How to Scale Digital Democracy?
How to Scale Digital Democracy?
https://gyazo.com/e0cbca9528fb418c1248e823bbcfbba9
For Thumbnail
Plurality Tokyo Namerakaigi Video
https://www.youtube.com/watch?v=07BQsWtF5KA&list=PLHsuZp6_Tsv9RZnyar7J7F8dpbHDwX5Uf&index=5
Cybozu Labs Study Session Video
https://youtu.be/9US9vrS18Yg
impetus
tkgshn.icon I need you to speak for about 20 minutes on the topic of "Scaling Digital Democracy through Community" (how to achieve large-scale collaboration using Pol.is and Community Notes examples). I usually talk for an hour, so this time it's 1/3 of an hour, so it's compact.
I share my thoughts from the perspective of understanding what's in the implementation.
No amount of time would be enough to explain the implementation details.
I don't think of democracy as something vaguely good.
To improve democracy means that there are bad things about "democracy in 2024."
We need to think about "democracy" not in a vague way, but break it down into details, where it is good and where it is bad we need to dig deeper.
Use the better leg of the current location as the axis leg, without moving the better part, and move the other leg toward the ideal.
5th century B.C.
Discussed at the meeting and decided on policy by majority vote
It's nice to discuss and decide together."
Who is this 'everyone'?"
Adult males who are citizens by birth. Excludes women, slaves, and aliens.
Approximately 10-20% of the total population, or about 30,000-60,000 people
By the way, Tokyo Dome holds 55,000 people.
Ishigaki Island, Miyako Island, Sado Island, and Amami Oshima have around 40,000-60,000 people.
I like the idea of giving every citizen the right to vote."
Who is this 'everyone'?"
Adult Men.
Women's suffrage in Japan, 1945, not much different.
They talk about the "one man, one vote" principle, but that hasn't been accomplished even at this point in time.
Isn't it against the principle of one person, one vote to limit it to adults?"
"Minors do not have the capacity to make rational political decisions because they do not have the judgmental capacity to make rational political decisions."
What about seniors with dementia who have the right to vote?
Even though "adult wards who are judged to have insufficient capacity to make decisions due to dementia or intellectual or mental disabilities" have the right to vote since the 2013 amendment to the Public Offices Election Law? (see Voting rights of adult wards) You claim that 17 year olds are less capable of making decisions than those people?
Why not give equal voting rights to all human beings, including 0-year-olds, and delegate to their parents or guardians those children who cannot express themselves?
Maybe we should change the one-person, one-vote principle, because it's not really that important."
and radical reforms proposed by Glen Weyl Quadratic Voting (QV) Three slightly different things came out of this inspiration, so let's sort them out here
QV of [Audrey Tang
QV of [Glen Weyl
A social implementation of Glen Weyl's QV described below, stripped down to fit reality For example, in voting for the Hackathon Excellence Award, one person is given 99 votes.
Voting strength is the square root of the number of votes cast.
If you cast 64 votes, you have 8 voting power
If you cast 16 votes, you have a 4-vote power.
You can make better use of your vote by voting for a variety of candidates in small increments.
The interesting thing is that they have "99 votes" instead of "100 votes".
I can't vote for one candidate for all of them because it's too close to the end.
If a friend said to me, "I'm in this contest, vote for me," if it were a conventional method, I would only vote for that friend.
This way, you get 81+16+1+1 or something like that, and your friends get 9 voting power, but you also get 6 voting power for other projects.
The incentive is to look at multiple projects with high synergy and naturally try to find the "optimal combination.
"Combined, they are worth more than the sum of their individual parts" = synergistic combinations emerge through voting behavior.
Proposal for QV in the first place
'You can cast more than one vote if you want, but the cost will be the square of the number of votes cast.'
Where did that "plurality vote" come from?
The proposal depicted in Radical Markets is "the continuing value of the right to vote." Don't you all take it for granted that if you don't vote, your vote is worthless?
Current voting rights have no "continuing value".
In every election, the question is, "Does it make sense to cast a white vote?" There is a controversy that
What to do when there is no candidate you want to vote for?
For example, there is a favored incumbent and no decent opponents.
Glen's QV offers a simple solution
We can save those votes and use them next time."
The current voting system ignores incumbents because votes not cast or white votes have no value.
However, with QV, "votes stocked without being voted" are voted for "opponents who might be able to defeat the incumbent who appears in the future," so the incumbent will do his best to increase his vote share.
In terms of social implementation...
(Aside from the fact that no incumbent may want to include this.)
Technically, it's not easy to keep secret ballots and keep the unused votes belonging to the individual and surviving.
Audrey's QV facilitates social implementation by discarding continuity value.
