Kleros and the Advantages and Limitations of Cryptoeconomics

Some notes on the Schneider and Buterin debate on governance and cryptoeconomics...

Acknowledgements and thanks to Federico Ast and Yann Aouidef for their review and feedback.

This article seeks to gives some of our perspectives on the nature of cryptoeconomics, and what it can accomplish and what its limitations are. These thoughts are particularly prompted by a recent article by Nathan Schneider and a response blog post by Vitalik Buterin that give their thoughts on this question, both of which feature Kleros in their discussions.

Note that Schneider particularly focuses on these points relating to the issue of blockchain governance, drawing on existing intellectual criticisms that excessive "economic logics can erode democracy" to argue that systems built entirely out of economic elements are intrinsically limited and that blockchain governance systems should include non-economic, "political" layers to overcome these limitations.

We will particularly focus on these ideas in regards to decentralized, blockchain-based dispute resolution systems such as Kleros, though we will also touch on more general ideas including questions around blockchain governance.

Definition of Cryptoeconomics

To begin it, it's helpful to reflect on what are the defining features of cryptoeconomics. Brekke and Alsindi define cryptoeconomics as "an interdisciplinary, emergent and experimental field that draws on ideas and concepts from economics, game theory and related disciplines in the design of peer-to-peer cryptographic systems."

Schneider, when considering definitions of cryptoeconomics, cites Coindesk, according to which cryptoeconomics is "an area of applied cryptography that takes economic incentives and economic theory into account". Schneider goes on to write on this theme, in describing Bitcoin, "the math secures the economy, which in turn motivates people to use the math".

We feel this gets to the heart of what cryptoeconomics is: namely it is a form of economics where the set of actions that are available to actors is defined by cryptography. One can sign a message or not, publish a block or not, but one cannot forge a signature corresponding to a key one does not control (without breaking a cryptographic primitive, in this case a digital signature scheme).

Then cryptoeconomic mechanism designers attempt to construct games out of cryptographic primitives where participants are incentivized to have desirable behavior.

Hence, one could even ask what are the advantages and limitations of cryptoeconomics even compared to traditional economics.

• On the one hand, you are limited to what you can build out of the cryptographic tools that are available. For example, there is a fundamental limitation in blockchain incentivization schemes that people can only be penalized to the degree that they have previously made some deposit. This contrasts with traditional court systems that can issue fines with more flexibility.

• On the other hand, many of the cryptoeconomic systems that have been developed have novel features, some of which would be challenging to reproduce in more traditional settings. For example, the ability of a contract to execute an on-chain reward in accordance with a Kleros ruling sidesteps many issues involving enforcement in traditional dispute resolution systems. This includes practical issues regarding having losing parties comply with rulings or regarding having trusted third parties to hold disputed amounts in escrow. As another example, flash loans are hard to imagine in pre-blockchain settings as they depend on the updating of who is seen as holding what amounts being programmable in a fairly rich way.

Limitations on Cryptoeconomics Coming from Limitations on Economics

Cryptoeconomics is based on an economic logic, which only makes up part of what typically motivates actual human beings in their choices. Human beings are generally not homo economicuses. To quote Schneider:

Relying on cryptoeconomic incentives limits governance to a narrow subset of the techniques that other kinds of institutions have used. Cryptoeconomics sees only a certain slice of the people involved. Concepts such as self-sacrifice, duty, and honor are bedrock features of most political and business organizations, but difficult to simulate or approximate with cryptoeconomic incentive design. Labor unions can produce economic benefit for members, but achieving and maintaining those benefits has required cultivating an “expanded community of fate” that is anything but self-interested (Levi, 2020). As Albert O. Hirschman (1970) famously showed, the most economically valuable forms of organizational loyalty often grow out of non-economic forms of relationship. When people complain that others seem to vote against their own economic interests, it should be a reminder that economic interests do not encapsulate the totality of human needs and wants (Haidt, 2012).

We would tend to agree that there are powerful human motivations that are not well-captured by economics, and then particularly not by cryptoeconomics. While economic factors are important, they are only part of what have allowed our societies to build the robust institutions that we have off-chain.

Note, nonetheless, that blockchain applications should not be exploitable by a homo economicus. Indeed, while human users may tend to make their decisions based on non-economic motivations, if there are strategies that generate a profit by following a strict economic logic, one can program a bot to follow them.

