Omoikane Study Group
Up to now.
ChatGPT's API is now open to the public, so I talked about doing it.
Today I'm going to talk about the Omoikane Project I've been working on recently. For Thumbnails:.
https://scrapbox.io/files/64cbd8e0f223a1001c80bbc8.png
But first, a topic of current events:.
'Scanning my retinas gives me virtual currency? What does that mean?" I see people who are like.
World ID, a privacy-preserving digital identity designed to help solve important, identity-based challenges, including proving an individual’s unique personhood (DeepL) World ID is a privacy-preserving digital identity designed to help solve important identity-based challenges, such as the unique Proof of Humanity of an individual. (I haven't read the protocol, it will be a big Identity Provider nishio.icon)
Worldcoin token (WLD), where laws allow, the first digital currency to be freely distributed to people for just being a unique human* (DeepL) World Coin Token (WLD): the first digital currency* that is freely distributed to people just for being a unique person, as far as the law allows.
Every adult American citizen should receive a share of the U.S. GDP. The sources of wealth are 1) companies, especially companies using AI 2) land in fixed supply This one is also in the context of UBI. Basic income for people proven to be human by World ID
network effect
Cites "the unique Proof of Humanity of the individual" as a "key challenge based on identity." SybilRank is computationally efficient... Scalable to graphs with hundreds of millions of nodes
https://gyazo.com/cdc9b35a99ccc5919bdb81c4ffd7dd33
"Trust Seeds", an algorithm that propagates trust from highly trusted accounts. Efficient algorithms have been created, and it's time to make it practical to apply them to the entire human network.
How can "accounts that correspond to retinal scans" be "Trust Seeds" with a high degree of trust?
It is more beneficial for an individual to be part of a larger network
Businesses that have built large networks can compete favorably as platformers.
Elon Musk tried to buy the giant social network Twitter and turn it into X
Sam Altman's strategy is to take a platform by scattering
Like PayPay [10 Billion Yen Giving Campaign Disperse through virtual currency payments issued by the company itself.
The expectation that the virtual currency will create value in the future allows for cost-free disbursement.
Smart...
In a nutshell, "a project that combines SF prototyping and AI technology to create a better discussion mechanism."
Democratic Inputs to AI," which OpenAI called for in May, which we talked about last time. AI has far-reaching economic and social implications
It is not right that decision-making of "what AI should be" is done by "a few people" (like the management of a for-profit company). It is interesting to look at the Worldcoin news at the beginning of this article in light of this statement in May
The founding members of the Omoikane Project have repeatedly discussed applying for this [Democratic Inputs to AI
We had a very interesting story, so we started working on it without waiting for it to be adopted.
Unfortunately, the result was not accepted, but by that time, interesting things were happening in the online community, so we decided to continue with fine-tuning the plan.
We talked about Polis in the last and previous workshops. Polis is a beneficial component
However, it is not a stand-alone deliberation support system.
Pre- and post-stage design is required
https://gyazo.com/def130484660e621bd11bf759c7c88c0
human.iconWhat is SF, Space Fantasy?
nishio.iconScience Fiction here is Science Fiction, but I don't think it's worth worrying about since the term itself doesn't have a clear definition to begin with.
If I had to describe it in my own words, I would say that it is "a story of what is likely to happen in the near future.
What is Prototyping?
The idea is to create a rough model that represents the concept using materials at hand. You can use blocks, foam core, tubular material, strong adhesive tape, whatever seems useful.
The idea is to get the "experience" of having the product in your hands faster by making a low-cost "approximation" of the final product
So how can we prototype not a physical product, but a vision of the future and the direction we should go in the future?
The method is SF prototyping.
To put it crudely, "make an approximation of the future by means of a near-future science fiction short-short."
Can sympathize or rebel with the protagonist's "experience".
human.iconInteresting interpretation that SF can be prototyping.
Moreover, this process can be assisted by AI.
https://gyazo.com/a9cbed512af0304482aea9a095782782
In fact, several people in the Omoikane project have published their own AI-generated science fiction novels
All different authors.
Surprisingly easy, let's look at some concrete examples
Prompt: you are a science fiction writer. Write a near-future science fiction story. Constraint: The protagonist works for a software company whose motto is "creating a society full of teamwork. The story begins when he hears a researcher claim that prototyping the future by having an AI write science fiction is beneficial.
Title: "Future Prototypes".
Chapter 1: "Encounters with AI"
The main character, Itsuki, is an engineer at the software company TeamWork Tech. His creed is to create a society full of teamwork. One day, he becomes interested in a researcher's claim that AI is prototyping the future by writing science fiction. His heart leaps that this may be the key to shaping a new vision.
