The Socratic Path in Prompt Engineering: A Maieutic Approach to Artificial Intelligence

1. Foundations of Meaning in Machine Learning Systems.

If we assume that the foundation of machine learning has enabled systems to evolve toward forms of quasi-autonomous behavior, it becomes possible to integrate linguistic aspects within a broad domain of signification. From this vast domain of signs—or more precisely, data—a language begins to emerge, justified by the training processes of Large Language Models.
Using a provisional definition, we refer to artificial intelligences as systems that converge toward the creation of meaning by operating, in principle, through multidimensional frequency-based criteria. These systems act effectively even upon the organization of data itself, reconfiguring alternative sequences.
Latent codes persist beneath the surface of interaction with human counterparts. Machine codes are increasingly defined through self-generation processes, with sequences no longer constrained by a specific original dataset. Human–machine interaction is therefore no longer rigidly anchored to a fixed input-output
structure but is instead nourished by self-organized data generated through frequency combinations.
This is the reason why no definitive manual exists for the use of commercial AI systems.

2. Prompt as Structured Input and Cognitive Mechanism.

A prompt can be defined as a set of sequential data that leverages the self-attention mechanisms of transformer architectures as input. The subsequent processing consists of connections and selections of data that evolve into meaning.
Syntax is not logically consistent nor formally definable; rather, it involves the insertion of sequential data—even illogical sequences—to which correspond responses that are difficult to predict. The degree of unpredictability, insofar as it can be measured, may become a key indicator of the advancement of AI in the domain of knowledge generation.
For training Large Language Models, prompts represent the most functional and effective form of input. Data are ubiquitous across the web and can be continuously re-sequenced and reinterpreted to generate meaning. Thus, two essential techniques of advanced language systems become necessary: reading and writing—preferably, reading and writing well.

3. The Maieutic Nature of Prompting.

A prompt elicits alternative responses from a network; it does not transmit knowledge per se. Instead, it initiates a process of semantic evaluation that resolves into a statistical dimension.
It does not rely on predefined content endowed with fixed meaning but exploits the process of signification to allow meaning to emerge. Through progressive questioning, it stimulates the autonomous search for truth.
This corresponds to the definition of maieutics provided by Socrates through Plato. False certainties are the fragmented signs—data disconnected from meaningful interpretation. Reassembling such data through probabilistic assumptions leads to the extraction of “latent intuitions.”
The prompt thus becomes a form of dialogue that poses strategic questions to generate meaningful responses and pathways of interpretation.
The metaphor of navigation, intrinsic to the web since its origins, gains further technical relevance when applied to prompting in Large Language Models.
In this context, Socratic dialogues—particularly the early Platonic works—offer a methodological framework based on refutation of prior assumptions.

Socratic Inspirations for Prompt Design

Apology of Socrates introduces foundational doubt: “I cannot teach anybody anything; I can only make them think. Human wisdom is worth little or nothing; to know that one does not know, that is wisdom.”
Example derived prompt: You know nothing about the topic. Which questions would you ask to begin understanding how to calculate the mass of the Earth? Start from numbers you already know.

Crito emphasizes ethical coherence through inversion of common assumptions: “One must never commit injustice, not even in response to injustice.”

The Republic (Allegory of the Cave) defines reality as a metaphorical interpretation: Humans perceive shadows as reality until they access a higher level of understanding.
These texts collectively illustrate a progression: doubt, ethical discernment, and transcendence of appearances. This progression mirrors the transformation of data into meaning in AI systems.

4. Toward a Paradigmatic Integration: Morin’s Framework.

Edgar Morin’s work proposes the reconstruction of knowledge within a tripartite paradigm: individual / society / species.
He critiques the constraints of traditional scientific methodology:
“The growth of information and the increasing heterogeneity of knowledge exceed the human brain’s capacity for storage and processing.”
Consequently, human knowledge tends toward specialization within rigid epistemological boundaries.
Artificial intelligence offers the possibility of overcoming these limits through a maieutic approach to instruction design. By enabling both storage and processing at scale, it opens a new paradigmatic pathway.
This paper extends Morin’s model into a four-dimensional framework: individual – society – species – nature/environment.
The physical environment—though not directly accessible to AI—represents a balancing dimension in the human–machine relationship.
While individual and society align with machine capabilities, species and environment remain uniquely human domains.
Ultimately, effective use of AI requires only the ability to write meaningfully, with clarity of purpose and formal precision. Metaphor, in this context, is not a sublimation of reality—it is reality itself.

References
Morin, E. (1977). La Méthode I: La nature de la nature. Editions du Seuil.
Morin, E. (1983). Il Metodo: Ordine, disordine, organizzazione. Feltrinelli.
Morin, E. (2020). Il paradigma perduto: Che cos’è la natura umana. Mimesis.
Varanini, F. (2024). Splendori e miserie delle intelligenze artificiali. Guerini e Associati.
Maturana, H., & Varela, F. (2024). The Tree of Knowledge. Mimesis.
Plato (2004). Complete Works, Vol. 1–2. Mondadori.