Guessing Is Not Judging: Why Crowdsourced Blockchain Decisions Are Not Arbitral Awards

In this post, we argue that decisions produced by crowdsourced blockchain dispute-resolution systems such as Kleros and Polymarket’s UMA protocol do not qualify as arbitral awards. Their defining feature is that they incentivize participants to predict what the majority of other participants will decide. Arbitration, by contrast, is oriented towards finding the right answer through independent judgment, deliberation, and reasoned decision-making. Guessing may produce the correct result. But it is categorically different from adjudication. The fact that a crowd sometimes reaches the right answer does not transform a process of prediction into a process of judgment.

Blockchain-based arbitration has been around for a while. The key idea is to substitute a traditional arbitral tribunal with a ‘swarm’ of jurors. Kleros is probably the best-known example. The dispute is reduced to a dichotomous decision problem. Prospective jurors self-select into specialized sub-courts and must stake the platform’s native cryptocurrency if they wish to participate in the vote. The more tokens a participant stakes, the greater the likelihood of being selected as a juror. Jurors review the evidence and cast their votes. The party supported by the majority prevails. Jurors who vote with the majority are rewarded, while minority jurors lose part of their stake, which is redistributed to the majority. Kleros openly embraces this game-theoretic model of dispute resolution. The idea underpinning the system is to incentivize jurors to think about what the other jurors will do and to vote accordingly. In other words, jurors are rewarded not for reaching an independently justified decision, but for correctly anticipating the decision of the majority.

A similar mechanism has acquired enormous practical importance through Polymarket, one of the world’s largest and most influential prediction-market platforms. Prediction markets generally, and Polymarket in particular, have experienced explosive growth in recent years and now process billions of dollars in trading volume. Users buy and sell contracts whose value depends on the occurrence of future states of the world, ranging from election outcomes and geopolitical developments to sporting results. Most contracts are resolved automatically. Problems arise, however, when there is disagreement as to whether the condition specified in a contract has been satisfied. In these cases, Polymarket outsources the decision to Universal Market Access (UMA), a decentralized verification protocol. Holders of UMA’s native cryptocurrency vote on the disputed question, with voting power determined by the number of tokens they hold. The outcome supported by the majority determines how the contract is settled. 

Kleros and UMA claim that their voting mechanisms incentivize participants to identify the honest, fair, and correct outcome. Both systems are built on the premise that a participant who seeks to maximize her reward should ask what the majority of other voters is likely to decide. According to Kleros, the best strategy in this exercise is to vote for the ‘honest and fair’ outcome. Similarly, UMA states that the ‘correct vote is determined by majority consensus’ and rewards token holders for ‘honest participation’. The underlying logic is straightforward. If most voters seek to identify the honest, fair, or correct outcome, then voting for that outcome should also be the best strategy for anyone seeking to maximize her financial return from the voting process.

The theory inspiring this logic was developed by game theorist Thomas Schelling in his analysis of coordination games during the Cold War era. In a coordination game, the objective of the players is to coordinate their behavior. They receive a reward if they succeed. Assume that we will receive US $1,000 if we manage to meet somewhere in New York City tomorrow without communicating with one another beforehand. Where would you go, and when? Think about this for a moment. Many people would choose Times Square at noon. Schelling argued that, in situations of this kind, certain solutions are more salient than others and therefore exert an almost magnetic pull. This may be because of mathematical symmetry, obvious fairness, geographical prominence, or some other distinctive quality. Splitting the difference in a price negotiation may be such a solution. So may be noon at a famous and centrally located meeting point. Schelling called such solutions ‘focal points’.

* * *

Are decisions reached by jurors in decentralized blockchain-based voting systems arbitral awards within the meaning of the UNCITRAL Model Law and national arbitration statutes? Kleros explicitly thinks so: ‘In short, Kleros arbitration process fits within the structures and frameworks of a legally binding arbitration protocol’. The issue is of considerable practical importance. Arbitral awards benefit from the powerful enforcement mechanism of the 1958 New York Convention and can be recognized and enforced in more than 170 jurisdictions. At the same time, awards may be challenged in annulment proceedings if they fail to satisfy mandatory requirements relating to jurisdiction, due process, or public policy.

