Feature article in Information Processing, the journal of the Information Processing Society of Japan
- Journal of Information Processing, Vol. 64 (2023) No. 9 “ChatGPT, a Tool for a New Era: Exploring Its Potential from 14 Perspectives”ChatGPT
- At the stage of approaching him, I told him that I wanted to put the manuscript in my Scrapbox, and he gave me his OK on the condition that it would be published after August 15.
- If you want to read the unedited version in the paper, just look at the paper, and I’ll incorporate the additions here that I couldn’t do for the paper.
From the subjectivity of one to the subjectivity of many
, not subjectivity or objectivity.
- AI-based knowledge co-creation = integration of individual perspectives ← This was written by GPT4
- Q: 「主観か客観か」というテーマに対して、筆者はどのような意見を持っていますか?
- A: False dichotomy だと思う
summary:
- Many people ask the question thinking that AI is the one who can give us objective truth.
- However, the author finds usefulness in conversing with beings who have a different point of view than his own.
- Truthfulness and usefulness are different.
- Diversity of viewpoints plays an important role in discovering blind spots.
- Developments in AI technology will enhance people’s ability to summarize and understand the opinions of many other people.
- This affects the structure of the organization and society.
- In an age when not only the “subjectivity of one person” but also the “subjectivity of many” becomes the object of thought, the form of intellectual production will also change.
- How would it change?
1. New co-creation experience with AI
.
1.1 Challenge to cook with chestnuts in the fire
.
-
When I heard about this project, the phrase “picking up chestnuts in the fire” came to mind.
- The strike was 3/4.
- On 3/1, OpenAI had released ChatGPT API.
- This is about 10 times less costly than similar conventional APIs.
- Many applications have been created to date.
-
As of this writing on 6/4, I am using ChatGPT Plus in Browse mode.
- Then there is the waiting line at Anthropic Claude.
- These words are not explained.
- This is because the situation will change by 8/15 when this article is published.
-
Knowledge of how to use a particular tool quickly obsolescence.
- It is not suitable for writing in an academic journal.
- It would be better to write Knowledge at a high level of abstraction to make it less stale.
- However, to understand abstract knowledge, you must have your own Specific Experience to support it. - one must see with one’s own eyes 。
- Related: Rootless Knowledge.
-
This article will be published on 8/15.
- The author does not know what the reader can experience at that point in the situation.
- ChatGPT Plus, for example, has stopped accepting new applications many times.
- We don’t know if what is available now will still be available tomorrow.
- Related:. - There is no guarantee that the door that is open now will remain open in the future. - strike while the iron is hot
-
Because of this situation, I felt it was a “chestnuts in the fire” situation.
- And “I don’t know about the future” is not acceptable.
- With the chestnuts they pick up, they must create “unique and delicious chestnut dishes” that are different from those of other chefs.
- Interesting challenge.
1.2 Journey in search of the next generation of intellectual productivity
.
-
First, the cooks would need to be introduced.
- He is the author of The Intellectual Production of Engineers, published in 2018, and is interested in Increased intellectual productivity.
- Featured in the December 2019 issue of Information Processing.
-
Human intellectual productivity is enhanced not only by software but also by methodologies and names.
- When a concept has a name that refers to it, it becomes easier for people to deal with it in their thinking.
- Additions:.
- The Four Elements of Human Enhancement.
- “Augmenting Human Intellect: A Conceptual Framework” (1962)
-
- On productivity through words: see Concept Handles.
- The concept is made easier to manipulate by adding the word handle to the concept.
- The Intellectual Production of Engineers p.36 (Column) To name the pattern.
- > Name the pattern.
- > Engelbart saw language as a means for individuals to break down their perceptions of the world into “concepts” for modeling the world. Language is used to map symbols to these “concepts” and to consciously manipulate them. This conscious manipulation of concepts, he believes, is what “thinking” is all about.
- The Intellectual Production of Engineers p.137 (4.5.3.3) Get the tools of your thinking.
-
In his Chapter 1 column, “Naming Patterns,” Douglas Carl Engelbart cites language as one of the ways “to augment human intelligence. By creating abstract models of external events and naming these models, we can manipulate them in our minds and think about them. This is the enhancement of intelligence through language.
-
- The Intellectual Production of Engineers p.137 (4.5.3.3) Get the tools of your thinking.
- The concept is made easier to manipulate by adding the word handle to the concept.
- The Intellectual Production of Engineers p.36 (Column) To name the pattern.
