Day 1: Tomaso Aste (UCL Centre for Blockchain Technologies)

Transcript

[Music] ok I want to talk about incentives and emerging properties in blockchain system today and I think that this is a very very central topic which has been a little touched so far and I believe that this community of industry field and that of people that are making blockchain nowadays should consider this as one of the main part in the design of present and new blockchain systems and

in particular let let me say a little bit about emerging properties because it was not exactly common language term and I mean the example that I like typically give is the following well living beings ourselves are made of atoms molecules cells organs and the interaction among them and you know we have quite a good understanding of the things work and not perfect but pretty good actually and if we put all our understanding together and list it down and we look at carefully what we completely miss is that we are alive and inside that being alive is an emerging property it's very

obvious it's very obvious to anyone with difference with something alive as something not alive but it is emerging it to it to the extent that we are not able to replicate it even if you know everything some some small organ is we not literally know everything about them except why they are alive and well clock chains are a little bit easier than living organism but not death easier because they have to do anyway with living organism humans and therefore I mean understanding the emerging properties of this system is really crucial and can be the difference between be alive or dead ok let's start from the starting point is one starting

point which is the paper of Satoshi Matunuck Nakamoto 2009 and design it is the anonymous cash system where which rely on majority on consensus by majority and the system is anonymous and in her writing the idea is that there is a community that validate transaction and agree on the content and then eventually put in the block and the block is a shield etc etc and an important point is that the operation on verification and hashing and closing the

block is remunerated and actually Satoshi herself say look this is one CPU one vote in analogy with one person one vote of the early of the 20th century so we have some element that I put in red over there and so the idea that consensus should be done by majority that is a design constraint is something that Satoshi decided the when decided to do this this this system also the fact that user should be anonymous this is again a design constraint I mean you want to have an animal system because

you want to replicate cache which is essentially anonymous then the community the reference community again is something related to the input into the system and again it is a constraint in this particular case of Bitcoin it was addressing a wide community anyone essential and the fact that there is a specific ashing mechanism to do that that is definitely a design technology system and to some extent is relatively marginal to the rest to the whole system and then remuneration the rate again is is defined by design and that is what we call incentives okay people are

remunerated to do that because that is a service very important for the security of the system well today I will about this last point mainly but at the end I will also work up with the rest so once if you want wood well we all know I think how the story went people started indeed I mean the idea of Satoshi was literacy initially the paper is very obvious that you know you stay there with your little PC and in a while you do the transaction you'll to do some service to the others in Deventer you make a little money out of it then when people realize that actually the money were not that little they start using GPUs refuse goes about 5 times faster than CPUs and concerned consume about

1/4 of them so it's much more convenient very soon they pass 2 F GPA which is field programmable gate arrays are essentially solid-state programmable machines that goes about 100 times faster than CPU and use about 20 times less electricity and very soon all these happen in 2012 by the way very soon they start people start building ASIC application-specific integrated circuit this is this is in itself is an emerging phenomena of Bitcoin we should we should consider that at that time I mean people like Cisco and IBM Exeter when they were

deciding to build an ethical technology machine they were setting up a team or than or 10 engineers a budget of between 1 to 100 million setting a timeframe of 5 to 10 years and you know design it very very carefully and then getting this fantastic superfast machine of course those machines were doing a little bit more typically than hashing but still this happened during one year by who-knows-who and in the country spontaneous way and this is the kind of mechanism that emerged and that they are an expected certainly I mean GPUs may have been conceivable at the time where Satoshi was writing his paper but certainly not the use of Isaac and of

course that the paradigm shifted completely from one CPU from a single user to something completely different and so what this incentive incentive for mining created they created a specialized activity users are not any longer equal there is no longer one person one where one CPU one both there is a class of people that do only mining and don't do anything else and we know that this is highly concentrated in the end of three four major mining companies groups and mostly based in China while we should remember that this is very typical of technology I mean if we look

at what you know if the people old enough to remember what was in turn at the beginning into the beginning was a very open place where you could do everything people at the e-commerce bubble everybody was believing that any small business would have been able to sell his stuff directly through his own page over the Internet and nowadays we have a few very large retailers they do this business for us very efficiently why because the kanonji cost a lot of money so the needs investment and there is a scale economy that if you are larger at the end your cost is lower and people are very happy to use your services this is what happened to Bitcoin as much it happened to others this is a consequence of the design an emerging consequence I by the way well I

think that everybody knows here but at the moment I said this was three weeks ago so probably is a bit different but at the moment we are generating 30 eita ashes per second this is 32 10 to the 18th is a large number actually the fastest supercomputer has more or less this flop velocity so it's it's really impressive is 30 billions of billions ashes per second and this is done for this what is called the proof of works or the one CPU one-vote idea and let me remark it just a little bit because that is important for the following things that I want to say is that the proof of work should the cost and also cost a