A related issue that Schneider highlights is that of Sybil attacks, where an attacker manages to create many parallel identities with which to attack a system. Schneider writes "At the root of these limitations is the blindness of cryptoeconomics to the identity and integrity of human users—a persistent, but not necessarily permanent condition."

This is an issue that is not unique to blockchain applications. One approach to address these issues, built from the tools of cryptoeconomics is the Kleros-backed Proof of Humanity system which, in addition to providing a Sybil resistant mechanism under which people are limited to a single profile, by providing a notion of "blockchain identity" could also eventually allow for people to interact with blockchains in a way that more closely resembles off-chain social interactions.

Designing a system that interacts with both humans and robots is challenging, as incentive design tends to have distinct subtleties in these settings (see for example the work of Hadfield-Menell and Hadfield). However, as interactions are increasingly online, systems that are designed to motivate (or at least withstand) both types of participants are increasingly relevant.

Cryptoeconomics and Kleros

In their discussions of how limits of cryptoeconomics apply to Kleros, Schneider writes:

Such a jury does not deliberate, does not seek a common good together; its members unite through self-interest.  

to which Vitalik responds:  

The implicit critique is clear: the Kleros court is ultimately motivated to make decisions not on the basis of their "true" correctness or incorrectness, but rather on the basis of their financial interests. If Kleros is deciding whether Biden or Trump won the 2020 election, and one Kleros juror really likes Trump, precommits to voting in his favor, and bribes other jurors to vote the same way, other jurors are likely to fall in line because of Kleros's conformity incentives: jurors are rewarded if their vote agrees with the majority vote, and penalized otherwise. The theoretical answer to this is the right to exit: if the majority of Kleros jurors vote to proclaim that Trump won the election, a minority can spin off a fork of Kleros where Biden is considered to have won, and their fork may well get a higher market price than the original.

Expanding this point, Kleros is designed so that, as much as possible, participants economic interests align with voting for their perception of the truth. In particular, various aspects of the protocol make attacks difficult, including collusion, bribes, and pre-revelation attacks (attacks where a participant precommits to a given vote before the corresponding vote reveal period). Some of the key tools that Kleros uses towards this end are the following:

• The ability to fork - as Kleros is open source software, anyone can make a copy of it. Particularly, if part of the community finds a Kleros decision highly objectionable, they can redeploy a version of Kleros where that decision is reversed and those that voted for it are appropriately penalized, giving them less weight in future decisions.

To the degree that this copy of Kleros is resultingly seen as likely to produce more just decisions, one can expect it attract future cases. Hence, even if an attacker or a cartel manages to take control of 51% of the stakes in a version of the Kleros court, they may be cut out of future decisions, with their stakes resultingly losing economic value.

As Vitalik writes: "the 'final backstop' that Kleros relies on, the right of users to fork away, itself depends on social coordination to take place - a messy and difficult institution, often derided by cryptoeconomic purists as 'proof of social media', that works precisely because public discussion has lots of informal collusion detection and prevention all over the place". Furthermore, in the upcoming v2 of Kleros there will a built-in mechanism to help the community coordinate in such situations.

• The appeal system - the appeal system is designed to make both bribes and pre-revelation attacks less effective and impractical because, if a case is resolving in a way that seems out of step with how it is viewed by the broad community, there is an incentive for participants to appeal it to some larger panel of jurors.

This larger panel is more expensive to bribe and none of the votes necessarily correspond to the attacker who pre-revealed. Critically, Kleros jurors are incentivized to be agree with the ruling of the last appeal round, and not necessarily with the other jurors in their own round.

Indeed, if one tries to collude with the other voters in her voting round, those voters are not guaranteed to be part of the final appeal round that ultimately decides the case; on some level, one wants to agree with a hypothetical group of jurors in such an ultimate round which haven't necessarily been selected yet, rendering collusion difficult.

• Juror behavior court - In the next version, we plan to have a special "juror behavior" court. Then if there is solid, but ultimately off-chain, subjective evidence that a juror violated court policies such as by revealing her vote during a commit period or accepting a bribe, third parties will be incentivized to "accuse" that juror of juror misconduct. This will allow Kleros to leverage its own ability to provide subjective judgments to provide a more nuanced enforcement of community norms that what would be available through purely automatic mechanisms.

