owned this note
owned this note
Published
Linked with GitHub
# Chaals' notes for trust circle at magician's meeting, Prague 2018
Daniel, starting out say your name and if tghere is a trust/reputation thing you like
Chaals, not really, beyond Stuff my friends tell me
Aeon, Group Norms
Fernan, five * system from BlaBlaCar
James, systems built out of things people have done
Tegan, Yalda, AirBnB's system (with issues)
Chris, Apple app rating system
Janek, trustlines - group of friends who say who they trust for how much (money)
RobD, ratings that have some kind of text justification, and stuff your friends say
Albert, Chris, not really.
Adam, something nice and clear
Tim, Github repo stars
Adam, github contributions profile
Alex, binary systems backed by some insurance
Kirill, ELO ratings
Bradley, discourse
Kazuaki, offchain trusteed execution
0ed, taking selfies of people and working out how far they are from you (codepiction)
Julian, depends on the use case. Transitive friend-based trust.
Ryan, contextual
Toby, people I know, and automated things that are relatively hard to game.
Max, not really
Sunny, Web of trust / graphical (pretty sure reputation only works on a local scale)
Pelle, graphs of trust, agree that it only works locally. Context part is really important.
Daniel, D&D alignment system (by the way, that's the right answer)
Daniel - when you measure for something, you get what you optimise for / Goodhardt's law is that people work to what they are measured on
... Nosedive episode of black mirror is important. Think it is worth talking about why 5-star is generally wrong.
Chaals - I don't believe in alignment, because in general I find people are less consistent. China's social credit system is pretty resistant to Goodhardt because gaming the system aligns with its goals, but I am not that fond of permanent monitoring to achieve that.
Daniel - when social credit is concrete and about simple rules, people generally like it.
?? why aren't we afraid of the financial credfit system?
[several people are]
Daniel - there is a general bias toward accepting things we have grown up with.
?? - What about the difference between objective and subjective rankings?
Daniel - nosedive was bad but airbnb is good. Categorising reputation systems, we have:
- Numeric ones like 5-star systems, where context is key.
This is reasonable for very constrained contexts. But in a multidiemnsional measurement, it doesn't work.
- Graphical systems based on trust graphs that are ("somewhat") transitive
- Dialectic (there is no right answer)
0ed - the monetary system reduces complex value to a single measure.
Daniel - Hayek wrote about how prices allow simple condensed markets that can price things, although they leave stuff out.
0ed - there could be global scores like that, too. Why would "people" be different from "things" in a market?
chaals - things are allowed to buy each other, but we take a dim view o people doing that (in a limited set of cases, to be honest)
Daniel - land, labour and capital are not fungible commodities in reality. If we imagine that things sort off work, so long as we don't then leave the market to destroy the planet and enslave the people.
Pelle - money is different from people's reputations because there is no algorithm behind ranking money, its value is just emergent. Whereas people who try to create a top-down measure of how to make everything work is like trying to put a single value on people, and that doesn't work well.
Kirill - any 1-dimensional system will be misused, for example by the powers that be. As happens with money, for example.
Adam (I think) - main difference between money and reputation system is that there is an inherent lack of incentive in the money system, whereas reputation has incentives that distort a market...
0ed - there is an interesting point. Money's value is emergent. Can we have a reputation system like that?
Daniel - we have one but it is hard to leverage its scale.
... wehich brings us to graphical systems. Instead of being averaged across a set of observations to get a number, you observe relationships between different people. These may not be scalar. you can run analysis on the graph to get data - e.g. for allocating budgets to projects.
... You can find communities if you look at them in social graphs.
Kirill - note that predicting behaviour from graphs of behaviour isn't as much fun when it gets used for things we don't lke.
Daniel - numeric rankings have a sense of a simple "better" and "worse", but graphs don't.
... The dialectic score (personality tests, D&D alignment, ...) doesn't have numbers, but there is existence on various axes. There isn't an objective score but they are good for subjective rankings.
Pelle - money is actually complicated like a graph. The graph model is important to understan emerget orders, and Graphs vary.
... looking at how different graphs combine, there are separate measures
Daniel, nice thing to ahve input for is that without a real value you seperate @@
?? - there is nosuch thing as "a china-wide algortihm", different factors affect score.
chaals - so do we dislike them all now?
[few people]number of downloads?
[folks]yeah, that is numeric
Daniel - I am big on dialectic, but it's really hard to do
Kirill - an idea for a limited set of things that could actually solve problems.
Daniel - like moving filtering on twitter to the edge so users decide what algorithm they want.
Chaals - the problem is that I am not going to look for things that I don't actually respect so I don't know that this will change anything for me
[is it better to have multiple algortihms? What's the benefit?]
[sharing e.g. feed lists - is that going to improve things or is it the same thing done in a graph model]
Bradley - is the inevitable outcome enabling segregation / discernment / ...? Can you recombine people?
chaals - adding some element of randomness can be pretty useful. Sometimes. And allowing for things that are local in time as well as context is important.
Daniel - there is a gap between the world we care about and how we represent it. The size of that gap is a proxy for noise in transmission, so what we say can cause problems and so it the way we hear it.
... Take that home and solve it...