fraction of the value that is translate translate translate patron acted and this fraction cannot be infinite decimal so what thirty billet thirty-eight ash are well we are talking about 1g agile we are talking about the consumption of around one gel for for jjigae ash so we are talking about 30 gigawatts of electricity consumption this is the way that the machines work which is the power plant similar to the one a picture that is one Giga ash Jiva give a watt power plant and at this

moment in these months Bitcoin is more or less consuming about between 0.5 to 5 barrel of oil per second which is a lot but actually is not that much is a cost or about 0.5 percent of the money transferred through the network 0.5 is exactly high if you've compared with other services that bank offer that cost essentially nothing of fixed fees however if you are a Chinese that want to launder money out from the country then 0.5% is a very very very low cost it cost about 30% doing it on the black

market let me remark a little bit the fact that proof-of-work should cost and I mean and I don't go through this this this slide there is a paper out but America is a small and North out but it's very simple I mean is to prevent double spending and it is this is the concept of basic security concept you know if you have a safe the cost to break the safe should be similar to the cox-2 the value that is inside the safe so the safe is safe if the cost to break it is similar to the value inside of course I mean there are always way around it but that's a very basic principle of security and in this case if for the proof-of-work is exactly the

same so at the present one block contain about 20 million u.s. dollar and people wait an average of 10 blocks let's say to get Scheuer the things are into the blockchain which means that you have to spend around 2 million per block if you want to be sure because replicated twenty ten blocks will cost you 20 million which is exactly the value inside so you cannot steal more than 20 million and to steal 20 minion it cost you 20 million that's the fuel security that will imply 10 percent cost to the approval work instead actually is much cheaper is about 1/2 a percent so what we have is that in another emerging

properties the community decided actually 10 percent is probably not worth it it will be probably is too high cost for little to some extent security risk and the economy ended up in a in a state in a in an equilibrium which is about a percent this means that the system is vulnerable but it also means that the system is cheaper I mean this is similar to I don't know these are clearly security on credit card is very very low I mean numbers goes around there and everybody you pay with especially with the paper one as your number that not analyze I mean the security cost is balanced by the fact that we have less cost of to

maintain security and less security then in the system let me see then if this mechanism again in Bitcoin is working or not ok so given that assuming the security is there and to some extent there has been no double spending attack so far is efficient for the purpose which is verifying and validating transaction and this is a study done already two years ago which in block caching time is an eternity so we don't take it as a picture of what's happening now but I mean this is an interesting study we

monitored the transaction for about one week during May 2000 exist in all transactions so we generate an order pertained it to be a - I mean we took all transaction in the system and then we look for the six months of six following months when each of those transaction appeared during that period caste didn't appear was getting into the into the blockchain and we had a surprise so we found out that although a large number of transactions do get in in the first and second blocks generated after the the transaction is broadcasted this the number after one hour is only 58 percent so 42 percent of

the transaction are not in the register in the blockchain after one hour so the time actually is not at all ten minutes is much longer and actually what it was even more surprising we check and check again the result was that actually only 80 percent ah in the blockchain after 30 day after one month so after my must still 20 percent of the transaction is missing well this was a huge surprise however then we look which transaction there were and indeed I mean as the common sense tell us they were very very small transaction this is this fraction of enough of fragmented coins that make it the dust into the system but still still 0.1 percent was not registered in the

river chain after one month and why I mean again was surprised and went to talk with some miners etc and they you know they did essentially the answer was of course I mean why should we care and we don't make money out of you know looking for transaction that we are not written in the hello chain and putting their there if people pay us fees then we might do it with some more diligence but even that is not our main source of revenue so I mean we are not really too much worried about that and that's again is an emerging properties as consequence of a system of incentives and just to

give it a full picture if we look at the transaction values is kind of obvious you see the the speed is faster when transaction are very high I mean if you've trance if you make a transaction a few millions I mean you you do make sure that is registered in the blockchain after an hour or so otherwise I mean you you start actively trying to do it or redo the transaction or whatever you want you can do and however there are some very large transactions still that way it's a long time before six hundred blocks before getting into the system and actually the surprise is that also transaction fee yes transaction fee make the if the insertion into blockchain faster but not

that fast again there is a large queue of transaction that stayed there for a very long time even if people have pay fees again remember a smaller value transaction but still are there so again and this is not this is not to some extent should not be a big surprise I mean minors are not remunerated to do this job and that is an emerging property that they don't do this job they don't do this job efficiently enough so let's go a little bit higher level and let's try to understand as designers as industry intrapreneurs that want to design a new blockchain system

or adapt existing blockchain system which are the main element well i in my from my point of view there are two fundamental elements one is the actual design and the other one is the merging properties and the actual design is a gif for the example for the Bitcoin our constraints and our inputs and I will specify them and next time next few slides and the matching properties instead are something much more complicated much more complex they are evolutive they are mechanistic change they are dynamical and they are very

