from 日記2023-10-20

関係を見出す実験

これの「関係を見出す」をAIが支援できる可能性、あるな…

実験0、6つのキーワードをomni-privateに入れてみる

→検索対象データに既にキーワード間の関連を考察した文章が入っているために、それを拾って再現してしまう

そこで生のGPT4を使うことにした

実験1、キーワードXを与えて、それに関連するキーワードを6つ列挙させる

この時に、期待に反する出力が出る場合は、キーワード単体で「人間の意図している意味」を指定できていないので、補足をつける

人間の側が知識不足で解像度の低い表現をしてしまっているので解像度を高めるということ

例: LLM

LLMって何？

"LLM"は、"Master of Laws"の略で、法学の修士号を指します。...

その後で各Xの間の関係を記述させる

僕のまだ気づいてない関係が出てきたら成功

Given keyword X, answer 6 different keywords related to X.

X=LLM(Large Language Model)

Certainly! Here are six keywords related to "LLM (Large Language Model)":

Natural Language Processing (NLP)

OpenAI

GPT (Generative Pre-trained Transformer)

Machine Learning

Text Generation

Deep Learning

X=Plurality(related to Digital Democracy)

Certainly! Here are six keywords related to "Plurality (related to Digital Democracy)":

Direct Democracy

E-Governance

Online Voting

Digital Participation

Civic Technology

Electronic Referendum

X="World ID"(A more human internet with global proof of personhood. Privacy‑First. Self‑custodial. Decentralized.)

Certainly! Here are six keywords related to "World ID (A more human internet with global proof of personhood. Privacy‑First. Self‑custodial. Decentralized.)":

Digital Identity

Self-Sovereign Identity

Decentralized Autonomous Organizations (DAO)

Blockchain Technology

Privacy-Preserving Systems

Digital Sovereignty

X=大勢の主観

The term "大勢の主観" is a combination of Japanese words: "大勢" (taisei) meaning "general trend" or "majority" and "主観" (shukan) meaning "subjectivity" or "subjective view".

Considering that context, here are six keywords related to "大勢の主観" (Majority's Subjectivity):

公共の意見 (Public Opinion)

集団思考 (Groupthink)

民意 (Popular Will or Public Sentiment)

主流の価値観 (Mainstream Values)

マイノリティの意見 (Minority Opinion)

社会的圧力 (Social Pressure)

X=川喜田二郎(Jiro Kawakita, who made KJ-method)

Jiro Kawakita is known for the KJ method (also known as the Affinity Diagram or Kawakita Jiro method). The method is employed to synthesize large volumes of information by grouping and categorizing related facts.

Here are six keywords related to "川喜田二郎" (Jiro Kawakita) and the KJ method:

KJ法 (KJ Method)

類似性 (Affinity)

情報整理 (Information Organization)

ブレインストーミング (Brainstorming)

グループワーク (Group Work)

問題解決 (Problem Solving)

X=Eugene T. Gendlin

Eugene T. Gendlin was an American philosopher and psychotherapist who developed the process known as "Focusing." This process emphasized the body's inherent wisdom and the role of bodily sensations in our understanding and expression of emotions.

Here are six keywords related to "Eugene T. Gendlin":

Focusing

Experiential Psychotherapy

Felt Sense

Client-Centered Therapy

Carl Rogers (since Gendlin collaborated extensively with him)

Philosophy of the Implicit

Find relationship between Xs

Let's explore the relationship between the various "X" topics you've mentioned:

LLM (Large Language Model):

This pertains to advanced machine learning models that handle language tasks, like OpenAI's GPT series.

Plurality (related to Digital Democracy):

This topic concerns the majority's voice or opinion in a digital democratic system, and how technology can facilitate or hinder representation.

World ID:

This is about a decentralized, privacy-first digital identity. While it's not directly related to language models or digital democracy, it represents a tech-driven solution for ensuring privacy and self-sovereignty.

大勢の主観 (Majority's Subjectivity):

This has thematic similarities with "Plurality", as both discuss the views or opinions of the majority. In democratic systems, the majority's subjectivity can play a role in decision-making processes.

川喜田二郎 (Jiro Kawakita):

Known for the KJ method, Jiro Kawakita's work relates to information synthesis. While it doesn't directly tie into the other themes, its focus on categorizing and understanding information could be seen as a manual counterpart to what LLMs do algorithmically.

Eugene T. Gendlin:

Gendlin's work revolves around understanding and expressing emotions through a connection with our bodily sensations. This is more psychological and philosophical than the other topics, which are more technologically oriented.

Relationships:

Technological Progress: LLM, World ID, and to some extent Digital Democracy signify advancements in technology and how they affect various facets of our lives.

Understanding and Organizing Information: LLM, KJ-method, and Focusing (from Gendlin's work) are all methods/tools for understanding and organizing either data (information) or human emotions (inner experiences).

Majority and Representation: Plurality in Digital Democracy and 大勢の主観 both touch on the theme of the majority's views, opinions, or subjectivity.

Personal Understanding and Expression: Both Gendlin's work on Focusing and the KJ method revolve around introspection, understanding, and expression, whether it's emotional or information-based.

In summary, while not all of these topics are directly related, they intersect at various points, especially around the themes of technology's role in our lives, the importance of understanding and representation, and methods for introspection and organization.

