Featured image of post Deliberative Consensus Protocols

Deliberative Consensus Protocols

Introduction: Scalable Group Decision-Making

A deliberative consensus protocol is a process that online groups can use to make decisions. It’s designed to produce good decisions that are fair and manifest the collective intelligence of the group. And it’s designed to work at scale.

This is not easy, especially on the Internet, where bad-faith actors can easily create bots and sock-puppets to try to manipulate the results. And even if everyone acts in good faith, collective intelligence does not always scale well. Large groups often fail to come to consensus on even basic facts, let alone optimal actions.

But in this essay I argue that intelligent group decision-making can scale with the right process.

A deliberative consensus protocol is like a blockchain consensus protocol, which induces users to give honest answers to questions of sufficiently uncontroversial facts (such as the state of the blockchain). But unlike a blockchain consensus protocol, which only works when there is no honest disagreement about the correct answer, a deliberative consensus protocol does not punish disagreement. Instead, it rewards honest opinions. It then uses a deliberative process to discover the most informed and unbiased opinion of the group.

Although it is like a blockchain consensus protocol, a deliberative consensus protocol does not require a blockchain. It can be used anywhere an online group needs to make decisions: governance of open-source projects and DAOs, direct democracy, open fact-checking and peer review, moderating online forums, or improving social-media ranking algorithms to combat misinformation.

But a deliberative consensus protocol can facilitate not just more intelligent decisions, but also more intelligent conversations, by promoting a more healthy conversation dynamic that incentivizes informed honesty, rather than dishonest conformity.

Fair Decisions

Blockchain protocols consistently (and amazingly) produce a consensus on the truth when it comes to questions of unambiguous, uncontroversial fact. For an intuition of just how this miracle occurs, read my essay on Truthtelling Games.

Unfortunately, no protocol can consistently give us the “truth” when the facts are uncertain or subjective, or when values or preferences conflict. The best we can do is try to create a protocol that produces the most fair decision.

What constitutes a fair decision? Well, let’s consider some examples of unfair decisions. In a criminal trial, if exculpatory evidence is withheld, that’s unfair. If the jurors are bribed to not give their honest opinion, that’s unfair. And if the jury isn’t representative of the defendant’s peers, that’s also unfair.

So I would propose that a fair trail would be one that discovers the 1) honest and 2) fully-informed opinion of an 3) unbiased jury.

This is what a deliberative consensus protocol is designed to do. It uses a deliberative process to discover what an unbiased (e.g. representative) sample of the group would honestly believe after they have considered all the most informative – or convincing – comments made by other users.

To accomplish this, the protocol uses three different technologies, that correct in turn for 1) dishonesty, 2) ignorance, and 3) bias.

1) Correcting for Dishonesty using Game Theory

First, to address the problem of dishonesty and coordinated manipulation, a deliberative consensus protocol can use game-theoretical mechanisms such as the Bayesian truth serum. BTS is an extraordinary mechanism (developed at MIT) that rewards users for giving honest answers even if users believe that most people disagree with them.

To make these game-theoretical mechanisms work, there must be some sort of reward. If a deliberative consensus protocol is not a blockchain protocol, and the reward is not cryptocurrency, then what is the reward?

In the deliberative consensus protocol, the reward currency can be attention. As I argue in The Law Of Attention, if nobody pays any attention to your posts in an online platform, you will stop posting. If your votes have no effect on other posts, you will stop voting.

So using a reputation system where reputation translates to the ability to influence the attention of other users, we can use an increase or decrease in reputation as the reward currency. This allows a deliberative consensus protocol to fully employ the tools of game theory and mechanism design, and engineer a protocol that creates an equilibrium where everyone answers honestly, because they expect other users to do the same.

2) Correcting for Ignorance using Deliberation

Second, to address the problem of ignorance, a deliberative consensus protocol can use an algorithm such as the Global brain algorithm to curate conversations that discover the most convincing arguments on each side of a question, in order to determine the opinion of users who have considered all of these arguments.

The global brain algorithm works by analyzing a threaded conversation tree and considering the upvotes and downvotes on each comment, depending on who has seen what other comments before they voted. It then filters and ranks comments to influence how much attention each receives, in order to deepen the most informed conversation threads while pruning uninformative threads that do not effect voting behavior.

The integration of a truthtelling protocol such as the Bayesian Truth Serum with a deliberative protocol such as the Global Brain Algorithm can produce an equilibrium at informed honesty, where users maximize their influence on the platform by voting according to their honest opinion given the information that has been shown to them.

This is the opposite of the effect of many social media algorithms, which create an equilibrium on dishonest conformity: where people are rewarded for posts that are easily recognized as conforming to the biases and dogmas of the group, even if that requires ignoring nuance and reason.

3) Correcting for Bias using Machine Learning

However, even if all members of a group are honest and informed of the most convincing information and arguments posted by other users, the result will still depend on who you ask, because people have different core values, belief systems, and preferences. And because participation in online discussions is necessarily optional, votes will often be a biased sample of the opinions of the group.

To address this problem of self-selection bias, an unsupervised machine-learning algorithm can be used discover latent factors that predictably affect users’ votes, and adjust for these biases so that the results are representative of the overall opinions of the group. A similar algorithm is used by X’s Community Notes to correct for political bias.


A deliberative consensus protocol does not solve all the problems of group decision-making. For example, the dilemmas in designing fair democratic voting mechanisms addressed by social choice theory still exist when there are conflicting goals and preferences. But a deliberative consensus protocol can enhanced online group decision-making processes by more effectively distributing information and producing a fair decision that represents the collective intelligence of the group.

But the greatest potential of these protocols may simply be the improvement in discourse. In a sense, the most important decision for any group to make is how to allocate their attention. In any online discussion platform, the global brain algorithm can be used to focus attention on the most informative conversation threads and on comments that stand up to scrutiny, thereby promoting deep, honest, informed, and intelligent conversations.

Originally posted on social-protocols.org.
Built with Hugo
Theme Stack designed by Jimmy