hard to predict from the constituents so constraints are what you want in the system or what you need from the system and these are limit and and these these limits limits for instance the availability of incentive mechanism I am focusing on in cities well because the topic of my talk but also because this is essentially the only quantity that we have to steal the system in one direction or in the other so for instance if we have a purely anonymous blockchain we cannot only provide the economic incentives because I mean we

don't know who the user have so we cannot price user because they behave well just because they will come with a different identity every time they use the system and if we want to use trust visit the incentives then we should introduce limitation to the anonymity and even more if we want to do social kind of institution and again if we want to add an anonymous system we should burn consume external resources just as proxy for reputation and resources or put large amount of money on stake and

centralized Authority can booster up the reputation system however this is to the price of anonymity and but what is very very important is that any property any property that we are trying to to establish in the system is either a an emerging spontaneously and endogenously emerging property from the system or that property will not be stable and will eventually disappear even if you see it from the beginning it will go away with the dynamics and this is seen in economic system all the time and what the main emerging profit

of this system well the main immeasurable sucess is cooperation by cooperation because in distributed system multi participant must contribute to the maintenance of the system this is the day the absolute necessity in distributed system and these ik and this cooperation can be am rephrasing essentially what I said before it can be obtained in two way one with enforcement and were all with remuneration or both eventually and the session there are no other ways and enforcement requires the two no participant identity I mean you don't have to know who actually they are

as individual but you have to uniquely identify participant and this also requires strong government governance body instead the remuneration emulation are economic incentives they can give be given to an animal system they can be given in completely stable system however they as I give the example because very hard to implement and in a very simple terms I mean if you give small incentives they will not be enough to persuade people people to participate to do things and if you give instead the too large then will definitely generate specialization and concentration and the fine line between the two is very very

hard to find and also typically participant don't like when people change the rules during the game and therefore there are not that many instrument to to steer the system once has been designed so to give an overview it's so what's that the input the constrain incidentally I mention already is certainly an amenity is a self social responsibility is regulation regulation certainly is a constraint is the grievance community the business model the reference market the governance model the scalability the the speed at

which you want to grow etc etc the technology protocols are the validation methodology first but not and the ographers certainly but and the security in general the speed is obviously one of the technological issues scalability again can go on this side as well and are also the incentive in incentives our economic incentive or institutional incentive or social incentive or perhaps others and which are the emerging properties weld emerging properties are certainly already mentioned specialization but there is not only specialization that is certainly

centralization and distributors distribution there is a reputation repetition is something that is generated in dodging Olinda to the system there is trust again that is generating dodging system there is failure there is failure and success I mean failure and success are emerging properties of the system and there is the forking risk which is related to failure and success and these are all in the design but cannot be determined a priori by the design our dynamic appropiate come out when people start using it and it's very very hard and not impossible very very hard to predict one from the other and give let me give you an example so if we design a system with the constraint of an amenity so full and

an eighty we have a strong constraint on the validation methodology essentially we can use only proof of work and so we can use only economic incentives and this is kind of likely to end up in specialization this is certainly what happened to be cold but is what's also up into many other coins that has been a concern system that has been proposed and if we release completely an amenity we do identify users and validators then we have other validation mechanism and certainly the instant institutional incentive and institutional punishment or status can run the system smoothly

these however unavoidably brings the system to our centralization instead of distribution you need the strong governance strong Authority that even if the system is distribute that it will be an emerging properties of the system and if we look at if we try to map I mean a few of the existing their protocols in particular Bitcoin and ripple and the Tyrian we see all this kind of element I already mentioned for Bitcoin Bitcoin is is anonymous in terms of users in synonymous also in term approval work the incentives are economic there are no social entities its end up with I and increasing specialization high consumption of

eternal external resources but it is a fairly distributed system Ripoll which is to some extent an opposite example is it is an animal for users or can be an animal users already however is not anonymous for the validators these allowed to give rewards to validators in a different way in term of reputation is social kind stitute or kind of a reputation and it can give also reward to use it to other mechanism and that this seems to work it to give a fixed level of specialization how can the low consumption of resources however it relies on trustable at least trustable validators which means generate

concentration in the general centralization and aetherium idiom again is a is anonymous observe anonymous in users and also in validators and of course as in places the smart contract in terms of rewards family incentive values economic incentive not the social incentive and it is it use it has some trick in the design to use less computational power but my argument on the cost of ruble works it's still valid so you cannot escape the cost in fact is moving toward this proof of stake

protocol and but again this end up and avoid ibly in high specialization in high consumption of attend resources except for the Provos take a choice and in a distributed them and in the case of material in high risk of forking while this was inside of design it might mean that might not necessarily be something that you don't like but this is certainly something that is an emergent property and so let's me I have a couple of minutes let me conclude with the take-home messages so we have blockchain system follows

design by design there you have some input one train and some some technological choices but the actual properties and functionality of the system are emerging properties they come from the design but they are not specified in the design and will be discovered when the system work and the main instrument to make these of to steer these emerging properties are incentives and and this is a very important part of the design and the only instrument essential that we can use after that technological choices are taken and thank you

[Applause]