感想

興味深い

僕自身が発展させるとき、暗黙に母語の日本語を使ってしまうが、今回は英語の得意なGPT4に英語で考えさせているので、意外なキーワードが出てくる

"information synthesis"とか

実験2

Kozanebaに入れてみる

過程での気づき

あー、「大勢」を「Taisei」と解釈してるな

Xと6つのキーワード

Relationships

説明も入れる

「大勢の主観」と「Focusing」の間に谷があるね

気になったキーワードを拡大しつつ、谷に橋を掛けられるか考える

introspection

inner experience

Public Sentiment - 主流の価値観 - Mainstream Value

この接続自体が面白いな

繋がった

sentimentはもちろんemotionだ

そして今のところemotionを持てるのはhumanだけということか

inner experienceが真ん中に来た

つまりまだ言語化されていないのはinner experienceをLLMやPluralityにどうつなげるかということ

Find relationship from "inner experiences" to each of "LLM" and "Plurality"

Let's delve into the relationship between "inner experiences" and both "LLM" and "Plurality":

Inner Experiences to LLM (Large Language Model):

Connection: "Inner experiences" refer to a person's subjective feelings, thoughts, emotions, and intuitions. LLMs, like OpenAI's GPT series, are designed to understand, generate, and respond to human language, which often conveys these inner experiences.

Understanding and Mirroring: While LLMs don't have emotions or consciousness, they can be trained on vast amounts of text that describe human emotions and sentiments. As a result, LLMs can generate language that mirrors or resembles human inner experiences.

Therapeutic Uses: LLMs can be utilized in therapeutic settings as tools to help individuals express and understand their inner experiences, even if the models don't truly "understand" emotions in a human way.

Bias and Representation: An LLM's training data could be biased or may not fully encompass the vast array of human inner experiences. This limitation makes it essential for users to be discerning when interpreting LLM responses related to deep emotional or personal topics.

Inner Experiences to Plurality (related to Digital Democracy):

Connection: "Inner experiences" in the context of plurality refer to the individual subjective feelings and opinions of members within a larger group or majority. In a digital democracy, these experiences can influence or be influenced by the collective voice.

Individual vs. Collective Voice: While "plurality" usually focuses on the voice of the majority, it's made up of individual inner experiences and opinions. The challenge is ensuring individual voices don't get lost in the collective, especially when they might offer important insights or dissenting views.

Digital Platforms as Expression Tools: Digital democracy tools can be platforms for individuals to express their inner experiences and opinions, contributing to the broader dialogue.

Feedback and Change: When a digital democracy system effectively captures the inner experiences and sentiments of its participants, it can lead to more responsive governance and policies that better reflect the needs and feelings of the population.

In conclusion, both LLMs and concepts of Plurality in Digital Democracy interact with the idea of "inner experiences" in distinct ways. LLMs attempt to understand and reproduce language related to human emotions and sentiments, whereas Plurality in Digital Democracy grapples with the challenge of representing and valuing individual sentiments within the collective voice.

更新

intuitionsいいね

Therapeutic UsesでClient-Centered Therapyが引っ張り出された

insightやdissenting viewをもたらす

全体

つまり人々のまだ言語化されていないInner ExperiencesをLLMのTherapeutic Usesによってinsightやdissenting viewをもたらす、それはよりresponsive governanceに近づくという形でPluralityに寄与するわけだ

それはKeichobotの延長線上にあると思う

これはFocusingの亜種のようなClean Languageというセラピーをベースにしているが、Focusing自体をもっと深く調べた方がいいかもしれない

振り返り

LLM(GPT4)は「関連を見出し」「思考の材料を作る」ことはできている

人間がそれをキャッチできていない

Kozanebaは僕の長年のKJ法の体験を踏まえて開発したツール

このツールの効果は下記の一行にとてもよく表れてると思う

つながりが可視化されることによって「つながりの不足しているところ」がわかる

「まだつながりが見出されていないところ」につながりを見出すための思考を集中することができる

気になってキーワードを拡大することによって「思考の道具」化が起きている

当初まったく言語化されてなかった「inner experience」が最重要キーワードへと育っていった

individualが出てきたのもよかった

experienceは根本的に個人的なものなので、それを引き出して流通可能なものに変換してやる必要がある

考察

現時点のGPT4で既に大部分の人間より「関係を見出すこと」が得意

この「得意」とは「案をたくさん出すことができる」ということ

ブレインストーミングの量での評価と同じ

人間と違って「こんなこと言っていいのかな」という心理的抵抗がないから案だしが本当に得意

人間の側の「たくさん出された案」の理解能力が不足している

特に1次元のテキスト形式では認知能力の限界にすぐぶつかる

Kozanebaによる2次元配置と関係線と重要断片の拡大などが認知能力を高めることに寄与している

テキスト形式のままでは「書かれてないこと」に気づくのが困難、関係線を引けば「線のないところ」が見えてくる

「たくさん出された案」に重みづけをすることは人間の主観的な解釈

「採用する/破棄する」の二択では人間は捨てることに恐怖を感じて残す選択をしがちで、それによって自分のキャパシティをこえた要素を抱え込んでしまう

「重要なものを拡大する」という戦略により、明示的な廃棄をしないまま、良いものに対して注意力をより多く割り当てることができている