Focal Points and Phony Focal Points

Kleros, UMA, and similar systems rely on Schelling’s focal-point concept to establish a connection between majority voting and an honest, fair, or correct outcome. Let us assume for the sake of argument that, if such a connection could be established, it would support characterizing the resulting decision as an arbitral award. The problem is that the premise is much weaker than it appears.

In many, and perhaps most, cases that reach a dispute-resolution mechanism, there is no focal point at all. If the correct or fair outcome were obvious, we would expect the parties to settle without resorting to adjudication. By doing so, they could avoid the costs of dispute resolution. The very fact that a dispute arises is usually evidence that reasonable people disagree about the facts, the interpretation of a contract, or the applicable legal standard. In such circumstances, there is no reason to assume that participants will naturally converge on a particular outcome.

Kleros states that ‘even if a particularly obvious option does not exist, some options will be perceived as more likely to be chosen by other parties and will effectively be chosen’. But this is merely an assertion. Kleros offers no explanation as to why such options should generally emerge. The assumption underlying Kleros and UMA—that voters will generally coordinate on the correct or fair solution—is therefore far less plausible than it first appears.

One possible response is that, even where there is no focal point inherent in the facts or the applicable law, participants may nevertheless converge on what they regard as a fair outcome. Fairness, on this view, functions as a second-order focal point. But this merely relocates the problem. What counts as fair is itself often contested. Different individuals may have very different intuitions about fairness, shaped by their experiences, values, and social backgrounds. In many disputes, disagreement about fairness is precisely what divides the parties. There is therefore no reason to assume that participants will naturally converge on a common conception of fairness any more readily than they converge on a common interpretation of the facts or the law.

If neither the facts, nor the law, nor a shared conception of fairness provides a focal point, participants are left with the task of predicting what other participants are likely to think. At the same time, the parties will typically seek to persuade voters that their preferred outcome is the correct or fair one. If an obvious focal point existed, there would be little need for such efforts. Their very existence suggests that no such focal point is available.

The problem runs deeper still. Even where a focal point exists, it need not coincide with the correct or fair outcome. Schelling repeatedly emphasized that successful coordination does not imply rationality, fairness, or truth. Communities may coordinate around conventions, biases, dominant narratives, misconceptions, or simple rules of thumb. Splitting the difference may become a focal point because it appears symmetrical. A particular outcome may become focal because it reflects a widely shared political belief or social prejudice. Focal points therefore need not track legal correctness or justice. They merely identify stable equilibria.

This gives rise to what may be called phony focal points. If participants believe that a majority of voters share a prejudice, a misconception, or a mistaken interpretation, they have an incentive to vote accordingly. The resulting equilibrium may be perfectly stable and perfectly wrong. The mechanism rewards convergence irrespective of the reasons that produce it.

The argument advanced by Kleros and UMA therefore breaks down at the outset. In many cases there is no focal point, and where a focal point exists, it may have little or nothing to do with correctness, fairness, or justice.

Guessing vs. Judging

The failure of the focal point argument is already enough to cast serious doubt on the claim that decisions generated by Kleros, UMA, and similar systems can qualify as arbitral awards. But the problem runs deeper. Even if the voting mechanism reliably produced correct outcomes, it would still not follow that the resulting decisions are arbitral awards.

An arbitral award is not simply an outcome. It is the product of a distinctive form of decision-making. Arbitrators are entrusted with determining the rights and obligations of the parties according to the applicable normative standard. Whether they apply national law, transnational principles, or equitable considerations (see Article 28(3) of the UNCITRAL Model Law on deciding ex aequo et bono), they are expected to seek the most justified answer available under the applicable standard. Their task is to get the answer right.

