The response by Ast, George, and Trotz leaves the central point of our argument untouched. Instead, it recasts the issue as a continuum between ‘guessing’ and ‘judging’ and then argues that crowdsourced systems can incorporate elements of ‘honest evaluation’, written justifications, and appeals. But our argument is not that real-world adjudicators are infallible, or that Kleros jurors never think about the merits. It is that the defining incentive structure of the mechanism rewards participants for predicting the collective outcome, whereas adjudication is oriented toward determining the best justified answer through independent judgment, deliberation and reasoned decision-making. Adding elements associated with adjudication does not eliminate this fundamental difference in what the decision-maker is incentivized to do.
There is no reliable focal point in the cases that matter
The first response is that a Schelling point need not be an obvious answer. According to the authors, the contractual standard, the parties’ ‘dispute policies’, Kleros ‘court policies’, and the evidence provide a shared informational environment that can make honest evaluation the focal strategy.
This does not solve the problem. The critical cases are precisely those in which the substantive answer is disputed. If the applicable standard were sufficiently clear, and the evidence sufficiently compelling, there would often be little difficulty identifying the appropriate outcome. In hard cases, however, there may be disagreement not only about the facts but also about the interpretation of the contract, the applicable legal standard, the weight to be given to competing considerations, and the application of the standard to the facts. In such cases, there will usually be no substantive answer with sufficient ‘intrinsic magnetism’ to serve as a focal point on which independent participants can naturally converge.
Nor do the items identified by the authors create such a focal point. They describe the information and procedural environment in which the jurors operate. But they do not resolve the substantive disagreement that the dispute is about. Telling jurors what rules apply, providing evidence, and establishing court policies does not make the correct interpretation of those materials salient. The substantive question remains precisely what is contested.
The response also assumes what it needs to prove when it says that ‘honest, standard-applying judgment’ can become the focal strategy. Why should honest evaluation become the equilibrium that participants can safely predict? If reasonable jurors disagree about what the applicable standard requires, or about how the evidence should be assessed, there is no reason to expect convergence on a single answer merely because everyone is instructed to be honest. The mechanism still rewards the participant who correctly anticipates what the other participants will do. The existence of a common rulebook does not transform that predictive incentive into an incentive to reach the most justified answer.
Reason-giving is not deliberation
The second response points out that Kleros jurors are ‘invited’ to provide reasons and that appellate jurors ‘see’ these and can take them into account. Appellate jurors may therefore be exposed to the reasoning of first-round jurors rather than deciding in complete isolation. But exposure to reasons is not the same as deliberation. It does not reproduce the deliberative structure of judicial or arbitral decision-making.
Adjudication is characterized by an exchange of reasons. Parties make submissions. Opposing counsel challenge them. The tribunal tests competing arguments. Tribunal members question one another and, in a collegiate tribunal, deliberate with one another before a decision is reached. The process is designed to expose weaknesses in arguments, correct errors, and improve the quality of the eventual decision through back-and-forth engagement.
That critical exchange is not a peripheral feature of adjudication. A process in which jurors independently cast votes and are rewarded for matching the majority is fundamentally different. That jurors are ‘invited’ to provide written justifications does not mean that they are required to engage in an exchange of arguments with one another. Nor does the mere existence of reasons attached to votes substitute for deliberation. A collection of individual explanations is not the same thing as a deliberative process in which reasons are confronted, challenged and revised.
The appellate mechanism does not cure this deficiency. If first-round jurors formulate reasons in the hope that those reasons will persuade appellate jurors, the question remains what the appellate jurors themselves are incentivized to do. If their decision is again governed by a Schelling-point mechanism that rewards them for anticipating the majority outcome, the underlying predictive logic has not disappeared. It has merely been moved to the appellate level.
Hercules is an aspiration, not an empirical description
The authors also misunderstand our reference to Ronald Dworkin’s Judge Hercules. We do not suggest that real judges or arbitrators possess unlimited time, perfect information or superhuman intellectual abilities. Obviously, they do not. The point of Hercules is normative: it captures the aspiration that characterizes adjudication. A judge or arbitrator seeks to identify the right, or at least the best justified, answer and to develop the strongest arguments that support it.
That aspiration matters even when it cannot be fully achieved. A human arbitrator may misunderstand the evidence, make a legal error, overlook an argument or reach a conclusion that later proves to be wrong. None of this changes the orientation of the adjudicative enterprise. The decision-maker is nevertheless supposed to ask: What is the best answer under the applicable legal or contractual standard? What arguments support it? How should competing arguments be assessed?
A participant operating under a Schelling-point mechanism faces a categorically different incentive: to select the answer they expect the other participants to select. Someone may happen to reach the correct answer while responding to that predictive incentive. But correctness of the outcome does not change the nature of the process that produced it. Guessing correctly is not the same activity as judging correctly.
Conclusion
The response therefore does not displace our central argument. Indeed, its final characterization is revealing. The authors accept that, if arbitration is defined by the established international arbitration framework, crowdsourced mechanisms ‘fit uncomfortably’ within that category. In plain language, that is precisely the problem we identify.
We agree with the authors on the broader practical challenge. There is considerable scope for technologically sophisticated dispute-resolution processes that are faster, cheaper and accessible where conventional arbitration is inefficient. But the development of such mechanisms does not require treating every decentralized process as arbitration. These systems may constitute useful and innovative forms of dispute resolution without producing arbitral awards.
The task is therefore not to stretch arbitration until it accommodates any mechanism that produces a sufficiently reliable outcome. It is to design new forms of legal dispute resolution that are efficient, resistant to manipulation, and compatible with appropriate minimum standards of procedural fairness and reason-giving. Those mechanisms may ultimately deserve legal recognition in their own right. But a process should not be called arbitration, and its output should not be treated as an arbitral award, merely because a crowd can sometimes coordinate on the right answer. The distinction between predicting what others will decide and independently judging what the parties’ rights require remains fundamental.
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.
Share:
With the support of
