Decentralized society

Finding Web3's soul

A digest of paper Decentralized Society: Finding Web3's Soul by E. Glen Weyl, Puja Ohlhaver, and Vitalik Buterin, written on May 10, 2022

Introduction

The Dao is the hearth and home of the ten thousand things. Good souls treasure it, lost souls find shelter in it. — Laozi, #62

Web3 has stunned the world by forging a parallel system of unprecedented flexibility and creativity in less than a decade.

Yet at the core of Web3 there are humans and their relationships. Because Web3 lacks primitives to represent such social identity, it has become fundamentally dependent on the very centralized Web2 structures it aims to transcend, replicating their limitations.

Examples of these dependencies include:

1. Most NFT artists rely on centralized platforms like OpenSea and Twitter to commit to scarcity and initial provenance.

2. DAOs that try to move beyond simple coin-voting often rely on Web2 infrastructure, such as social media profiles, for sybil resistance.

3. Many Web3 participants rely on custodial wallets managed by centralized entities like Coinbase or Binance. Decentralized key management systems are not user-friendly for any but the most sophisticated.

Souls and Soulbound Tokens

Our key primitive is accounts, or wallets, that hold publicly visible, non-transferable (but possibly burnable or revocable by the issuer) tokens. We refer to:

1. the accounts as Souls, and

2. tokens held by the accounts as Soulbound Tokens (SBTs).

Imagine a world where most participants have Souls that store SBTs corresponding to a series of affiliations, memberships, and credentials. For example, a person might have a Soul that stores SBTs representing:

• educational credentials,

• employment history, or

• hashes of their writings or works of art.

In their simplest form, these SBTs can be self-certified, similar to how we share information about ourselves in our CVs. But the true power of this mechanism emerges when SBTs held by one Soul can be issued or attested by other Souls, who are counterparties to these relationships.

For example:

• the Ethereum Foundation could be a Soul that issues SBTs to Souls who attended a developer conference

• a university could be a Soul that issues SBTs to graduates

• a stadium could be a Soul that issues SBTs to longtime fans

Note there is no requirement for a Soul to be linked to a legal name, or for there to be any protocol-level attempt to ensure one Soul per human. A Soul could be a persistent pseudonym with a range of SBTs that cannot easily be linked. We also do not assume non-transferability of Souls across humans. Only the SBTs are non-transferable.

Applications

Souls would create a verifiable, on-chain way to stake and build social reputation of an individual. How can we spend our social reputation?

Art

Souls are a natural way for artists to stake their reputation on their works. When issuing a tradable NFT, an artist could issue the NFT from their Soul. The more SBTs the artist’s Soul carries, the easier it would be for buyers to identify the Soul as belonging to that artist, and thereby also confirm the NFT legitimacy.

Credit and lending

SBTs that represent education credentials, work history, and rental contracts could serve as a persistent record of credit-relevant history, allowing Souls to stake meaningful reputation to avoid collateral requirements and secure a loan.

Loans and credit lines could be represented as non-transferable but revocable SBTs, so they are nested amongst a Soul’s other SBTs—a kind of non-seizable reputational collateral—until they are repaid and subsequently burned, or better yet, replaced with proof of repayment.

Private key recovery

Recovery methods today, like muti-sig recovery or mnemonics, have different tradeoffs in mental overhead, ease of transacting, and security.

Social recovery is an emerging alternative that relies on a person’s trusted relationships. SBTs allow a similar, but broader paradigm: community recovery, where the Soul is the intersectional vote of its social network.

Human agency is fundamentally social, and individuality is primarily constituted by the unique set of social connections and identities one adopts. In this sense, flourishing individuals are crossroads of different communities, not self-sufficient islands. Zoë Hitzig & E. Glen Weyl

Maintaining and recovering cryptographic possession of a Soul requires consent of the Soul’s network. By embedding security in sociality, a Soul can always regenerate their keys through community recovery, which deters Soul theft.

Souldrops

Thus far Web3 has largely relied on token sales or airdrops to summon new communities, which yield little accuracy or precision. Airdrops, in which tokens are algorithmically given for free to a set of wallets, mostly fall to some combination of existing token holders and wallets—easily attacked by sybils, encouraging strategic behavior and the Matthew effect.

SBTs offer a radical improvement we call souldrops: airdrops based on computations over SBTs and other tokens within a Soul. We can imagine a non-profit whose mission is to plant trees dropping governance tokens to Souls who hold a mix of environmental action SBTs, gardening SBTs, and carbon sequestration tokens—perhaps dropping more tokens to the carbon sequestration token-holders.

