Thoughts on reading the [DeCartography Whitepaper](https://scrapbox.io/tkgshn-private/DeCartography Whitepaper):
-
Regarding the mechanism:
- The discussion on building a social graph and the mechanism of decentralized oracles is interesting.
- However, I had some doubts about combining these two.
- In fact, is this not a decentralized oracle?
- As for the direction of decentralized oracles, I think there is something like Social Distance in the conventional sense (tkgshn).
- It is presented as a social graph, but I’m not sure what it exactly offers (tkgshn).
- Q. Is manual tagging the optimal method for labeling?
- My impression is that if a proper mechanism is created, it will work reasonably well, but it doesn’t seem elegant (blu3mo).
- When I say “manual tagging,” I mean it only in the context of creating a social graph (tkgshn).
- But what if we need Common Sense for use cases like temporal annotation?
- That’s exactly what I found well articulated in [/tkgshn-private/From Users to (Sense)Makers: On the Pivotal Role of Stigmergic Social Annotation in the Quest for Collective Sensemaking] (tkgshn).
- It’s like establishing something agreeable among humanity (tkgshn).
- Labeling is just ≒ compressing data, so there might be more mechanical ways to do it (blu3mo).
- Like the approach of pol.is?
- You mean using descriptions? (tkgshn) Like converting descriptions into natural language and measuring them with vectors? (Actually, pol.is only worked within the same language)
- Ah, but don’t people vote on the ones written in natural language? (tkgshn)
- I’m not sure what you mean by “descriptions” (blu3mo).
- Can you explain more? (tkgshn) (tkgshn) (tkgshn) (tkgshn) (tkgshn)
- I think the discomfort I wanted to express is more specifically articulated in the “issues I thought of” below, so please refer to that (blu3mo).
- The task of “having people choose natural language tags” is very dependent on language, so it’s not good because it reduces information or introduces subjectivity.
- I completely agree, I will take this into consideration! Thank you (tkgshn) (tkgshn) (tkgshn).
- I think the discomfort I wanted to express is more specifically articulated in the “issues I thought of” below, so please refer to that (blu3mo).
- Like the approach of pol.is?
- Issues I thought of:
- One issue with manual tag selection is that the information is reduced to a level that humans can interpret as words (=tags).
- When asking for the tag “Public Goods,” all the differences within it are completely eliminated.
- It’s difficult, what would be the best approach? (tkgshn)
- Moreover, clustering in general is an act of reducing information, so it’s not good to rely on human language here (u7693).
- Yes, if there is a method that does not rely on labeling with language, it may be possible to avoid clustering.
- Also, there seems to be an issue with the arbitrary setting of tag options.
- For example, by excluding certain tags from the options, it becomes easy to manipulate the plurality judgment in DeCartography.
- It overlaps with the previous issue, but generally clustering is often done with Unsupervised Learning, so using something like k-means should work? (u7693)
- https://ja.wikipedia.org/wiki/データ%E3%83%BBクラスタリング (u7693)
-
Clustering is a type of data analysis method (especially a multivariate analysis method). It is an unsupervised data classification method that automatically classifies given data without external criteria. Also, an algorithm for that.
- Since the goal is to evaluate the diversity between addresses, there seems to be no need to do clustering in the first place (blu3mo).
- That makes sense (u7693).- Speaking of which, it might be dangerous to assume that addresses are clustered (u7693).
- One issue with manual tag selection is that the information is reduced to a level that humans can interpret as words (=tags).
- My impression is that if a proper mechanism is created, it will work reasonably well, but it doesn’t seem elegant (blu3mo).
- However, I had some doubts about combining these two.
- The discussion on building a social graph and the mechanism of decentralized oracles is interesting.
-
Other possible approaches:
- For example, we could ask a group of people questions like “How similar are these two addresses?” and plot their answers in a high-dimensional space for distance calculation. This method could solve the problem of reducing information to the granularity of words. It would also avoid the subjectivity of tag selection, making it better (similar to pol.is).
- I see, that’s possible. If we repeat this process multiple times, we can imagine creating a two-dimensional or higher-dimensional space (tkgshn).
- Since the question of “How similar are these two addresses?” involves subjective human judgment, I thought it might be better to rely solely on machine-based judgment (u7693).
- I’m curious about your perception of “subjectivity” and the concept of Democracy (tkgshn).
- Well, it’s kind of separate. I wonder how it is with oracles.
- So basically, it seems like we are mapping it to a vector space that can be expressed in human language (u7693).
- I think it’s fine to rely on machines if they can achieve the goal. However, I feel that determining the “good” diversity in voting requires information from the human world (intuition) (blu3mo)(tkgshn).
