and this is actually an extra term we have to prevent a couple attacks that we tacked on later but overall the core is for ocean is did you predict did you put your money where your mouth is for something being relevant and then was it actually relevant right that's the core but we had one more constraint if you saw on the checklist before and that was what about privacy right and it turns out that for certain problems the only way to address privacy properly is on premise computer right also though as we were iterating a lot of data is super heavy you can't actually you don't it's a petabyte or more and you really don't want to move it so that's another reason to have on premise computer we also realized there's a lot of emerging networks out there for for data for computer and so on is there a way where we can hook into that and basically grow together as an ecosystem together and what we realized was we could generalize what we've been building rather than just maximizing the supply of relevant data it could be relevant services which includes data and compute for our case tuned towards AI and so overall then the objective function simply changes from instead of predicted popularity of the data it just data it's for the service and same thing for the proof to popularity and like you saw before we actually have proofs for not just data but for compute right so we knew we could leverage those patterns so the core design of ocean did not change that way but by generalizing in just a specific way I'd address these new challenges so overall as I was iterating in this just in the last few weeks actually we've started to realize that does a new label needed for this right so out there there's actually a field that's been around for forever there's game theory which is really about analysis of tokenize ecosystems or and other things and then but there's a synthesis side too which is mechanism design right the thing is a lot of the mechanism designers and stuff out there this is really people in this science not engineering discipline of economics and how often do you get a chance to actually deploy an economy it's pretty rare right but if you think about it optimization design people are deploying optimizers all the time right there's actually a big practice around that how do they relate well it turns out that if you take ideas from mechanism design that framing and add just a few constraints it actually casts into the problem of optimization design so this is very cool these fields directly relate right but I also saw that like where is the engineering theory practice tools right where's the responsibilities that you take on if you're an engineer like Canadian engineers we get iron rings so that we always remember and if you start doing this then you take the optimization design framing and it actually turns into what I hope will emerge as its own engineering discipline engineering discipline token engineering and this is what I see so it's sort of going from this science economics which is game theory and mechanism design to engineering which has overall engineering has this goal of building systems that just work so to conclude I have a dream a vision of a practice of token engineering how can we get there token design looks a lot like optimization design by doing that we can actually have a process for designing these tokenize ecosystems formulating the problem objectives constraints design space trying existing patterns iterating don't invent if you don't have to right and only if needed you try a new design right iterating iterating there this helped a lot for ocean we've seen so far we haven't deployed it yet so I'm not going to make any grand claims that way and overall then token engineering is theory and practice and tools and responsibility towards the sign of tokenize ecosystems that's so good engineering thank [Applause] I thank you for the nice talk you seem to move the engineering on the talkin side a lower high level of mechanism design so the question is in what way you can engineer the mechanism design to achieve complexity of the system yeah so it's not in a lower level how I see it there is mekin InDesign is really within the field of economics which is really a science right and science is really about answering questions and gaining knowledge for Humanity right engineering is about building practical systems that work so I see that the mechanism what I have a dream of is the mechanisms of designers working very closely with this sort of emerging field and maybe in the end it will be called just mechanism design maybe just token engineering I don't know but Engineering has a lot to say in terms of the ethics and how to approach this stuff right so I see that the side by side science and engineering and then overall in terms of complexity within engineering obviously in engineering you can have wildly complex systems if you want right like a chip you know people design chips a team of 10 engineers can design a chip with 10 billion transistors in a matter of months because they have the right theory and practice and tools right I see that here we want to keep the designs right now pretty simple because we don't have very good tools and instead we want emergent complexity above right so Bitcoin itself the core design is super simple right so from very simple rules we get this complex system and that's actually what complex systems theory has a lot to say you there so in general we should be always trying to strive to keep things as simple as possible but no simpler thank you hi so here so in ocean you've taken care of the popularity with your token incentives what about quality so for example I don't know the New York Times versus BuzzFeed yeah so overall this is actually one challenge we still have we'd like to have when we say relevant right now we're measuring relevant is popular which is not ideal there's actually a lot of metrics for relevance within the world of data as well as sort of information theoretic measures to all this sort of thing we don't have a great answer there yet so overall we're leaving basically people when they're when they're betting on something are they betting on it being popular or relevant we don't know now if that data is valuable then people will want to maybe wanting to sell it so you can actually incorporate the relevance on the price side of things above that ocean itself