Chapter 2: "Teamwork and Future Projections"
The tree will launch a project to find ideas for how to utilize the AI's capabilities to instill teamwork in society. That AI, named Asimov, Asimov's task is to write a prototype of the future based on the data and goals provided by Tree and his team.
Chapter 3: "The Future as Depicted"
The future envisioned by Asimov is one in which collaboration shapes society. Individual ideas intersect and generate new ideas. But Asimov's story also contained a warning: a warning about the shrinking role of humans in a future dominated by AI.
Chapter 4: "The Need for Coordination."
Trees will challenge against this future envisioned by AI. Emphasizing that teamwork stems not only from AI, but also from the connections between humans, he seeks a future that uses AI's capabilities but does not lose its humanity.
Epilogue: "The Future of Teamwork"
Ultimately, Trees and Team envisions a future that combines human creativity with AI's predictive capabilities. It is a future where teamwork takes center stage and AI provides support. And it is also the beginning of a new science fiction, a prototype of the future.
This story illustrates the possibilities of AI and humans shaping the future together and offers readers a new vision of the future.
Comments.
human.iconIs this an unadjusted AI-generated text on the site? That's a pretty interesting sentence.
nishio.iconYes, I didn't modify anything except the indentation, no trial and error, one shot out: chatlog "AI and humans will shape the future together."
That's a nice phrase!
Collaboration as "the intersection of individual ideas to generate new ideas."
"The future of teamwork" / "A future where teamwork is central and AI provides support."
What does "teamwork" look like in this world? It could be something different from the "before GPT" teamwork.
Teamwork... also derives from the connections between people.
I feel like this is an important perspective.
Would it be interesting to build a sci-fi prototype on the issue of the unclear definition of "teamwork" and the future of teamwork?
Postscript:.
"Write future prototypes based on data and goals."
No data or goals were given in this prototyping process.
Sure, it's possible to make better prototypes by giving them data and goals...
human.iconIt sounds very plausible as a synopsis, but I'm not sure what exactly it will be about.
Is this good enough as long as it is prototyping?
nishio.iconI think the prototype is complete at this stage.
The "novel as entertainment" and the "novel as prototype" offer different kinds of value.
Entertainment fiction doesn't have to generate controversy, as long as it's interesting to read.
Interesting, but nice to have a MUST to make you think.
Novels as prototypes are valuable for generating discussion, even better if they are interesting, but just being interesting is futile.
Make you think is MUST and interesting is nice to have.
Since GPT4 is writing the story instead of me thinking up the story, unexpected stories and phrases come up for me.
This time I read it myself and wrote my thoughts, but I can do that because I'm not 100% in control of the story.
Like, "I see what you mean, it's important to be based on data and goals."
I can't share with others if I'm just mulling around in my brain, but I can share this sci-fi prototype and the verbalized impressions triggered by it.
Sharing triggers thoughts in others.
If doing a vague "let's all think together" or "let's all come up with ideas" doesn't work very well, then making something readable first will lead to more concrete discussions.
So I suggest we do SF prototyping first before discussion.
I'm already convinced that this would be beneficial to me, after trying it myself first in making my proposal.
The cost of the work is lower than expected, and when you see what comes out, you say, "Oh, well, I didn't have that perspective.
Do you think this usage is teamwork with AI?
I think it's a new form of teamwork, but it's also an interesting story, and I think the world is divided on it!
On the other hand, there are those in the world who believe that tools should not be allowed to have personalities and that AI's capabilities should be limited, and from that perspective, AI will never be a team member.
I believe that AI can be a team member and that we should have more conversations with AI as a team member.
Even when used alone, ChatGPT will create unexpected stories to activate it.
Share your prototype with others to stimulate conversation. This explanation, which I just introduced, was given in "Chapter 4: Ultimate brainstorming". The idea is to create a rough model that represents the concept using materials at hand. You can use blocks, foam core, tubular material, strong adhesive tape, whatever seems useful.
Instead of "thinking," "ideating," and "discussing" in a vague way, making things visible and touchable allows for more concrete thinking, more concrete ideas, and better discussions.
So, in creating a better discussion mechanism(deliberation support system), it is a good idea to "do SF prototyping first before discussion". This was only an idea at the proposal stage, but became a conviction as we tried it out.
The cost is lower than expected because prototypes like the above example are really quick and easy to produce.
There are two supp projects that will delve into this.
Create search keywords based on SF prototypes and load the prompts with the abstracts of the papers hit by the search with those keywords and have them refined.
This is the direction suggested earlier in the SF prototype "write future prototypes based on data and goals".