- So, would it be necessary to use humans or natural language models?
- Using humans’ natural language models would be great (u7693).
- However, natural language models are also subject to arbitrary model selection, so it might be better to use crowds instead (blu3mo).
- Rather than asking the question “How similar are these two addresses?”, I felt that using natural language models created by crowds would reduce the subjective aspects (intuition) (u7693).
- This point can be generalized to “how to create a natural language model that eliminates subjectivity” and can be applied to distributed systems methodology (unknown) (u7693).
- [/tkgshn-private/Can distributed oracles be replaced with large-scale natural language models?](https://scrapbox.io/tkgshn-private/Can distributed oracles be replaced with large-scale natural language models?) (u7693)
- This point can be generalized to “how to create a natural language model that eliminates subjectivity” and can be applied to distributed systems methodology (unknown) (u7693).
- Also, I want to explore more about what “good” diversity in voting means (u7693).
- That’s true (blu3mo)(blu3mo)(blu3mo)(blu3mo)x100
- See below, 636b156f79e1130000b1e256
- Is it really good? (u7693)
- That’s true (blu3mo)(blu3mo)(blu3mo)(blu3mo)x100
- Rather than asking the question “How similar are these two addresses?”, I felt that using natural language models created by crowds would reduce the subjective aspects (intuition) (u7693).
- Well, it’s kind of a meta thing: [/tkgshn-private/DeCartography Whitepaper#6369fb4009c5f20000fd83e1](https://scrapbox.io/tkgshn-private/DeCartography Whitepaper#6369fb4009c5f20000fd83e1) (tkgshn)
- However, natural language models are also subject to arbitrary model selection, so it might be better to use crowds instead (blu3mo).
- I’m curious about your perception of “subjectivity” and the concept of Democracy (tkgshn).
-
Q. What is the purpose of tagging information in the first place?
- Is it about considering that two votes with different tags are more valuable than two votes with the same tag?
- That’s correct (tkgshn)(tkgshn)(tkgshn)
- So, it has nothing to do with the social graph?
- It seems like it’s about determining address diversity using both ①social distance information based on the social graph and ②tagging information (tkgshn).
- Is it about considering that two votes with different tags are more valuable than two votes with the same tag?
-
More about the philosophical aspect:
-
Understanding from (blu3mo):
- Based on the premise of [/tkgshn-private/50 actually independent thinkers are worth more than 1000 NPCs who all consume the same media and vote the same way](https://scrapbox.io/tkgshn-private/50 actually independent thinkers are worth more than 1000 NPCs who all consume the same media and vote the same way), which is a thought of vitalik, it is argued that votes from two people with different tags or from people who are distant in the social graph are more important.
-
Well, I guess that makes sense when you put it that way (tkgshn).
-
However, I would like to know more details about the basis of [/tkgshn-private/50 actually independent thinkers are worth more than 1000 NPCs who all consume the same media and vote the same way](https://scrapbox.io/tkgshn-private/50 actually independent thinkers are worth more than 1000 NPCs who all consume the same media and vote the same way). - 自分は追えていないが、vitalik周辺の人々が色々議論していてstableな土台があるんだろうなと予想 - この主張/思想の土台、referenceがあるなら知りたい(blu3mo) - 「多様が良いという思想は前提なので、その前提に関しての疑問はそっちを参照してね」というやつ - 例えばvitalik（誰でも良いけど）がこの主張の根拠を長文でかいてたりするなら、それがreferされてると良いなと思った(blu3mo) - (tkgshn)それは、結構むずくてEthereum財団とかまず見るのがいいのかな？ - https://www.neweconomy.jp/features/audible/164207 - クリプトのここ二、三に関しては - NFTサマー（2020） - DeFiサマー（2021）がきた後に若干落ち込んでる - 今までDAOとか作られてきたけど、思想面が全く非中央集権になってなかった（組織構造とかも含めて） - そこで、プロジェクトトークンがエクイティ（をただ、トークンにしたもの）みたいなところがあって、「これは単なる証券化では」みたいなのもあった - そこで、どんどん多様化してきて、やっと「もっと人間的にちゃんとしてるっぽいやつの方が良くね？みたいな - 戻ってきた感 - [/tkgshn-private/ソーシャルグラフやSoulboundトークンをベースに多次元ID（plural id）を構築し、Quadratic Fundingの脅威である”結託”を防ぐ](https://scrapbox.io/tkgshn-private/ソーシャルグラフやSoulboundトークンをベースに多次元ID（plural id）を構築し、Quadratic Fundingの脅威である”結託”を防ぐ)