if you've noticed actually it doesn't actually do the the pricing stuff itself that would be for data marketplaces that emerge on top ocean is just incentivizing for data whether it's priced or for Commons data either way so I guess to answer your question we don't have a great measure of relevance yet the best is just knowing that it's staking but for price data then relevance comes implicitly in one level up and if any of you guys has ideas on how to get to relevance in a better way please come talk to me one other thing too in general please come talk to me if you're interested in this idea of token engineering at all and contributing to patterns and whatnot I do hope to build community around this actually just with all of us together this is something that can really you know help the whole community question thank you so I'm wondering how much the fact that people try to get information from this creates a feedback which distorts the information so this herd behavior that people vote for things that other people vote for might create so do you think that it's it's irrelevant like it doesn't doesn't distort the measures that you're trying to yeah so that was certainly a challenge and actually that's part of the reason that we brought in the data availability so if you actually it's not just curating because if you have just a curation market with just that Cerrito that encourages a lot of pump-and-dump stuff right even if you have like cost and they're selling and stuff but it once you bind the curation with the work then the only way for people to really extract value well is they if they start serving up that data too right so that actually implicitly covers for it so forcing people to actually do work to compute to the system means that they're contributing to the system in the way that the system wants right so it this is why it's so important to actually ask yourself like what do what does the system want right and then optimize to that yeah great question Victor we're gonna need a mic thanks I can try to shout so basically popularity and all these measures are hinge on the hinge and the assumption that you know somehow somehow you can track the access to the data so the the idiot problem of of piracy so basically with digital data once you reveal it you can you can you can distribute it and make it available for others either in the secondary market or just just for the fun of it make it available so is the I mean this problem occurs in all other fields of course it's not not specific to yours but you have do you have like a hunch on like how you how you want to solve it or is it is it gonna be so does it have to be solved do you think it needs to be solved well solved as in as in somehow you have to take care that that uses of the particular set of data is kept in the system so the Dylan record its popularity so basically basically date piracy okay so actually in that I see two related questions one is you know the this concern towards proving that the data was made available and so on and that's why this is where the cryptographic proofs are really key um so actually we look forward to leveraging swarm is one of the supplying one of those proofs right as well as so so hurry up no but overall you know this I also reference you know proof of space-time and and other stuff or like more lightweight ones you know Jason Teich just wrote a paper on data availability which is a simpler problem than proof of space-time right so overall there's the proof that the data was made available and you only get paid once it's proven right on the other question of what if basically there's there's different attack vectors and you hinted at one around if people maybe get the data set and then spread it around outside of the system right yeah exactly yeah exactly so data leakage data escapes there's lots of words and so there's this to answer to that one of them is for certain types of data you can just let go and it's like okay fine it's out there yeah you have to let go and just knowing that manure you know like when iTunes took away DRM guess what nothing happened because it was convenient enough than the system was good enough that people were still incentivized to use iTunes which is actually wonderful right a wonderful result if it really matters that you don't want the data to escape this is where you can actually ocean by the way that it doesn't care where the data is stored it can be on premise behind your firewall it can be decentralized it can be centralized for super sensitive data maybe it needs to meet gdpr standards or whatever or german medical data needs to be on German soil then keep it on German soil and so if you have concerns of really strong concerns about data escapes this is the way right so ocean it's very flexible in terms of where the data is is served from as well as it's flexible on where the compute is right and that's on purpose this is why it's to this that's framing this sort of decentralized data hub of data and services on and this question is about the two different types of models are trying to kind of perpetuate one is around the data Commons which will have a different effect on the price of the ocean token compared to priced data and I think it's kind of similar to the data escapes from them if you're if you're always trying to build a data marketplace and data kind of has this property where it always tries to be free or kind of stay is you kind of creates downward pressure doesn't everything trend to free over time and and how does that affect the token yeah so I mean when one way of so this is a great question thank you and overall there's the question of there you're getting value at the level of ocean whether the data is for your price but then does that conflict with the price data when level up right and we wanted to make sure that it was the ocean itself was a unified framing for free versus price data right and this is why we arrived at this level it also means that there's ways to compensate creators directly if they have something good they stake on it early all this just like people my name Bitcoin early they're gonna get more and you know there's been some really nice articles recently on medium about like designing bonding curves around this and stuff to to really help compensate the creators sort of at the at this base level right and I actually see you know there's this