'Tokens of continuing value? We already have that, don't we? It's gold!"
I get an allergic reaction when I describe it as "buying votes with money."
Let's look at it as crowdfunding and a guessing game for idols.
Functioning as a social implementation
AKB48 General Election 2009]
Fans earn votes by purchasing CDs.
At this time, "number of tickets purchased = number of ballots = voting power".
If the voting power is the square root of the number of ballots, it becomes QF.
Established in 2017
Start redistributing funds to open source public goods through Gitcoin Grants around 2019. Approximately $700,000 in the first year, followed by rapid growth and redistribution of approximately $21.4 million in 2022 (link). What's wrong with buying votes with money in the first place?
Intention of the title "for Funding Public Goods
https://gyazo.com/7de5f0eae25deb83ad3ae020628a2fd5
In QF, 1 million yen (=1000 voting power) from 1 person and 100 yen (=10 voting power) from 100 people have the same power.
https://gyazo.com/b7b59fcc5cc066b7adaf26378206f3f6
1,000,000 yen goes to the treasury and will be redistributed in the future, 10,000 yen per person if divided among 100 people.
People who think plan A is good are paying 10,000 yen each to people who think plan B is good (they can pay up to 100 yen).
Even those who think Plan B is better are paying 10,000 yen for one person who prefers Plan A, divided among 100 people.
For irreconcilable decision options, "pay more for everyone who makes you drink" is chosen.
Voting and Crowdfunding
Reflection on the past
Many people associate the word "democracy" with the "voting" "ritual" in "Democracy in 2024" of "writing the names of politicians on paper and putting them in a box.
But from the perspective of designing social mechanisms, we talked about how voting and crowdfunding have a common structure.
Both are "mechanisms for gathering information from the many and consolidating opinions."
We need to take a closer look and break down the concept of "voting" from this perspective.
The "ritual of writing a politician's name on a piece of paper and putting it in a box" is just one of the major implementations currently in place.
What was the vote?
In the past, there were no computers, so aggregation algorithms were undeveloped.
An easy-to-implement and widely used tally method is the majority rule algorithm: "write down the choices, vote, count the number, and choose the one with the most choices". The Athenian arcon (regent) was decided by lot.
They think there's a lot wrong with majority rule.
Plato "The judgment of the ignorant masses disturbs the state. The ideal is to be governed by those with expertise." Aristotle. "Popular rule disregards minority opinion and is easily swayed by emotion." James Madison "We need separation of powers and federalism to prevent tyranny of the majority." John Stuart Mill. "Simple majority rule could undermine individual liberty and minority rights." There are many problems, so let's focus on one of them this time.
Pointing out that "Voting is too little information to be sent in the first place."
I want to send more
But conventionally not accepted.
Starting around 2023, the development of LLMs (large-scale language models) is enabling better aggregation algorithms
Experiments are being conducted on a variety of applications, and various results are being achieved.
https://gyazo.com/8aed1a6ee239c672d1e504cdb48d0e9e
These two are ballot-based methods that do not use LLMs
Polis became famous when Audrey used it in vTaiwan.
People vote "for, neutral, or against" an opinion.
Opinions themselves can also be voted on (depending on the setting).
Community Notes is used by X/Twitter. GitHub People vote "useful/useless" for the notes.
You can even post the note itself (after accumulating some trust points).
Processing Mechanism
Polis
Missing values are filled with the mean
PCA (Principal Component Analysis) to be 2-dimensional = people are placed in 2 dimensions.
Then clustering by k-means
CN has a problem with very many missing values.
Learning a "very simple neural net" using the given data as the teacher data (abridged explanation).
As a result people are placed in one dimension(= roughly in line with the political left and right)
Make the weight of people's political bias and unrelated note support the note score.
Talk to the City (TTTC) is an LLM-based methodology
Appeared in early 2023, then released as open source in October 2023
In Japan, Mr. Anno used it in his campaign for Governor of Tokyo, Nittele used it in the Lower House election, and the Tokyo Metropolitan Government used it to formulate a long-term strategy, and it is gaining recognition.
Processing Mechanism
LLM converts people's opinion text into a higher dimensional vector
Make it 2-dimensional for visualization purposes (using UMAP) The two dimensioned items are clustered into groups, and the LLM generates a description of the groups.
https://gyazo.com/1e1a56060aab7c997e9d0999fab108a8
Polis and Talk to the City fusion line
The first half is voted on like Polis, and the LLM generates commentary on the groups created, like TTTC.
Overlaying "opinion vectors" extracted from political party manifestos is a feature not found in conventional Polis or Talk to the City.
Data collected and visualized for 4,400 people in the 2024 House election: GitHub TTTC, Public Opinion Map and Beyond
What features were in Polis, not in TTTC, and what features did the poll map cut from Polis?