- > Name the pattern.
- > Engelbart saw language as a means for individuals to break down their perceptions of the world into “concepts” for modeling the world. Language is used to map symbols to these “concepts” and to consciously manipulate them. This conscious manipulation of concepts, he believes, is what “thinking” is all about.
- On productivity through words: see Concept Handles.
-
Jiro Kawakita’s “way of thinking” gave birth to many methodologies and names, including the KJ method.
- However, since the book is from 1967, it does not incorporate the developments in digital technology over the next 50 years.
- I created “The Intellectual Production of Engineers” because I thought we needed a book for the next 50 years.
- We did not expect ChatGPT to appear five years later.
-
This technology will have a major impact on the way intellectual production is done.
- Some disappear, some remain, and new ones are born.
- What will be new? (Fig1)
- Fig1: Migration = part disappears, part remains, new birth
1.3 Intellectual Production and its Evolution through Interaction with AI
.
- The reason I was approached was an experimental log I wrote on Scrapbox, a kind of wiki, on Feb. 23.
- The experiment is to have ChatGPT play the role of a “listener who repeatedly asks questions” and organize his or her thoughts by answering those questions.
- Additions:.
- Let’s see, I can’t remember which one, or if it was mentioned in March for the February experiment.
- Nodal point of thought 2023-02-23 Yeah, but I don’t get the context from this one.
- Summary: An experiment to have ChatGPT play the role of a “listener who repeats a question”.
- Let’s see, I can’t remember which one, or if it was mentioned in March for the February experiment.
- Additions:.
- Letting the machine ask questions is familiar to the author.
- In the counseling technique described in “The Engineer’s Art of Intellectual Production,” the counselor uses only the language used by the client and a few question patterns. - clean language.
- Human counselors are expensive, so we created a CUI version in 2014 with the idea of having a computer do the same thing.
- It is being offered as a web service “Keichobot, a chatbot that brings out your ideas” starting in 2021 so that others can use it.
- By making it a web service, it can be accessed from a smartphone, allowing people to talk and deepen their thinking while bathing or taking a walk.
- What we did on 2/23 was an experiment to let ChatGPT play the same role.
- The experiment is to have ChatGPT play the role of a “listener who repeatedly asks questions” and organize his or her thoughts by answering those questions.
- The ChatGPT API was released the following week on 3/1.
- We have experimented with having them summarize the chat logs and suggest next actions, and have found it useful.
- As of 6/4, there has been no full-scale integration due to slow response times and other issues, but time will tell on performance issues.
- We feel that in the future, chat will be the baseline, and that it will be necessary to explore what can be added to chat to increase intellectual productivity.
- PS: As of 8/15, Agents living in Scrapbox. was created and Scrapbox, where AI lives. has 3
- This is not a “receive input from a human and quickly return output” implementation of chat, but a form of reading and writing Scrapbox
- The other thing I do is Think about AI Question Box UI.
- PS: As of 8/15, Agents living in Scrapbox. was created and Scrapbox, where AI lives. has 3
1.4 A new form of brainstorming using Scrapbox and ChatGPT
.
- Following the release of the ChatGPT API on 3/1, we created a tool to connect Scrapbox and ChatGPT.
- The author has 60 books worth of text data in his Scrapbox, which he vector searches and gives to ChatGPT as reference material.
- Postscript:.
- 2023-03-08 I connected my Scrapbox to ChatGPT.
- 2023-03-09 Scrapbox ChatGPT Connector
- 2023-03-24 Story study session about connecting my Scrapbox to ChatGPT.
- When I published this tool, a number of people published their own Scrapbox-generated data.
- With different data, they will react differently to the same agenda.
- In other words, they are virtual personalities with differing opinions.
- Brainstorming sessions can be held 24 hours a day, 7 days a week, whenever it is convenient for you, bringing together virtual personalities with diverse perspectives.
- Examples:.
- 2023-03-11: AI Panel Discussion
-
(Nishio) I am Nishio, the moderator. Human. Today, with the cooperation of /tkgshn, we have prepared ChatGPT with this Scrapbox/nishio, tkgshn’s Scrapbox and Qualia-san’s/qualia-san Scrapbox loaded respectively, and we will consider them as individual personalities and have a panel discussion! The AI output will be unedited and the moderator will do his/her best.
-
- 2023-03-15: [/villagepump/Scrapbox ChatGPT Connector roundtable mode](https://scrapbox.io/villagepump/Scrapbox ChatGPT Connector roundtable mode).
