Reputation in a multi-identity world

A person's reputation is a collection of the beliefs and opinions held about that person. In a sense, it is a model that is formed based on a person's past actions to try and predict how they will act in the future.

A person's reputation often determines their opportunities for collaboration within a network, so the way that we collectively create reputation has major implications on people's lives. Many people are excited about the potential for new forms of reputation to emerge in web3.

If we have a public ledger that measures all activity that takes place within a network, we have new potential to create accurate, verifiable, and scalable reputation systems. An on-chain based reputation system is incredibly exciting!

Blockchains introduce pseudonymous and (to a lesser extent) anonymous ways of interacting with the world. If we do not know who a person is, we can't reference their previous actions, and we can't make meaningful predictions about how they will act in the future. As Andrew Beal put it in his recent post:

"The DeFi ecosystem is a new quadrant of the financial system, but from a reputation standpoint, it is an island, separated from the mainland.

There is no identity. All the work you’ve put into establishing your reputation in the traditional financial system doesn’t apply because the DeFi ecosystem doesn’t know who YOU are. It’s like transferring high schools halfway through your junior year. Everything you did your first two years is gone, and you have to start over with a clean slate."

Andrew goes on to discuss the tactical impacts of a financial system without reputation (definitely give his essay a read), but to me, there lies another interesting question worth exploring: what is the meaning of reputation in a world with multiple identities?

For most of us, our entire identity is tied to our name. Our name is how we are addressed, recognized, verified, and judged.

But the internet has proliferated the use of multiple identities for a single individual. We are interacting with more and more people in extremely different contexts, sometimes necessitating the need for different identities. For example, the separation of a professional identify from a social-media identity might allow you to post things on platforms that would not be condoned by your employer or peers (see Slate Star Codex).

With blockchains, the atomic unit of identity is no longer a name but an address. If you use more than one wallet or layer 1 protocol, you likely have more than one address. Having multiple identities becomes the default rather than an inconvenient opt-in.

In a world with a single identity, all actions can be traced back to a single individual, and thus, a holistic reputation can be constructed for a person. In a world with multiple identities, a reputation is only tied to a single identity, providing only a snapshot of an individual.

In the previously mentioned piece on reputation, Andrew uses the example of a credit score to display the role of reputation on a person's life. Imagine you are a bank reviewing loan applications:

  • Person A: has a good credit score - they never missed a credit card statement, paid off their car loan last year, and has no outstanding debt.
  • Person B: has a bad credit score - they were late on 4 of their credit card payments last year, just opened up two new card accounts, and is 3 months behind on their mortgage.

All else being equal, the bank would obviously rather give a loan to Person A, and would offer them better rates. But what if these are actually the same individual under different identities?

When a bank looks at your FICO score, they are fairly confident that the score is representative of the totality of your financial history. If a crypto liquidity provider reviews your ARCx credit score, they will be much less certain that they have a holistic understanding of you - who is to say you don't have 5 different alt-wallets with vastly different behavior?

All of this begs the question - what are the implications of a reputation tied to an identity rather than an individual?

There is an argument to be made that reputation tied to an identity could be a more accurate predictor of future action than a reputation tied to a whole individual.

Identity based reputation is more context dependent. If I want to know how someone will act in a certain situation, it might be more informative to observe only their actions that are relevant to the situation I am trying to predict. We would essentially be narrowing the types of data the predictive model has access to, which could potentially result in more accurate predictions.

If I am considering giving a DAO grant to a developer: is it net beneficial or harmful to be able to look through their social media accounts? On one hand, you might find information that informs your decision, on the other hand you might be influenced by biases that are irrelevant.

Context specific identities will allow people to develop completely separate reputations in different parts of their life. It feels like this will lead to an increasing separation of art from the artist.

Growing up, I was warned not to post any potentially controversial photos on social media because it could prevent me from getting into college - should a teenager's college acceptance be dependent on their social media posts?

On the more extreme end, would we be better or worse off if we never knew that the singer of "Thriller", "Billie Jean" and "Beat It" was an alleged child molester?

In a digital world of multiple identities, for better or worse, these things may truly be separated.

But this world also raises a new attack vector for bad actors. The downside of ruining a reputation is lessened if I have or could create other ones.

Consider the credit score example. If I ruin my FICO score, no bank will give me a favorable loan moving forward. This creates a game theoretic check that creates consequences for bad financial behavior. If I ruin my web3 equivalent of a FICO score, I only damage the ability of that specific address to receive a loan, while I, the individual, can simply use another address.

Reputation being context dependent changes the calculus of the expected value of an action and potentially enables more bad or risky actions to be expected value positive.

We will need to develop new mechanisms to preserve the power of a reputation and to protect against bad actors.

Many projects already use proof of uniqueness (e.g., BrightID) to prevent against Sybil attacks. If a reputation application requires proof of human, then the system effectively protects against people maintaining multiple identities.

In systems where the downside of a reputation is less explicit, such as social situations where people do not have skin in the game (e.g., predictions), we can use staking / market mechanisms to punish bad actors. Whereas in web2, potential downside for wrong predictions is reputational risk, in web3, the downside is tied to financial outcomes, making it costly to be a bad actor.

A multi-identity world will bring new challenges as we develop on-chain reputation systems. Perhaps this leads to more accurate predictions of behavior and privacy that is more preserved. Perhaps this leads to opportunities for bad actors to exploit.

Regardless, we are moving towards a digital, pseudonymous world that is going to drastically change our meaning of identity and reputation. These changes will need to be further explored if we hope to create usable reputation systems for work, collaboration, and socialization.