21D/ Decentralized Reputation

From IIW

Decentralized Reputation

Thursday 21D

Convener: Johannes Ernst


Tags for the session - technology discussed/ideas considered:

Reputation, authentic data, decentralization

Discussion notes, key understandings, outstanding questions, observations, and, if appropriate to this discussion: action items, next steps

Challenge: the motivated malicious user is willing to dedicate far more time and resources to manipulate the reputation systems then a normal user. This leads reputation systems to be dominated by those with the most extreme views.

In lieu of a whiteboard:


Links from the audience

  • Sam Smith: presentation on a github pdf ; more papers on reputation are here

  • Timothy Holborn: medium link

    • example about ‘repetitional’ related problems linked to ‘identity’ associated decision making (ie: associating DNA to birth certificates by default, rather than ancestry.com)

    • please decouple this comment from my identifier, for purposes of privacy / dignity protection

    • friend link

    • causality & social informatics implications. complex version

    • Notes about ‘identity’ (human agency / personhood, AI Ethics related considerations) - also this and this

  • Juan Caballero: github pdf

  • Not only algorithm, but the people who pick the algorithm

  • Lots of money often equals high reputation because it can be paid for

  • Open Federated Learning

  • As soon as you allow non-transparency, people are tempted to introduce bias and cannot learn from others


Rouven Heck1 to Everyone (10:04)


Nader Helmy to Everyone (10:04)

This is one of the oldest problems in identity 🙂 there’s a lot of prior art

To put it lightly

Johannes Ernst (Indie Computing) to Everyone (10:05)


Me to Everyone (10:05)

Nader Helmy to Everyone (10:07)

The reputation of “x” for what purpose? It’s not a universal score, there is presumably a specific context to which that claim is being made

Andy Morales to Everyone (10:07)

RottenTomatoes review bombing is a great example too

Steve to Everyone (10:08)

Can we record the meeting?

Juan from Spruce to Everyone (10:08)

are there slides?

TelegramSam to Everyone (10:08)

diagram in notes.

Andy Morales to Everyone (10:08)

We are in the notes, Johannes is diagramming there

Nader Helmy to Everyone (10:08)

No slides but here are Johannes’ notes: https://docs.google.com/document/d/1rDMMm2AB75nH4YhMDOpbLXUgRkgLIh1VWwhJJNzF71M/edit

Juan from Spruce to Everyone (10:09)

nice thanks, qiqo bombed and was showing me a "request access" page somehow! 🤦‍♀️

Timothy Holborn to Everyone (10:09)

question reference: https://miro.medium.com/max/3122/1*OD62QRiDXQd5mBdR_MiFpQ.png

Juan from Spruce to Everyone (10:10)

do people want to record?

John Phillips - Sezoo to Everyone (10:11)

my preference is yes (if only because I'm running on very little sleep so my ability to absorb new content is limited! 🙂)

Juan from Spruce to Everyone (10:11)

^ +1

Michel Plante to Everyone (10:11)

recording preferred

Timothy Holborn to Everyone (10:12)

“the distinction between reality and our knowledge of reality, between reality and information, cannot be made” Anton Zeilinger

Juan from Spruce to Everyone (10:12)

can someone get the host-code and hit "record to cloud"?

hostcode = 232323

google founders :(

Andy Morales to Everyone (10:14)

But will the data EVER be securely attributed on the internet?

I just…don’t ever see that happening?

John Phillips - Sezoo to Everyone (10:14)

Do we really want this? Just because we can, does that mean we should? What problem are we solving for here? If we have concerns that "identity" is too often the table stakes for interactions, surely we should be more concerned if reputation becomes the table stake?

Juan from Spruce to Everyone (10:14)

huge +1 -- sam's papers on this are great reading, highly recommend on this.

@Andy, we're all dealing in probabilities between 0 and 100 that one or the other decentralized technology will get us close enough to that goal :D

Juan from Spruce to Everyone (10:15)

@John Philips, Dave Huseby has a session on that very question happening now in another room :D

Andy Morales to Everyone (10:15)

Love that, but there’s a high high chance that actors that do NOT want to be recognized will do everything in their power to at least obfuscate the systems no?

Juan from Spruce to Everyone (10:15)

(I agree the social dangers are huge!)

Eric Weber to Everyone (10:16)

Don't you have to ask A,B,C,D for rating

Andy Morales to Everyone (10:20)

But this is a different situation because landlord has to abide by clear laws ruled by public organizations, person submitting a review anonymously on the internet does not have any potential compliance issues

John Phillips - Sezoo to Everyone (10:21)

Answering my own question further up in the chat (just because we can, doesn't mean we should)... if we assume that our reputation is being rated by others, whether we like that or not (think Equifax etc.), then having an understanding of how our reputation is measured, and having some control over how we present it, *could* be good. However there are way too many dystopian possibilities here. We could create a "rod for our own back".