It's "the ability to post comments after viewing the visualization"
Currently TTTC and public opinion map "stops at visualization"
https://gyazo.com/21beb7358d7da00e2bf24eccec79fbe9 by mashbean (I'm not sure I agree with the diagram itself from an engineering standpoint because of the functional overlap due to Polis and TTTC being black box components.)
I agree with you when you say "There needs to be an interface where users who see the visualization can be triggered by it to write their opinions and vote..."
The loop should keep spinning all the way around as an ongoing mechanism.
On the other hand, as a realistic project, it is easier to make it a lively event with a clear end.
How to fill the gap here is a challenge.
Cooperation has depth and breadth, these are tradeoffs, and technological advances push the range of what is possible.
Originally titled "How to Scale Digital Democracy?"
https://gyazo.com/a7e741fc4fcebd8ba02b51f997ce24f5
The "voting" that is often associated with the word "democracy" has already been scaled considerably.
Wide but shallow
There's also the line, "Let's broaden it a bit more, maybe minors, foreigners, etc."(1)
Audrey's "Voting is too little information sent in the first place, isn't it?""More information" is a deeper approach(2)
From "Choose from the candidates" to "Vote directly for or against the agenda" or "Submit your opinion in writing".
(The fact that JOIN allows high school students to send out messages also contributes in the direction of expanding it.)
This direction stands out.
Because we are starting from a "wide" place.
There are other approaches.
https://gyazo.com/1d9ebb5769c6c0e31e2d04831b3cc998
Productivity of discussions when you are talking to two like-minded people.
Productivity when discussing in small groups
How can we "deepen" (3) and "broaden" (4) this further?
Operated 24 hours a day for 5 days with no location restrictions and answered 6,000 questions.
Unlike humans, AI ans have no physical limitations.
Work and child rearing.
https://gyazo.com/4e453420bbbc08b3a9671fa498ef022d
Information sharing by being able to see people around you instead of having a one-on-one conversation with the AI
Allows the general public to see "what questions other participants are asking and what responses they are getting".
https://gyazo.com/de2c94f79f955838d374e0cc904e2eb5
Unlike (1), where AI only talks and does not listen to people, or (2), where AI listens to people but does not feed back its data, AI An has been structured as (3) from the beginning.
This is one of the group discussion methods, and is conducted at Cybozu with more than 100 people.
Participants propose a theme of their choice in front of everyone.
Everyone goes to the booths on topics of interest to each of you.
No one is forced to participate in a discussion in which they have no interest.
However, if you set an unpopular theme, fewer people will talk to you (Vitalik's subjectivism). (Depending on the number of people and time constraints, in the case of Cybozu)4 sessions of 30 minutes per session, there were 13 booths.
https://gyazo.com/9907d2b196a02dbb72d9f266e9f7880f
A society is made up of a number of overlapping groups, not a single mass of society, nor individual people in isolation.
I think Open Space Technology fits very well with this idea of the Plurality book.
Dividing 100 people into 13 groups averages 7-8.
Close discussions in small, face-to-face groups.
Doing it four different times with four different groups would put you in four different groups.
Group Discussion Information Sharing
There are many situations to discuss in groups, not only in Open Space Technology.
There is a strong need to "know what other groups are discussing."
But information sharing there is not smooth.
How can I get a good summary of the conversation in a group discussion to be shared?
https://gyazo.com/b396aeb578a17b1fcfb5309bb84680dd
For example, one participant in each group will record the event with his/her smartphone and upload it to YouTube unlisted after the event is over.
YouTube will do the transcription for you.
So far, at this point, we already have the tools and can make it happen operationally.
Transcriptions are collected programmatically and summarized by AI
If you create a code to collect, you can create a summary report with Talk to the City.
It would be good to summarize with a link to the timestamp of the video so we can quickly check the original discussion.
Discovering that different groups are talking about related content and teaching the participants in each group
This is technically possible, no implementation yet.
AI feedback in a few hours is possible at this point, and the time it takes will become shorter in the future.
https://gyazo.com/3aeba9eb57caa45165af87dccbc57406
New forms of communication being researched by Mr. Aoyama (Bluemo) Continued from "Sharing Information in Group Discussions."
The smallest form of "human group discussion" is "a conversation between two humans."
Productive discussions between two like-minded people are "very narrow but deep cooperation."
When AI is able to adequately replace humans, it will be able to have equally deep conversations on its own.
Time Constraints" are no longer a factor = Free from Time Constraints.
"Group Discussion + AI" and "Human-AI one-on-one" both have advantages and disadvantages.
Group discussions provide an opportunity for "participation" even for those who do not proactively speak up.
