AI and Meno's Paradox: How Can We Ask and Know What We Don't Know?

AI and Meno's Paradox: How Can We Ask and Know What We Don't Know?

Deep Research

1. Introduction

"Meno's Paradox" from Plato's dialogue Meno presents the dilemma: "If you don't know something, you don't know what to search for, and if you already know it, there's no need to search." In the modern era, this paradox is being re-examined in relation to artificial intelligence (AI).

This article explores the possibilities of knowledge acquisition and exploration in the AI age, using Meno's paradox and the concept of tacit knowledge as starting points.

2. Comparative Perspectives on Meno's Paradox

2-1. Key Points of the Paradox

In "Meno's Paradox":

  1. For things unknown, we don't even know how or what to search for.
  2. For things already known, there's no need to search.

This results in the contradiction that "learning is impossible." Plato attempted to resolve this with the Theory of Recollection, positing that human souls "already know something before birth." This implies that some "clue" lies hidden between "complete ignorance" and "complete knowledge."

2-2. Connection to Tacit Knowledge

Meanwhile, Michael Polanyi's concept of "tacit knowledge" addresses the question of how we can learn "knowledge not yet verbalized" by assuming the existence of unverbalized knowledge and skills. In the process of "getting the knack," insights and discoveries emerge that cannot be derived from explicit knowledge (formal knowledge) alone.

These discussions share a common thread: how to explore the unknown and break through the limits of verbalization.

3. Can AI Users Only Ask About What They Already Know?

3-1. Prompts to AI Depend on "Known Language Expressions"

Current large language models (like GPT) operate by users inputting text (prompts) and receiving responses. However:

  • It's difficult to accurately query concepts the user doesn't know or unestablished concepts
  • The AI model itself is built from training data (past linguistic resources)

This raises the concern: "Without entering keywords we already know, can AI really provide answers?" Yet this is an unavoidable problem inherent to human language use. Similar to Meno's paradox where "we cannot ask about what we don't know," even in the AI era, how to take that first step of inquiry remains a challenge.

3-2. AI's "Generative Capability" and "Exploration Expansion"

However, large language models possess these characteristics:

  1. Generative ability: The model combines learned knowledge to present insights and answers unexpected by users
  2. Broad associative exploration: Suggests related concepts or combinations humans might not consider, providing clues for new questions
  3. Iterative dialogue: Users can "counter-question" AI responses, interactively refining and redefining concepts

For example, when asking AI to "list unsolved problems around field A," it can present "related issues being discussed but not yet established" or "ideas from adjacent fields." This generates new perspectives the user hadn't anticipated, spawning new questions and research themes.

In other words, the "unexplored hypotheses and perspectives" AI presents can serve as "clues to unknown territories" in the Meno's paradox sense. Users aren't in complete ignorance; by simply inputting a "rough keyword or field name," they can have AI "evoke" broader ideas and related areas.

4. Can We Go Beyond "What Humanity Has Ever Conceived"?

4-1. AI's Limitations Based on Data and "Novelty"

AI (especially large language models) is built by learning massive text data, fundamentally dealing with knowledge derived from "existing human intellectual products." Thus, it's difficult for AI to spontaneously generate insights about completely unexplored territory with no trace in training data.

However, by recombining and editing elements scattered across training data in new ways, AI can present perspectives humans hadn't consciously addressed. Such "novelty through combination" can open doors to themes humanity (at least mainstream) hasn't considered. This aligns with the view that new inventions and discoveries emerge from "combinations of existing elements."

4-2. "Exploring New Knowledge" Through Dialogue and Practice with Humans

As Polanyi's "tacit knowledge" and Zen's "beyond words" suggest, knowledge and insights aren't always confined to existing verbal expressions. Important roles are played by:

  • Physical practice
  • Dialogical trial and error
  • Intuition and inspiration
  • Collaborative work (master-apprentice, joint research)

If AI complements these elements, it might be possible to gradually externalize and clarify vague questions or images through interaction with AI, generating new concepts.

Through this iterative process, "questions humanity has never conceived" may gradually take shape, reaching ideas beyond mere "reuse of the known."

5. Conclusion: "Questions About the Unknown" and the Future of Learning in the AI Era

  1. Meno's Paradox

    • The dilemma "you can't ask about what you don't know" remains a fundamental problem in the AI era.
    • But like Plato's "Theory of Recollection," we can assume we possess some latent clue to knowledge, not a complete blank slate.

  2. Tacit Knowledge and AI

    • Polanyi's "tacit knowledge" raises questions about acquiring and transmitting non-verbalized knowledge and skills.
    • AI can assist in gradually externalizing and developing fragments of human sensations and ideas not fully verbalized through dialogue.

  3. Possibility of Going "Beyond the Known"

    • Large language models depend on "trained data" but can newly combine and present existing knowledge, offering unexpected threads of insight.
    • Through iterative questioning and answering with users, previously unconscious unknown questions may emerge.

  4. Key Approaches

    • Present keywords or directions to AI (however vague) to handle unknown territories (first clue).
    • Examine returned ideas and information while adding further questions, redefining concepts.
    • Through this process, the cycle of "ask → gain insight → redefine → ask again" repeats, expanding exploration into unknown territories.

6. Closing

Meno's paradox—"you can't ask about what you don't know"—confronts us even in the AI era. However:

  • Starting from small known clues,
  • Utilizing AI's associative and generative capabilities,
  • Through iterative dialogue and practice It's entirely possible to venture into unknown territories.

The ideas and related information AI presents may help unearth "knowledge we haven't consciously extracted yet" and form unknown questions—serving as aids for "recollection" and "discovery." Viewing such processes as "devices that complement tacit knowledge," there's significant potential to expand humanity's intellectual frontier.