DAO of Souls

DAOs could mitigate sybil attacks with SBTs in several ways, by:

• computing over a Soul’s constellation of SBTs to differentiate between unique Souls and probable sybils, and denying any voting power to a Soul that appears to be a sybil

• conferring more voting power to Souls who hold more reputable SBTs

• issuing specialized proof-of-personhood SBTs, which could help other DAOs bootstrap sybil resistance

• checking for correlations between SBTs held by Souls who support a particular vote, and applying a lower vote weight to voters who are highly correlated

The latter idea of correlation checking is particularly promising and novel. A vote supported by many Souls who all share the same SBTs is more likely to be a Sybil attack and—even if not a Sybil attack—such a vote is more likely to be a group of Souls who are making the same error in judgment or who share the same bias, and so should reasonably be weighted less than a vote with the same numerical level of support but from a more diverse base of participants.

Informal Twitter poll evidence suggesting that people already find the idea of taking diversity into account in decision-making mechanisms intuitive.

Correlation score

Mathematically, if A and B are two agents, P is a project, $X_{A \rightarrow P}$ and $X_{B \rightarrow P}$ are the contributions of A and B to P, and $X_A$ and $X_B$ are the contribution vectors of A and B to all projects, then the contribution of A and B to P should be discounted by the similarity of the contribution vectors $X_A$ and $X_B$. Popular measures of similarity are cosine similarity and Pearson similarity.

If A and B have a low similarity, we assume that they are highly independent agents, and give them close to the maximum subsidy whenever they do contribute to some project together. But if A and B contribute to the same project frequently and/or in large amounts, we assume that they are highly coordinated and are acting somewhat more like a single agent, and discount the subsidies to projects that they co-fund.

To measure similarity among the two contribution vectors, the authors propose the following correlation score, which is akin to (non-normalized) cosine similarity:

$\sum_P \sqrt{X_{A \rightarrow P} \cdot X_{B \rightarrow P}}$

The correlation score is intended to reflect to what extent two participants contribute to the same projects. If two participants A and B both contribute x to some project, then their correlation score is increased by x. If they contribute different amounts, their correlation score increases by the geometric mean of their two contributions.

Artificial Intelligence to Plural Intelligence

Large scale non-linear neural network models (such as BERT and chatGPT) could also be transformed by SBTs. Such models hoover volumes of public or privately surveilled data feeds to produce rich models and predictions, such as code based on natural language prompts.

However, most surveilled data creators aren't aware of their role in creating these models, retain no residual rights, and are viewed as incidental rather than as key participants.

Moreover, data hoovering divorces models from their social context, which masks their biases and limitations and undermines our ability to compensate for them.

SBTs offer a natural way to program economic incentives for provenance-rich data while empowering data creators with residual governance rights over their data. At the same time, model-makers can track the characteristics of the collected data and their social context - as reflected by SBTs - and find contributors that offset biases and compensate for limits.

Risks

Souls can go to Heaven... or Hell.

While we have selectively highlighted the potential unlocked by DeSoc that we find promising, it is important to remember that almost any technology with such transformative potential will have a similar potential for destructive transformation: fire burns; the wheel steamrolls; the television brainwashes; cars pollute; credit cards trap in debt, and so on.

Here, the same SBTs that could be used to compensate for in-group dynamics and achieve cooperation across differences could also be used to automate red-lining of disfavored social groups or even target them for cyber or physical attack, enforce restrictive migration policies, or make predatory loans.

Just as the downside of having a heart is that a heart can be broken, the downside of having a Soul is it can go to hell and the downside of having a society is that societies are often animated by hatred, prejudice, violence and fear. Humanity is a great and often tragic experiment.

As we meditate on the possible dystopias of DeSoc, we should also contextualize these possibilities within other technological enabled dystopias:

• Whereas Web2 often relies on top-down artificial bureaucracies to confer identity, DeSoc relies on horizontal peer-to-peer social attestations

• Whereas DeSoc empowers Souls to encode their own relationships and co-create plural property, Web2 intermediates social connections or monetizes them with opaque algorithms that can polarize, divide, and misinform

• DeSoc sidesteps top-down, opaque social credit systems. Web2 forms the basis of them

• DeSoc treats Souls as agents, whereas Web2 treats Souls as objects

Thus, DeSoc does not need to be perfect to pass the test of being acceptably non-dystopian; to be a paradigm worth exploring it merely needs to be better than the available alternatives.

Whereas DeSoc has possible dystopian scenarios to guard against, Web2 and existing DeFi (Decentralized Finance) are falling into patterns that are inevitably dystopian, concentrating power among an elite who decide social outcomes or own most of the wealth.

The direction of Web2 is deterministically authoritarian, accelerating the capacity of top-down surveillance and behavior manipulation. The direction of today’s DeFi is nominally anarcho-capitalist, but is already falling into network effect and monopoly pressures that risk its medium-term path becoming authoritarian in much the same way.

DeSoc, in contrast, is stochastic social pluralism — a network of individuals and communities that come together, as emergent properties of each other, co-determining their own future.

Looking at Web2, the outgrowth of DeSoc can be analogized to the rise of popular participatory governments out of centuries of monarchy. Participatory governments didn’t inevitably give rise to democracy; it also led to the rise of communism and fascism. Similarly, SBTs don’t make digital infrastructure inherently democratic, but are democratic-compatible depending on what Souls and communities co-determine.