- 何となくそんな気はするが、とても曖昧な主張に感じる - 全然ない、やってみないとわからん。だからもっといろんな人からつっこまれたい(tkgshn) - やってみてどうだったら、それが良いと言えるのかが知りたい(blu3mo) - （評価方法が定まってないなら実験をする意味がないので） - 究極、[[人力オラクル]]として「今までは[/tkgshn-private/SAD](https://scrapbox.io/tkgshn-private/SAD)としてQuadratic Fundingに対する[[談合]]を検知する費用も時間がかかっていたものが、この仕組みで早く・安くできるかみたいな(tkgshn)(tkgshn)(tkgshn) - [/tkgshn-private/Gitcoinがシビル攻撃を防ぐために掛けている費用感](https://scrapbox.io/tkgshn-private/Gitcoinがシビル攻撃を防ぐために掛けている費用感) - あ、Decartographyの目的は[[シビル攻撃]]の防止だった?(blu3mo) - いや、そこからだとユースケースが強そうかなという感じ。実際にGitcoinのチームも興味持ってもらってるなら、すぐに検証できるところでやれた方がいいかなと(tkgshn) - シビル攻撃①談合防止の用途と、②多様な投票を良いとしたい用途は多分別の話だと思うので、後者の評価方法もあるなら知りたい(blu3mo) - 例えば、つながりのない2人が同じ投票をしたとして、 - ①談合防止の目的の場合: 別に良い - ②多様な投票良いとする場合: 同じメディア見たNPCだろうから重み付け下げたい - という理解(blu3mo) - そうだよね〜それは分かるんだけど、まぁなんか疑問とかわからんことあるけどやってみないとわからんくね？というのが一貫した感想かな。むしろどうしたらいいと思ったりしてるのかを知りたい(tkgshn) - 「やってみないとわからん」は同意、ただやってみて結果が出た後に（特に②の論点に関して）それをどう評価するのかが気になった(blu3mo) - （例えば学会でAIモデルの研究を発表するなら、モデルを評価するところも含めて論文に書く） - crpytoの世界だと、結果をフォーラムとかに公開して評価は他の人に任せる、みたいな方法論だったりする?(blu3mo) - いや〜まぁそれはそんなことないんじゃないかなw(tkgshn) - 別に評価をしてもしなくてもいいし..
-
-
I haven’t looked into it myself, but I expect that there are various discussions among people around Vitalik and that there is a stable foundation.
-
If there are references or foundations for this claim/idea, I would like to know (blu3mo).
- “The idea that diversity is good is a premise, so if you have any questions about that premise, please refer to it.” That’s what I mean.
- For example, if Vitalik (or anyone else) has written a long article explaining the basis of this claim, it would be nice if it is referenced (blu3mo).
- (tkgshn) That’s quite difficult, but maybe it’s a good idea to start by looking at the Ethereum Foundation?
- Regarding the past two or three aspects of crypto:
- After the NFT Summer (2020) and DeFi Summer (2021), there has been a slight decline.
- Although DAOs and the like have been created, the ideological aspect has not become completely decentralized (including organizational structure).
- So, there are aspects where project tokens are like equity (but in token form), and there were also some arguments like “this is not just securitization.”
- As a result, there has been increasing diversification, and finally, there is a sense of returning to “Isn’t it better to have something that seems more human and well-organized?”
-
I have a vague feeling about it, but it seems like a very ambiguous claim.
- I don’t have any evidence; I won’t know until I try it. So I want more people to criticize it (tkgshn).
- If we try it, I want to know what criteria would make it good (blu3mo).
- (If the evaluation method is not established, there is no point in conducting an experiment)
- Ultimately, as a human oracle, it’s like “In the past, it took time and money to detect collusion against Quadratic Funding as SAD, but can we do it faster and cheaper with this mechanism?” (tkgshn)(tkgshn)(tkgshn)
- Oh, is the purpose of Decartography to prevent Sybil attacks? (blu3mo)
- Well, it seems like a strong use case from there. If the Gitcoin team is also interested, it would be better to test it in a place where we can verify it quickly (tkgshn).
- Sybil attack ① for preventing collusion and ② for wanting diverse voting are probably separate discussions, so if there are evaluation methods for the latter, I would like to know (blu3mo).
- For example, if two unrelated people vote the same:
- ① For the purpose of preventing collusion: It doesn’t matter.
- ② For the purpose of diverse voting: They are probably NPCs who saw the same media, so I want to give them less weight.
- That’s my understanding (blu3mo).
- Yeah, I understand that. But well, I have some questions and things I don’t understand, but we won’t know until we try, right? That’s my consistent impression. I actually want to know what you think we should do (tkgshn).