Even if you don't have "something to talk about" from the beginning, you will find what you want to say as you listen to other people's discussions.
One-on-one between a human and an AI" doesn't require consideration of the other party's feelings.
https://gyazo.com/e0cbca9528fb418c1248e823bbcfbba9
There are diverse forms of communication, not limited to "one-on-one human-AI".
AI An's town meetings are one of them.
AI participates in a variety of communication and loosely bridges the gap.
What is "scaling?"
Making a small group bigger?
Is it to increase the average level of activity in a society by creating many small, active groups?
To loosely connect small groups of people with "information and communication channels that are neither 100% disconnected nor 100% shared"? summary
How to extend digital democracy "deeper and wider" for better social decision-making?
It is necessary to divide democracy into elements rather than a single black box
Voting = writing one's name on a piece of paper," "one person, one vote," and "majority rules" are assumptions.
A history of questioning "conventional wisdom" from Classical Athens to the present day
New voting and decision-making methods
Quadratic Voting/Quadratic Funding: Ideas to change the principle of one person, one vote.
Talk to the City: AI visualizes and summarizes huge amounts of text
Open Space Technology: Networking through intense dialogue in small groups
AI Anno: Town meetings free from time and place constraints
AI-based "hybrid dialogue"
Recording and transcription of group discussions → AI summary → Discovery of related discussions
New forms of discussion and psychological safety through "human x AI" communication
Consider what "scaling" means.
It should not simply be about increasing participation.
Isn't the goal to improve the quality of "cooperation" by achieving deeper discussions and wider collaboration?
Enriching society as a whole by connecting many of the intense dialogues that occur in small groups
Small Steps Starting at the Grassroots
We don't wait for someone else to make it for us, but introduce it ourselves, little by little, to suit "our place".
--- memo
Making
ver.2
I had exactly 20 minutes to practice my presentation, but I somehow got 15 minutes for my presentation. w
cut from ver.2
Concept of [MetaPolis
Visualization is not the end of the story.
Hopefully after seeing the visualization, people will be inspired to speak up by it.
Discussions "build up".
A system where a common ground for discussion is created and improved.
ver.1
How to achieve [large-scale collaboration
About Polis and Community Notes, technologies for scaling democracy
Replies do not scale.
1000 replies to one person's post
There's no way you can read.
And there's a high percentage of verbal abuse.
Mute all together
This mechanism is not enough for useful opinions to be exchanged and discussions to deepen.
TTTC
Advances in LLM have made it possible to filter out rhetoric and group similar opinions
You could say it's just visualization.
What are the next steps?
It's hard to function if you just pass it on the web.
+Broadcast
Nittele Case Study
Human commentary included.
Can be used as a springboard or topic for discussion with an expert (in this case, a politician)
+Internal deliberations
Tokyo Case Study
As a major public comment
More information will be available on 1/31 and I'll write that.
Polis
Some kind of novel UI
I won't let you reply.
Order is shuffled (information-maximizing algorithm, must be checked)
Not yet rooted in Japan.
The "Public Opinion Map" is a mobile-first UI redesigned for the Japanese environment.
The introduction should have been done by Mr. Tonfi first, so I won't do it here.
The internal tally server and other functions of Polis are used as they are, while AI-based cluster commentary and party icons are displayed.
Between Boat Match and Polis, which are relatively well rooted.
Public Opinion Map 2024
UI that is a little easier to understand
Submissions divided in favor and against.
No direct replies.
The good ones get upvoted.
Community Notes
matrix decomposition
What's the difference from majority rule?
Drawing and Explanation
Meta, Fact Check
On April 7, U.S. Meta announced that it will discontinue the use of independent fact checkers to examine the accuracy of posted content on its Facebook and Instagram sites. It will be replaced with a "community note" system that leaves comments on accuracy to users, similar to that used in the U.S. social media X.
Some people just cut out the "abolish fact-checking" part and say something like, "That's outrageous! but the way they cut it out itself is biased.
Fact-checking by a small number of people to determine what is correct does not work very well, is not trusted by today's social media users, is not efficient, etc., so it is better to "connect a large number of people with an algorithm to check correctness".
The following English-language papers on expert evaluations of COVID-19-related notes in community notes are available
"Do Community Notes work? A case study of COVID-19 related notes": in this study, COVID-19 related community notes posted between 2022 and 2023 were medical experts evaluated the COVID-19-related community notes submitted between 2022 and 2023 and found them to be 97% accurate.
This paper demonstrates that community notes are highly reliable in providing information on COVID-19.
A decent rebuttal.
Emotional rebellion, outbursts
---
This page is auto-translated from /nishio/デジタル民主主義をスケールさせるには？ using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I'm very happy to spread my thought to non-Japanese readers.