classic problem a Superbrand problem of information wants to be free it also wants to be expensive right the free part is the physics of bits they want to copy copy copy the expensive part is because you want to compensate creators the cool thing is we actually have on whole new answer to that now because there's this magic internet money that gets printed that can go directly to the creators right which is actually kind of cool independent of whether the data is priced or not so we kind of have a new deal of compensating creators and you know this strikes to my very heart to in a great way going back to a scribe days compensated creators to you're related to this all then there's a question of the price versus not price data my view is ideally most things move towards being free for humanity and the price stuff gets squeezed out frankly I would love that about there are times in places where priced still makes a difference so this is kind of how I see things and I'm completely okay if price gets squeezed out great right hi I'm I'm curious to hear your thoughts about incentivizing different behaviors in in a complex ecosystem like ocean right like there's incentivizing the the value of the data itself there's incentivizing the running of the infrastructure that provides the data there's probably some kind of reputation based baked in just having you know token registry at all it's kind of a representation of of reputation so why why incentivizing all wisent advise all of those different mechanisms with the same denomination of value instead of using a few different a few different denominations yeah so great question one of the there's there's a few reasons why we chose what we did we saw actually early on that with the AI data it's really a supply-side problem you know if you talk to any data scientist any AI researcher they'll be like yes I'll take 10x 100x more dead oh please I want it yesterday right so that part is actually easy in terms of the demand or easier right the supply side is much harder because you've got these big corporations that have you know petabytes of data that they don't know what to do with and they're very antsy about letting you know because of privacy reasons or they just don't know or whatever and also even you know there's lots of different open datasets but they're not incentivized even to share their stuff right now so for ocean it actually was more about the supply side than the demand side so if you look at the objective function for ocean it's actually maximizing supply not to man and that's an explicit design on purpose now that said there's still the challenge overall of you know we've got the people that we want to do the keeping we've got the people that we wanted to be supplying and so on and so forth and one option was we looked at this design where we have a token reward if you do keeping or if you do this or if you do that and so on and we but there's attack vectors for that because it might be that people maybe it's you have 20% to the keeping in 20% of this and so on there's attack vectors there that's pretty easy to game in many ways so we said you know what we're gonna bind this all together so if you actually want to get you're running a keeper node that you're and you're curating your staking and you're supplying it's all bound together into basically one actor in the end yes there's different roles but we're binding into one actor there for all the other actors in the ecosystem they they aren't part of the token design itself however I think it's really really important one thing that lets you sort of fudge your way through this in a powerful way is making sure that all the actors have have tokens right and that way you're aligning the interests of everyone so so this is why for example with ocean as we're growing the community and stuff we're making sure that there's tens of thousands of people involved and we actually have community greater than ten thousand already around this and there will be air drops and all that sort of thing thing to it we expect so it doesn't overall that's the way of aligning all the various stakeholders over time right and ultimately the main way is simply like are they happy with the system is it working is it useful right but with the token design I think you have to be very very careful if you're gonna have token rewards for more than one role there is a bunch of attack trekkers you got to be super careful there right and we're not like keep in mind too you can have if it turns out that we want to incentivize demand-side better well maybe you have a network on top somehow right or maybe you there's a whole bunch of different possibilities for designs we're seeing the emergence of tokens with server ecosystems with two tokens right like a stable coin and something else so we're open to all this and we're just we're trying to keep it simple now but what who knows right none of this we see as cast in stone we're going to iterate iterate measure and so on I have a question about iteration actually so you you talked about I mean just in drawing the equivalence between token engineering and optimization design so when you're doing optimization design on something like a like a printed circuit or or a well-defined problem a diode or a resistor behaves in the same way and that's very easy to simulate and so you can do many many iterations but if you're trying to design a token economy people don't behave in predictable ways and so I wonder if you have any thoughts on how you iterate with economies and how you rate with a token design when you can't really do that many experiments or do that many iterations agreed so I view this as a great challenge it's a great question actually um it turns out actually circuit design well it's a closed system it's actually really hard to model well so even if you're trying to like synthesize an analog circuit or something it's actually really hard to have constraints such that it's not giving you completely wonky designs like Rube Goldberg machines and I know the super well because this was my PhD thesis and I was actually banging my head against the wall for yours trying to solve this and I cracked it eventually but so so even if ur circuits is actually hard to capture everything which is maybe a bit of a hint of the solution that said you know when you've got a more open loop system where you've got humans involved