- I agree with “we won’t know until we try.” However, after trying and seeing the results (especially regarding the second point), I was curious about how to evaluate it (blu3mo).
- (For example, if you present AI model research at a conference, you would include the evaluation of the model in the paper)
- In the world of crypto, is it like publishing the results on a forum and leaving the evaluation to others? (blu3mo)
- Well, I don’t think it’s like that, haha (tkgshn).
- It’s not necessary to evaluate it or not.- Hmm, first of all, we should validate it as an Oracle data source and then see if Gitcoin can use it effectively. That’s the idea for now (tkgshn).
- I agree with “we won’t know until we try.” However, after trying and seeing the results (especially regarding the second point), I was curious about how to evaluate it (blu3mo).
- Yeah, I understand that. But well, I have some questions and things I don’t understand, but we won’t know until we try, right? That’s my consistent impression. I actually want to know what you think we should do (tkgshn).
- For example, if two unrelated people vote the same:
-
Well, in that case, it makes sense to take the hypothesis and throw it out to the forum (blu3mo).
-
The flow would be something like a Graph API, right? Something that businesses can use (tkgshn).
-
I only understand that much, haha (tkgshn).
-
From there, the community will evaluate it (or unfortunately, it may be ignored) (blu3mo).
-
Yes, exactly. We really don’t know, and it’s also a matter of luck where we can get feedback. But it seems interesting, right? (tkgshn)
-
But is there a high “expectation value” or something like that? (blu3mo)
-
I want to know more about that or have it explained (blu3mo).
-
Why do you think the expectation value is high?
-
It feels like there’s an atmosphere of “If you provide an Oracle, you’ll make money” (tkgshn).
-
[/tkgshn-private/Understanding Oracles](https://scrapbox.io/tkgshn-private/Understanding Oracles)
-
It seems like a good idea to implement it for now, expecting serendipity, but it would be better to have an idea of what we are evaluating in order to take the optimal action (blu3mo).
-
For example, a research question like “Does the diversity of ideologies manifest in the social graph and tagging?” (blu3mo)
-
[/tkgshn-private/Assumptions about wallet transactions](https://scrapbox.io/tkgshn-private/Assumptions about wallet transactions) touches on this definition. Also, it’s mentioned in the original draft of the whitepaper (tkgshn).
-
It seems interesting to explore the difference between academic methodology and crypto methodology (blu3mo).
-
Pairwise coordination subsidies: a new quadratic funding design could be a useful reference for thinking about collusion (tkgshn).
-
[/tkgshn-private/Pairwise coordination subsidies: a new quadratic funding design](https://scrapbox.io/tkgshn-private/Pairwise coordination subsidies: a new quadratic funding design)
-
It’s not about Sybil attacks, but preventing collusion (tkgshn).
-
For example, if we were to present a counterexample, let’s say 99% of people think A, but 1% think B. Does B get reflected in the tagging? (blu3mo)
-
I think the goal is to reflect minority views the most, but it’s difficult to present them as tagging candidates (tkgshn).
-
(This is more of a critique of the system rather than a counterexample, so the focus is different) (blu3mo)
-
“A. Not reflected.” Basically, since it’s permissionless, anyone can fork it without permission, and if it’s logical, it’s natural to design a process for objection (tkgshn).
-
So, if we have transparency about how the judging process works, we can negotiate that with Gitcoin, I guess, but I’m not sure (tkgshn).
-
I don’t have knowledge about the crypto world, so it’s difficult for me to come up with good counterexamples (blu3mo).
-
Actually, counterexamples can’t be pointed out without experimentation, so it’s more about presenting the points of discussion (blu3mo).
-
Q. Does the diversity of ideologies manifest in the social graph and tagging?- I wonder how far the mechanism of “thinking what people think” will survive. (tkgshn)
-
And so on.
-
It seems that having a strong logical connection and counterexamples here would make the DeCartography mechanism more convincing.
-
As a starting point, I remember that DeCartography was about finding the variables of QF when distributing funds, and it doesn’t make sense if it’s only the votes of certain project stakeholders. (u7693)
-
In other words, the “good” diversity in voting would vary depending on the votes. (u7693)
-
In the case of DeCartography, it is good for the voters to be involved in diverse projects. (u7693)
@kanair_jp: It’s interesting that there is already a recognition that algorithms choosing something is fairer than people choosing it arbitrarily. Creating trustworthy AI technology is often discussed, but it may be that AI itself becomes the trusted entity in society.
The assumption of treating a physical body as ‘one person’ doesn’t quite fit