However, these mechanisms are all essentially cryptoeconomic as these are the natural tools that are available to Kleros in a blockchain environment. We have actually seen in some experiments that even in Kleros cases, majorities of participants rejected bribes that were designed to guarantee them payoffs greater than what they could have received by voting honestly. Hence, even on the blockchain, people seem to have some notion of honor, though its effects are likely much weaker than what one might expect in a in-person, less anonymous setting.

In contrast, Vitalik writes about the non-economic tools available to real-world courts:  

Non-financialized courts, on the other hand, do prevent collusion in two key ways:

• Bribing a judge to vote in a particular way is explicitly illegal.

• The judge position itself is non-fungible. It gets awarded to specific carefully-selected individuals, and they cannot simply go and sell or reallocate their entire judging rights and salary to someone else.

The only reason why political and legal systems work is that a lot of hard thinking and work has gone on behind the scenes to insulate the decision-makers from extrinsic incentives, and punish them explicitly if they are discovered to be accepting incentives from the outside.

We agree that these are powerful tools in maintaining the integrity the traditional courts. Note, however, that the subtext of Vitalik's argument is that judges are still human beings who are motivated partially by economic factors that must be taken into account in the design of a judicial system. Indeed, researchers in the field of law and economics have been able to apply frameworks of economics to fruitfully analyze the behavior of judges, often modeling judicial utility in terms of social factors which, while non-financial, are still not necessarily perfectly aligned with "the common good".  For example, Richard Posner, himself a US appellate court judge, has analyzed the motivations of judges in terms of reputational effects and using metaphors to players of games. Furthermore, data analysis on judicial decisions, particularly including that developed by legaltech projects towards predictive tools, has shown that judges’ decisions can be influenced by a number of idiosyncratic factors, such as time of day and results of recent rulings.

Nonetheless, developing on Schneider's emphasis on the roles that "self-sacrifice, duty, and honor" play in human decision making, the "specific, carefully-selected individuals" who wind up as professional judges in well-functioning court systems tend to be people who have spent their careers in an ideological environment that instills in them the importance of the rule of law. Such judges ultimately build an important part of their self-identity around being impartial. So, faced with a given bribe and all else being equal, a judge who has to betray her idea of how she views herself to take the bribe is going to be far less likely to be corruptible.

However, while these non-economic defenses that traditional justice systems possess are powerful, they are also expensive and subtle to put into place. The process to train, select, and then correspondingly compensate judges is expensive and each judge can only handle a limited number of cases. This has lead to backlogs in civil courts in many jurisdictions and many small-scale disputes are never considered by formal dispute resolution process.

The fundamental principle of Kleros is that participants, who individually are not necessarily particularly trustworthy, can be sufficiently motivated by economic interests within a well-designed incentive system to provide useful rulings in cases that  are not well addressed by traditional dispute resolution systems.

Indeed, building such a system out of cryptoeconomic mechanisms has several advantages:

• As already mentioned above, Kleros rulings can be automatically executed on-chain removing a layer of uncertainty regarding whether losing parties will comply (or alternatively issues around finding a trustworthy third-party escrow service to hold funds during a dispute) that is present in many traditional dispute resolution services.

• The code of Kleros smart contracts, being public, sets out a transparent process that the parties to disputes know will be followed. Furthermore, Kleros jurors themselves, as they should be confident that the economic incentives that are motivating them are respected, can have confidence the processes set out by the smart contracts will be followed.

• The selection of jurors, in particular, is made according to a random number generator that anyone can verify followed clear, on-chain logic.

Schneider and Vitalik both write about the role of blockchains in creating trust. Scneider writes:  

And what became of trust? De Filippi et al. (2020) conclude that while the distributed-ledger technology underlying Bitcoin and its progeny has not fully escaped trust, it has produced a new kind of “confidence machine.” The partial shift from trust to confidence transfers governance burdens from people to technical systems, inviting that renaissance in designs for systems of governance.

Trust is particularly relevant in a dispute resolution system such as Kleros, where it is essential that the process has a certain level of credible neutrality. The trust instilled by "technical systems" may not be the same as that that one has in traditional judges selected through a rigorous process. However, this type of confidence can scale more readily to the types of small-scale disputes currently not be adequately addressed by existing systems.

Kleros as a Tool in Governance

Touching on the subject of governance specifically, the primary motivation of Schneider's article, one possibility that is currently being explored is the idea of using Kleros as a tool for "judicial review" in DAO governance.