- thanks @Juan, I'll try and clone myself!

Trev Harmon to Everyone (10:21)

@Andy, we'd like to believe that's the case, but it isn't everywhere. This renter example creates a very concerning extension of the power differential between tenant and landlord.

Dan Robertson (he/him) to Everyone (10:21)

But how do I know the landlord is not a synthetic entity, who has been made to appear reputable via a Sybil attack (i.e. many fake renters who have asserted they rented from the faux landlord)?

Andy Morales to Everyone (10:22)

@Trev fair.

But see, “they’re a public legal entity"

This example depends on “public legal entity” as a core concept

Juan from Spruce to Everyone (10:23)


Timothy Holborn to Everyone (10:23)


Brigitte Piniewski to Me (Direct Message) (10:24)

Do you have an example of when the system is in place, then people will work to give you better data?

Michel Plante to Everyone (10:25)

Landlord + tenant use case: https://domilabs.io/

Trev Harmon to Everyone (10:26)

I think that the landlord/renter example only really works with consistent regulation. However, this is just a hypothetical example.

Andy Morales to Everyone (10:26)

The problem with these things (decentralized reputation systems) is ALWAYS that we need a group of humans doing investigative work SOMEWHERE. Whether it is in a centralized public organization, or in a series of private organizations that build visualizations of the networks of reputation, or volunteers...

I’m more interested in how THAT labor will organize

Andy Morales to Everyone (10:27)

I think that’s ultimately the most important thing

Juan from Spruce to Everyone (10:28)

there's a lot in glassdoor on SV employers

outside of california it's a graveyard

here in berlin it's almost worthless, <5 reviews for companies with >100 positions

Andy Morales to Everyone (10:28)

Blind is a better example imo

Because you need to have a company email to be able to submit a review

Andy Morales to Everyone (10:29)

But it anonymizes you

Rouven Heck1 to Everyone (10:29)

Before we have computation on top of encrypted data - can we really build privacy preserving reputation systems?


Dan Robertson (he/him) to Everyone (10:29)

I think glassdoor is also better for larger companies than smaller. If you’re at a three-person company, you probably don’t feel safe to be honest in posting publicly about how you feel about your employer — since “…by a Product Manager at XYZ Co” may literally identify you uniquely.

👆Or Blind app, etc.

Timothy Holborn to Everyone (10:31)

causality & social informatics implications. complex version is https://medium.com/webcivics/theoretical-relationship-between-social-informatics-systems-and-quantum-physics-reality-check-6ce3781d1a29?source=friends_link&sk=f0d5323abc9355ad1edb7e15f1e60f41

Timothy Holborn to Everyone (10:31)

Notes about ‘identity’ (human agency / personhood, AI Ethics related considerations)




mary104 to Everyone (10:31)


Trev Harmon to Everyone (10:32)

We also still have the issue with bias in AI.

Charles Lanahan to Everyone (10:32)

@Timothy I'm getting "oops that page can't be found"

Dan Robertson (he/him) to Everyone (10:32)


Bart Suichies to Everyone (10:32)

What reputation are we talking about?

Mary to Everyone (10:33)

and who made the algorithms and what are their biases

Timothy Holborn to Everyone (10:33)

@charles - i’ve tried the links, perhaps someone else can confirm if there’s a problem?

Charles Lanahan to Everyone (10:33)

oh wait, haha user error

Charles Lanahan to Everyone (10:33)

I copy/pasted both links as one link

Timothy Holborn to Everyone (10:33)

friend link to https://medium.com/webcivics/the-semantic-inforg-the-human-centric-web-reality-check-tech-50e2fa124ed4?source=friends_link&sk=27043bff446e23466b9bae786fa614e0

Trev Harmon to Everyone (10:33)

@Mary, exactly. We don't need AIs "inadvertently" disadvantaging certain groups because of those biases.

Mary to Everyone (10:34)

@Trev yes

TelegramSam to Everyone (10:37)

the discussion about the topic generally is excellent, but one point: we can make reputation 'portable' by issuing credentials. This is obviously limited, but also right in front of us.

we don't have to fully solve the issue to benefit from some 'portable' reputation usecases.