- 2023-03-11: AI Panel Discussion
- Compared to blogs, Scrapbox seems to work better with virtual personification because it is a mechanism that helps short articles to be refined over time.
- Related: Scrapbox is a purifier of ideas.
- It would also be interesting to turn a classic classic into a virtual persona.
- In the past, in order to converse with the great men of the past, humans had to read books and create emulators in themselves.
- The cost of creating that emulator has dropped dramatically.
- See also: The Intellectual Production of Engineers Column: No Two-Way Communication with Books.
-
By chewing up the contents of the book once and turning it into a diptych, and then reconstructing it myself, I nurtured my own understanding and created a model of Jiro Kawakita’s way of thinking. I called this “creating a Jiro Kawakita emulator within myself.
-
1.5 Varieties of subjectivity: correctness and usefulness are different things
.
- In this use of “brainstorming with virtual personalities,” individual statements need not be objectively true. - [Correctness and usefulness are two different things. By looking at the opinions from yourself and from Different Perspectives, you notice the blind spot. This awareness is useful.
- Jiro Kawakita said, “All data is a lie.
- Everything has the potential to be wrong, which means that it is important to gather different ways of seeing and judging things from different perspectives.
- Put another way, it is an approach that moves from “one subjectivity” to “many subjectivities,” rather than eliminating subjectivity and proceeding to objectivity.
- On the other hand, there are those in the world who expect ChatGPT to return an objectively correct answer and are disappointed.
- Such an AI would be useful, but at this point in time homo sapiens civilization has not yet produced it.
- Lamenting it is futile.
- Because it is not something that can be controlled by lamenting.
- The beneficial use of the tools available today is an area of skill and can be controlled by the user.
2. Social impact and system legitimacy
.
2.1 Direction of Social Change and Plurality
.
-
They also want me to write about the possibility of social change in this project.
- There is no argument that social change will occur. - As costs fall, structures change. 。
- The question is “the direction of social change.
- We need to consider which direction is best for change and steer the ship so that it does not go in the wrong direction.
-
The key word on this issue would be Plurality.
- It is a concept proposed by Audrey Tang, Minister of the Digital Agency of Taiwan, and others.
- There is concern that productivity gains from AI will make government too powerful.
- Therefore, AI must be used to strengthen the people’s power to achieve balance.
- This direction is called “Plurality.
- Plural (plural)” is used as a synonym for “Singular (singular)” in “singularity”.
- It is a representation of a single central government versus a diverse population.
2.2 Broad Listening: Enhancement of the ability to “hear” a large number of opinions
.
-
I mentioned earlier that naming concepts strengthens intellectual productivity.
- I think the term “broad listening” used by Audrey Tang in the Plurality discussion is instructive.
- It’s about “listening” to a large group of people extracting the essence from the distribution of their peers’ opinions.
- She believes that the technology to support this will enhance democratic careful deliberation on a large scale.
- Related:.
- Audrey Tang spoke at LibreCon 2016 Four Steps for Deliberative Discussion explicitly collecting people’s “feelings” - Polis used to support careful deliberation is described as a [sentiment gathering platform
- Related:.
-
From another perspective, this is about overcoming human Limitations of Cognitive Ability.
- In other words, the augmentation of human intellectual and productive capacity through technology.Human Augmentation
- The development of information and communication technology has made it possible for one person to communicate his or her views to a large number of people.
- For example, it is easy for the president of a company with 1,000 employees to reach out to all employees through video distribution.
- However, if 1,000 employees send a 10-minute video to the president, it will take the president 7 days x 24 hours to watch it.
- So listening to the opinions of the masses is the bottleneck.
- See also: A more detailed discussion of the original source of this figure can be found at Plurality and Polis Study Group.
- This diagram seems to be useful for explaining broad listening, so I’ll call it CC0.
- Example of use: Updating Society and Democracy through Civic Tech.
2.3 Eliminating bottlenecks through digitization will change the social structure
.
-
The organization needed a layered structure because a thousand flesh-and-blood people couldn’t listen.
- So we had to put middle management in between to summarize the information manually.
-
Indirect democracy became necessary because we could not listen to everyone.
- So people needed to have delegates in between to represent their views and argue on their behalf.
-
Audrey Tang compares elections to the poor Internet, which can upload only a few bits every four years. - digital native is not satisfied with that situation.
- The development of technologies that summarize the “subjectivity of the many” and enable broad listening will change the structure of society.