The idea here is that DAOs can have constitutions that any governance updates must respect, and then one can use Kleros' capacity as a subjective oracle to determine whether a given governance proposal is, in fact, in line with these requirements.

The parallels of such a structure to the separations of powers in nation states evokes a comment by Schneider:

Recent cryptoeconomic practice appears to be reinventing some old wheels of institutional life. The Kleros judiciary, the board-like Graph Council, the constitutionalism of 1Hive, the protocol politicians—they are not the same as their old-world counterparts, but their reappearance also suggests a growing recognition of the need for political institutions in some form.

The application of Kleros here still fundamentally rests on the cryptoeconomic tools out of which Kleros is built. However, even just the framing of asking Kleros jurors to look at a question in a context of "judicial review" prompts a specific mindset among jurors. Kleros is a Schelling point system, and as is well-established, the way one asks a question can have a significant impact on what Schelling (focal) points people gravitate towards.

Already, this "judicial review mindset" is a different way of looking at an issue than one might in the perspective of a pure token governance vote. So the different branches of governance under this scheme can interact with a given governance proposal in very different ways, while all being built out of cryptoeconomics, as they interact with human psychology differently.

Hybrid Systems and Kleros

Schneider's proposed solution to the limitations of cryptoeconomics in governance is to layer cryptoeconomic tools with institutions that follow a political, non-economic logic. He writes:

A napkin sketch of classical, never-quite-achieved liberal democracy (Brown, 2015) would depict a market (governed through economic incentives) enclosed in politics (governed through deliberation on the common good). Economics has its place, but the system is not economics all the way down; the rules that guide the market, and that enable it in the first place, are decided democratically, on the basis of citizens' civil rights rather than their economic power. By designing democracy into the base-layer of the system, it is possible to overcome the kinds of limitations that cryptoeconomics is vulnerable to, such as by counteracting plutocracy with mass participation and making visible the externalities that markets might otherwise fail to see.

and

I have argued that pairing cryptoeconomics with political systems can help overcome the limitations that bedevil cryptoeconomic governance alone. Introducing purpose-centric mechanisms and temporal modulation can compensate for the blind-spots of token economies. But I am not arguing against cryptoeconomics altogether.

Vitalik agrees, arguing that these non-economic elements can be effective even while being used sparingly if they are well-placed:

This seems broadly correct. Financialization, as Nathan points out in his conclusion, has benefits in that it attracts a large amount of motivation and energy into building and participating in systems that would not otherwise exist. Furthermore, preventing financialization is very difficult and high cost, and works best when done sparingly, where it is needed most. However, it is also true that financialized systems are much more stable if their incentives are anchored around a system that is ultimately non-financial.

We would also generally agree with the value of interweaving cryptoeconomic systems with other types of institutions towards having a more robust whole which can reflect the economic as well as non-economic motivations of their participants.

In fact, the idea of embedding cryptoeconomics as a tool within a broader, non-economic system is similar to a recent approach taken by a Mexican arbitrator, Mauricio Virues Carrera, who has pioneered an approach where arbitrable clauses in existing contracts specify a traditionally accredited arbitrator, but these clauses indicate that decisions will outsourced to Kleros.

Then, one can achieve Kleros' ability to scale to smaller disputes within a framework of traditional arbitration. This approach was recently applied in a tenant-landlord dispute that was decided by Kleros that was subsequently upheld by a Mexican court. Carrera writes about this approach:

The idea of using Kleros as a ‘tool’ rather than a 'system' for dispute resolution fundamentally allows to embed it into (practically any) other mechanisms (such as arbitration. This approach can create a hybrid system that equally benefits from current legal frameworks and traditions (for example, the strong and wide roots of commercial arbitration) and the new technologies and dynamics, brought by Kleros and digital-based dispute resolution systems, into the civil justice global landscape. There is now a successful case of application of such a 'hybrid' model. A Kleros decision was recognised and ultimately enforced as a valid arbitral award before the Mexican courts.

Here you can dowload the full report about the case where a Mexican court accepted a Kleros ruling as legally valid.

Conclusions

Reflections on the advantages, but also the limitations, of cryptoeconomics has informed both how Kleros has been designed as well as the settings in which it is being applied.

We broadly agree with the conclusions of Schneider and Vitalik that cryptoeconomics has a lot to offer, particularly when combined with non-economic logic that takes it beyond purely "financialized" applications. Indeed, Kleros can be a valuable component in these hybrid settings.