Juan from Spruce to Everyone (10:40)


Me to Everyone (10:40)


Juan from Spruce to Everyone (10:40)

<3 non-rivalrous good

Timothy Holborn to Everyone (10:41)

example about ‘repetitional’ related problems linked to ‘identity’ associated decision making (ie: associating DNA to birth certificates by default, rather than ancestry.com) https://www.dailymail.co.uk/news/article-9106197/Researchers-uncover-British-foreign-aid-workers-fathered-children-abroad.html

Timothy Holborn to Everyone (10:42)

please decouple this comment from my identifier, for purposes of privacy / dignity protection: https://drive.google.com/file/d/1Vni97O6Pa1YETLE0S3NzV4CsWwDo5YLl/view?usp=sharing

(above it says repetition rather than reputation).

Andy Morales to Everyone (10:45)

Lol my computer shut down

Andy Morales to Everyone (10:45)

Didn’t mean to mic drop and leave

Johannes Ernst (Indie Computing) to Everyone (10:46)

I wondered where you suddenly had gone!

Juan from Spruce to Everyone (10:47)

Plugging a colleague's session on this topic: 24I - Self-sovereign Applications on “Authorized P2P” Infrastructure: Kepler Design Progress Report (added after session generation this morning) - 1.45 PT/4.45 ET/Very late CET

Andy Morales to Everyone (10:47)

Someone catch me up! Did someone suplex my argument!?

Eric Weber to Everyone (10:49)

Is the objective reference the sum of a lot of subjective references?

Me to Everyone (10:50)

I call this "User Permissioned Trust"

Juan from Spruce to Everyone (10:54)

Do I have to plug my colleague's session again?

I think the ZCap/OCap people would say that granular and caveated delegation could get us most of the way, if not all the way, to those kinds of crawlers and discovery mechanisms and authorization languages...

Juan from Spruce to Everyone (10:55)

(I'm trying to bait Adrian to answer here ;D )

Zorigt Baz to Everyone (10:57)


Juan from Spruce to Everyone (10:57)


Dan Robertson (he/him) to Everyone (10:58)

I think we have until 15 past the hour, no?

John Court to Everyone (10:59)

I am not convinced from any of this that reputation will become decentralized through current SSI technology. The centralised ecosystem can switch to issuing VCs and maintain their audience because word of mouth from individuals no matter what form is inherently untrusted when monetary lose could be involved. What seems to be needed is an ecosystem to build trust in individuals somehow.

Andy Morales to Everyone (11:00)

+1 to the above

I also want to mention “decentralized trust systems” have been built by women for YEARS. They were called “gossiping” and was never paid precisely because it was not backed by money or powerful institutions

Me to Everyone (11:01)

cryptographically authenticated but fully pseudonymous trust and be built over merely from behavior. This is the type of reputation that allows trust in the history of behavior. Similarly it allows rapid loss of trust given any variation in behavior. Think Victorian ear female authors with male pseudonyms

Andy Morales to Everyone (11:01)

You wanna design a dececnralized trust system? Just go to a hair salon.

Juan from Spruce to Everyone (11:01)


Aaron Goldman to Everyone (11:02)

Gossip is a valuable protocol

Charles Lanahan to Everyone (11:02)


Zorigt Baz to Everyone (11:02)

That’s social media

Juan from Spruce to Everyone (11:02)

I have a non-IIW meeting I have to run to. thanks all!

Andy Morales to Everyone (11:02)

No, social media has other weird incentives, like faking that your life is better than it is.

Trev Harmon to Everyone (11:02)

Thanks for the discussion everyone. Have another meeting to run to.

TelegramSam to Everyone (11:02)

There is only IIW.

Andy Morales to Everyone (11:02)

At the hair salon you are looking at your worst


Andy Morales to Everyone (11:03)

Have to go :( bye everyone!

Eric Weber to Everyone (11:04)

There is certain chance that your reputation will de-anonymize your pseudonym

John Court to Everyone (11:05)

So a non-authenticated form of individual trust could be seen to be the current fad of `social media influencers`. Can we somehow build on how that has somehow happened ?

Dan Robertson (he/him) to Everyone (11:05)

Great point, Sam, re: not needing to be associated with a natural human individual

Timothy Holborn to Everyone (11:06)


Mary to Everyone (11:06)

same with scammers who steal financial payment artifacts like atm cards, etc

Me to Everyone (11:07)

If you pseudonym is a cryptographic random string of sufficient entropy then there is no correlation to the string only from the body of work itself.

evanwolf to Everyone (11:08)


Timothy Holborn to Everyone (11:08)

probably important note: (noting therein also, wikidata) https://youtu.be/l0P4Cf0UCwU