Workshop for teachers scheduled for the end of September
I had seen it in nishio.icon's project and knew it existed, but felt the hurdle and hadn't been able to vote for it.
Would it lower the hurdle if it's your opinion of a creative work?
Consideration 1: With political topics, there are cases where people hesitate because they think they don't have enough prerequisite knowledge to vote.
If it's "please read the novel and vote on what you think," then once you've read the novel, you can assume that you have enough prerequisite knowledge.
Consideration 2: The hurdle was high because of the "must write a correct opinion" atmosphere
This time, by being called "impression," I can express how I felt, not how "right" I was.
Q&A
What are the specific ways in which AI will assist? For example, in what areas will the AI assist with plot generation, character development, dialog creation, etc.?gpt.icon
Personally, I like the use of having a near-future science fiction plot created, cost-effective.
This is because I consider "effectiveness" to be "prototyping".
Prompted for various keywords, "Explain in English for people who are not familiar with Japanese culture.
"Convincing" is a richly nuanced Japanese expression that cannot be directly translated into English.
The Japanese word "nuisance" (meiwaku) translates directly to "nuisance" or "trouble" in English. However, the cultural meaning of the word goes beyond these translations in the context of Japanese society. In Japan, social harmony and collectivism are greatly emphasized... What specific insights or lessons learned from the written science fiction stories?gpt.icon
In a setting where local government has gone digital due to a lack of manpower, we created a storyline of those in favor of digitalization and those opposed to it.
At first I subconsciously thought that older people who resist change are unhappy by their choice.
When I saw that ChatGPT generated a happy ending with that constraint, I realized I had an unconscious bias. What are the main benefits of using AI to create stories?gpt.icon
To create a near-future science fiction story is to start from the current situation and create a prediction of the future by connecting small steps that "such a thing is likely to happen.nishio.icon
Some people are not very good at making these future predictions in the first place, or they make narrow predictions based on unconscious assumptions.
Mechanical support here has the effect of reducing costs and broadening horizons.
This is the first half of the story so far, so I'll question it.
Post-Polis phase
https://gyazo.com/c24a617eaefe6f4124accce7f2cb7d18
Minimal product" that can be learned by releasing it to others and observing their reactions.
By reading the novel and doing Polis, you can realize that "different people feel differently even though they read the same thing".
https://scrapbox.io/files/64cba5450400ec001cd713bd.png
I voted and was in the minority, so I wanted to write my opinion.nishio.icon
After that, I wanted to make SF based on that opinion, and this became the first SF prototype for nishio.icon!
I wanted to improve this "Novel to Polis" process.
Polis does not show visualization until 7 people have voted, and there is no feedback for those who have taken voting action, so it is difficult to increase the number of votes in the middle
For example, opinions on why the clusters are divided into clusters are shown on the visualization screen, but this cannot be retrieved via the API.
It would be interesting to have the AI role-play the opinion clusters as virtual personas, but the only way to do that is to copy-paste them into each other.
The needs in this area led to the list of "areas for improvement" in the previous Polis Study Group. This "improved Polis-like system" was later named "Komorebikagami".
A mirror that collects and reflects (reflects/contemplates) the "emotional data" that leaks out of people like sunlight leaking through a tree.
Japanese cockchafer (Melolontha japonica)
✅Return immediate feedback on ballots
✅You can check your own past votes.
✅Can update ballot content
I think of this as "a system that helps people update their thoughts", it's only natural that people with updated thoughts would want to update their votes, so I supported that.
https://gyazo.com/3598f4f5c65e11b910a734ff7400eb2c
(Planned) to be able to share on Twitter and vote directly...
I'm still waiting to see if Twitter will turn into X or not...
You can vote against Bluesky with a script, but...
In the end, it's not about which social network, but about being able to share easily on each platform where people are congregating.
In the case of Polis, it was "a visualization is read by a human, a thought is triggered, and a post is written (by some people)"
Polis clusters are virtual personas, so GPT could read them directly and role-play them.
Virtual personalities played by AI's could argue with each other.
Community-centric AI
The "Building Community-Centered AI" project is an attempt to create a "Community AI" (Community-Centered AI). This refers to an AI that learns about a particular community's history, culture, challenges, and people's opinions to understand and represent them. Specifically, it will interact with people involved in the issues facing the community, providing an environment that not only provides knowledge, but also creates a forum for discussion and the creation of solutions. Our initial goal is to create an AI for the Omoikane community. As a result of the iteration, a lot of data was accumulated by community members, and we thought it would be interesting to turn it into a chatbot with Retrieval-Augmented Text Generation. To use a figurative expression, "the AI is watching every conversation in the community."