Ronald Dworkin captured this ideal through the figure of ‘Judge Hercules’, a hypothetical judge endowed with unlimited time and intellectual ability. Hercules seeks the interpretation of the law that best fits and justifies the legal system as a whole. The example illustrates Dworkin’s famous ‘one right answer’ thesis: adjudication is guided by the aspiration to identify the most justified answer to a dispute, even if real-world judges and arbitrators may sometimes fail to do so. The aspiration to correctness is constitutive of the enterprise. In Kantian terms, it functions as a regulatory idea that guides and structures the decision-making process.

Crowdsourced voting systems invert this orientation. The participant in Kleros or UMA is not primarily concerned with identifying the most justified answer. She is concerned with identifying the answer that other participants are likely to choose. The question is no longer: ‘What is the correct outcome?’ It becomes: ‘What outcome do I expect the majority to support?’ The focus shifts from the merits of the dispute to the anticipated behaviour of others.

The distinction is not just empirical. It is epistemic and normative. One process seeks justification; the other seeks successful prediction. Even where prediction happens to generate the correct result, it remains fundamentally different from adjudication. Guessing and judging are not the same activity. A participant who reaches the correct answer because she successfully anticipated the behaviour of others has not engaged in adjudication. She has engaged in prediction.

Guessing vs. Deliberating

The contrast becomes even clearer at the procedural level. Arbitration is not merely a mechanism for aggregating preferences. It is a process of deliberation. Members of a tribunal exchange arguments, challenge assumptions, test competing interpretations, and attempt to persuade one another through reasons. In doing so, they can rely on well-established methods and theories of legal reasoning and argumentation. Even where a sole arbitrator decides the case, the requirement to provide reasons performs a similar function. The award itself becomes part of the deliberative process because it exposes the decision-maker’s reasoning to scrutiny and criticism.

Crowdsourced voting systems operate differently. Their objective is not to facilitate deliberation but to induce convergence. Indeed, open discussion may undermine the very logic on which these systems rest. Kleros, for example, deliberately prevents jurors from observing one another’s votes before the voting process is complete. Restrictions of this kind are intended to preserve the conditions necessary for coordination around a focal point.

The result is that votes are counted rather than arguments weighed. Participants do not seek to persuade one another through the force of better reasons. They seek to predict how others will vote. The objective is therefore coordination rather than persuasion. Adjudication, by contrast, is fundamentally concerned with the quality of reasons. A tribunal reaches a conclusion because it regards one argument as stronger than another. Crowdsourced voting systems reach conclusions because one outcome attracts more support than another. These are fundamentally different decision procedures.

This difference matters because deliberation is not an accidental feature of arbitration. It is one of the principal mechanisms through which adjudicative institutions test, refine, and improve the quality of decisions. A process that replaces deliberation with prediction may generate convergence. But convergence is not the same thing as judgment.

Conclusion

Blockchain-based dispute-resolution mechanisms undoubtedly offer practical advantages. Yet judging is not simply a matter of reaching an outcome. It is a process directed towards finding the right answer to a dispute. That process requires independent judgment, reason-giving, and deliberation. Arbitrators do not ask what others are likely to decide. They ask what outcome is best justified by the facts, the applicable norms, and the available arguments.

Crowdsourced blockchain voting systems are built around a different logic. Their participants are rewarded for anticipating the decision of the majority. The resulting outcome may sometimes be correct. But a process of prediction does not become a process of adjudication merely because it occasionally gets the answer right. Judging is about weighing reasons in the search for the most justified answer. Guessing is not. For that reason, decisions produced by crowdsourced blockchain voting mechanisms should not be regarded as arbitral awards.

Horst Eidenmueller is Statutory Professor for Commercial Law at the University of Oxford and Professorial Fellow of St Hugh’s College, Oxford. 

Anna-Sophie Hochguertel is DPhil in Law student at the University of Oxford.