- Related:. - Elections are slow communications, sending 5 bits every four years. - Digital natives don’t think that once-every-four-years upload bandwidth is enough.
-
If costs fall, the structure will change.
- A system that was optimal under the existing cost structure may not be optimal under the new cost structure.
- Let’s look again at Fig. 1 “Part disappears, part remains, and a new one is born.
-
- Reprint: Fig1: Migration = part disappears, part remains, and a new one is born
-
- What is not being realized now because the cost is not worth it will become newly feasible as a result of lower costs.
2.3 To future students exploring unknown knowledge
.
-
Some may be worried about the future.
- Conduct experiments and weave the results into words.
- This is about tapping into “knowledge that has not yet been verbalized.
- This contribution to civilization is unlikely to lose value as the performance of AI that has learned “already verbalized knowledge” improves.
- the interface between the world and AI and the job of sensing the world and giving it to AI will not disappear in the foreseeable future.
- Related: Work to open up the forest of what has yet to be written.
- Conduct experiments and weave the results into words.
-
However, it will be necessary to be aware of whether the readers of one’s information dissemination are flesh-and-blood humans or AI.
- Information dissemination in a form that AI cannot read will be of decreasing value.
- As a book author, I am conflicted as to whether I should continue to invest time and effort in book writing.
- Books would be the “disappearing thing” in Fig 1.
- I accepted this article on the condition that it can be published in my Scrapbox.
2.4 Individual behavior legitimizes the system: objective or subjective? Subjective?”
-
The system is folk,
- People who thought the new system was better would use the new system,
- As people come together, the legitimacy of the system increases.
- PS: Talk about “what people choose to do” working as [voting - mobile voting 、 subjectivism 、Panarchy
-
For AI-related papers, the first thing to be done is to register the pre-review papers (preprints) with arXiv. - [Speed of information sharing” was selected over “quality assurance” by “peer review.
- This is not a good move from the point of view of existing non-open access journals.
- But people’s actions function as a kind of vote, legitimizing the system.
-
If OpenAI created a preprint server that ChatGPT preferentially learns from, would people post there?
- Would a company submit a manual for its product if it could submit something other than an article?
- Would people post blog posts?
-
When the usefulness of the submitted text can be speedily judged by the subjectivity of a large number of people,
- Will people also choose “speed over quality”?
-
For a subject to which science has yet to provide an objective answer,
- When AI can output the essence of the “subjectivity of many”,
- Would people choose subjectivity without waiting for objectivity?
-
In retrospect, we can say that this project was designed to gather the subjective views of each author regarding ChatGPT.
- Maybe the answer has already been given.
Yasukazu Nishio (Regular Member)
- Senior Researcher, Cybozu Labs, Inc. Director of MUTOH Incorporated Association. Doctor of Science, Master of Technology Management (Professional)
- Humans recognize in Vehicles of Civilization that the human physical body is only a cache, and that the information they emit to the open access field is their main body.
- Related:. - Humans are temporary caches
Subsequent Events
- 5/25: OpenAI launched a grant called [Democratic Inputs to AI
- It aims to improve the democratic process
- 6/20: OpenAI’s competitor Anthropic issued a preprint on applying LLMs to “Scalable Deliberation” using Polis.
- 7/6: release scope was being expanded from 2022-12 Twitter’s Community Notes is now being used in Japan [src https://twitter.com/CommunityNotes/status/1676767470961532928?ref_src=twsrc%5Etfw%7 Ctwcamp%5Etweetembed%7Ctwterm%5E1676767470961532928%7Ctwgr%5Eacbf59b89e01a5ac45d1921e1ec5d2e27f242b7f%7Ctwcon%5Es1_&ref_url=https%! 3A%2F%2Fwww.arabnews.jp%2Farticle%2Ffeatures%2Farticle_94560%2F]
- Community Notes collects subjective usefulness for individuals with the question “Is this note helpful?
- 6/4~present Omoikane Project.
- As of 8/15, seeking to create a better discussion mechanism by way of SF prototyping (letting people create near-future stories to trigger discussions) with the support of AI.
- One of the above “[Scrapbox, where AI lives.
- Anyone can join: [/omoikane/join omoikane](https://scrapbox.io/omoikane/join omoikane).
- From One Subjective to Many, Not Subjective or Objective: Q&A
- I’ll answer later.
This page is auto-translated from /nishio/主観か客観かではなく、一人の主観から大勢の主観へ 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.