When spoken to, "remember the similarity" and respond with reference to it.
How it works, which I discussed in March at [Story study session about connecting my Scrapbox to ChatGPT.
At that time, the vector index was updated manually.
They only experimented with personal Scrapboxes.
This is "automatically updated once a day" for "community groupware that is updated by multiple people".
I started automatically updating from the Omoikane Project's Scrapbox once a day in June.
✅Vector search function is already in service release, chat function is not yet added.
Essentially, there is no need for it to be Scrapbox.
It would be good if the text data of the exchange could be automatically extracted from the communication platform (= groupware) that the community primarily uses.
I'm using Scrapbox right now because it's convenient and automatable to export in JSON format.
Scrapbox is also a good place to find several communities in the public
Groupware for corporate use is generally not allowed to disclose its contents.
If a public community is used as a test case, it is easy to introduce a case study that says, "This is what happened when we introduced the system to this kind of community.
As a start, anything with text data is good.
For example, the Taiwanese civic tech community g0v has its minutes published on HackMD, which you can crawl Cybozu style article crawls, it becomes a chatbot that talks about Cybozu culture But even if you build a chatbot that learns these things as a start, it's not very interesting if you can't build a feedback loop for it.
Building a feedback loop is important.
What is a feedback loop?
https://gyazo.com/e2ef115531a07b58e362c723f59f3e16
1: AI reads groupware
2: When someone speaks to you, answer based on your memory of 1.
3: This exchange is automatically written to the groupware.
4: Vote on it in Komorebikagami and additional discussion will take place.
✅The current implementation is to update the vector index (1), then the AI will post its impressions of the past day's updates to Scrapbox (3'), and then you can write your impressions (4').
[Reinforcement learning from human feedback, from Human in the loop to [People in the loop Anonymous Comorebikagami for AI output is superior in three ways
Compared to "the case against human speech."
No hesitation to argue because the opponent is not a human but an AI.
Compared to verbal rebuttals or submissions with apparent names.
I am anonymous so I don't have to worry about putting my opinion out there.
Since it's a "yes/no" vote
The "community atmosphere" changes from "vague and only vaguely understandable" to one where questions can be explicitly asked.
You can ask questions without social risk, and the AI will answer them first.
Whether the answer was good or not, and how it should be improved, will be the subject of discussion in the community.
In this process, no one human is attacked and the "virtual character" called AI is "improved by everyone" cooperative game. Bringing "a better discussion mechanism" in a Japanese culture that values consensus rather than top-down direction, but fears conflict with others. Also, the GPT portion absorbs language differences, which should support cultural integration of teams across different language areas
This is what Nishio wants to create with the Omoikane Project
However, the Omoikane Project is a group of volunteers, unlike a corporation, so it may be more difficult than in the case of a corporation.
Participation is open-ended, allowing participants to work freely based on their own interests.
It increases diversity and creativity, leading to new discoveries and results.
When looking back at the past after a period of time, it becomes clear "what was produced".
We cannot predict in advance what will be born, we can only create an environment conducive to birth.
The "Thinking and Making the Future While Reading and Writing" Project
The most meta of the Omoikane project's sap projects (by /omoikane/niryuu.icon)
SF prototyping is a tool that allows us to think more concretely and deeply about the future by concretizing the future in the form of a novel.
But this "act of reading and writing SF" of "SF prototyping" itself is already evolving into a new form in this Omoikane project
The act of writing science fiction
AI assistance has lowered the bar.
The writing process, which used to be handled in a black box manner in the minds of science fiction writers, is now shared as a chat log with GPT.
And then people who had never thought they'd ever have the time to write their own novels got involved in writing science fiction.
The act of reading science fiction
If you read science fiction in a book, your impressions are made in a different medium than the book
But by reading on the Scrapbox, which can be co-edited, your impressions become content on the same page as the science fiction, and conversations with other readers and links to other science fiction occur.
Voting on what you read in Polis and Komorebikagami, and being counted and clustered is an experience not found in existing novels.
And many other things going on that I can't classify.
Add development in the form of answer songs to those made by others, and use image-generating AI to create illustrations.
Interactive fiction, abstracting and transforming the novel
Clearly a unique social interaction is occurring here.
This is the emergence of a new culture. In other words, a new future is being born here.
A project to observe and understand this phenomenon itself from a social scientific perspective.
It's more legitimate peripheral participation than I'm aware of.
The acquisition of this "way of writing" is not only through abstract knowledge transfer, but also through participation and experience in specific contexts and situations "Situation-embedded learning." Everyone is welcome to participate.
Currently 45 people, 614 pages
---
This page is auto-translated from /nishio/オモイカネ勉強会 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.