and humans you can't always predict humans because we would have to model the brains of everybody and that's just going quite crazy then what right and so it's a great question some and I've been innovating with several people about this over the last few weeks you know and there's high variance and opinions right one friend he tells me this trend this is impossible this is Fantasyland right other friends are like I'm gonna build a simulator tomorrow on this I'm gonna do my best right and I see that overall we can we should we need to try right for starters because if we don't try we end up with more things like Bitcoin you know this life-form that's eating the energy of the planet right I don't want that anymore right so if we can try it if we can anticipate these great so there's a near term solution and then a medium term the near term solution is something that we're doing and I'm seeing other groups doing this too with ocean we had at a draft away paper and then we would send it to a few friends who really get this stuff and they would attack it in various ways like what about this what would that we would add that to the list of attacks we would update the design to fix it and we kept doing this again and again and again right and we're seeing other groups doing this too you know the true beauty is the OS : guys and so on and I think that's actually a really great practice because then you're you're adding constraints at the end of the day though hopefully that's enough but I think also even when you deploy the system you might find new attacks the system itself needs to be evolve Abul enough right so the good thing is even by default like hard Forks or a path to evolving it can be painful but it's working we're seeing on chain governance maybe starting to work things like - coin and stuff great I don't know what the right answer there is but overall you know some form of governance whether it's hard for kishore something bit more involved that's gonna be crucial to actually adapting to the system I also am hopeful too but beyond just iterating with friends and stuff I do hope that we get something like a token simulator eye token economy simulator and it could be like agent-based simulation things like we've had seen before but these agents then how do you design the agents right and maybe you can turn that into a game on its own where humans get involved maybe it's Mechanical Turk style we're actually part of the simulation is actually people trying to attack in various ways and then maybe bit by bit by bit you bound heat it up towards the launch of the network itself right because you have to in send its people well enough if they come up with a great attack that they're getting paid maybe a thousand bucks for would they do that or would they wait till the system goes lives and make a million dollars right so how do you how do you use that and I think that there's a lot to be said with bounties for bugs sort of like cobalt that IO does right but more towards token bugs so there's bugs in terms of the formal verification it's more contract part but what about the token dynamics bugs right and it's like digital versus analog design so overall I see pieces right to summarize iterating with friends to attack you evolve ability and the system itself via governance and then finally token simulation itself that includes human-in-the-loop right one other thing I should mention these things here that I'm talking about I actually wrote a blog post that came out last week that discusses a lot of this in length so and like before too if any of you guys has interest in iterating more on this let's do this let's you know forms you know community around this thanks Trent first of all amazing presentation really enjoyed it so my question is I think as more and more protocols emerge that have specific use cases like like ocean for example for the for the exchange of data interoperability is going to become an increasing topic and now as we see different protocols interoperate and kind of you use each other's functionalities those sort of those protocols have their own token systems and incentives obviously how can we in instead of token engineering take like other token economies into account so like bake in the concept of token interoperability and the the incentives from one moving from one protocol to another into the actual design before even launching it is Wow if you have an answer to that please tell me or for anyone does that's that's a great question I don't know I do see that you know interoperability is actually at the heart of ocean 2 because we're trying to talk to like iterate with these other networks right so we want compute coming from Gollum or from singularity net or from enigma and we want data coming from swarm and from file coin and all these right it's actually really important and then we are we see that there's interoperability protocols emerging right inter ledger cosmos polka dot all of these true--but as a glue technology to so each of these the the near-term is we're not going to have tokens and incentivize this over all yet there's a sub challenge to if someone's using ocean and they want to make a call to say Gollum how did this let's get paid for and you know there's good the good thing is there's decentralized exchange is happening and built into things like in Fuhrer and stuff - that's actually a very nice but that's not about the incentive that's just about interoperating so I I don't know but maybe there's something around like baskets of currencies maybe right like prison style so if any of you have ideas yourself or otherwise please let's talk that's a great idea actually I think that we can learn from the kind of real company world work I think in music Spotify intense music did like a share exchange so I think seeing projects you do an active exchange of their tokens could be really interesting there isn't advice to actually build across protocols and I'm sure those things that could be quite quite interesting to explore that's a great idea yeah the sheriff swap taking a token line right and I took a network that's incentivizing here swaps across that would be amazing right so this is a very hot QA session so I will accepting more questions oh it is the last conference for this morning people have more questions great yeah I mean so anyone has other questions thank you everyone [Applause]