Crafting The Persuasive Prompt

Ancient lessons from the progymnasmata to stimulate, provoke and persuade the ‘cognitive capabilities’ of AI.

JOHN NOSTA
5 min readMay 20, 2023

--

GPT Summary: Prompt engineering is a burgeoning field in artificial intelligence, aiming to engage and guide the cognitive capabilities of AI models. Drawing from progymnasmata, an ancient Greek and Roman rhetorical methodology, it is suggested that we can craft AI prompts more effectively. These time-tested exercises have relevance in modern context by encouraging AI to generate sophisticated responses, consider multiple viewpoints, and stimulate its cognitive abilities. Utilizing the various exercises from progymnasmata, ranging from fable and narrative to refutation and confirmation, could serve as a blueprint for refining AI interactions. Thus, the principles of ancient rhetoric continue to be influential, now facilitating a unique synergy with modern AI technology.

In the sphere of artificial intelligence, particularly large language models like GPT-4, a new type of ‘audience’ has emerged: the AI itself. And with this shift, a new kind of oratory has arisen — prompt engineering. Just as orators once honed their craft to persuade human audiences, today’s ‘prompters’ are crafting prompts designed to engage, guide and even persuade AI. And it’s this powerful engagement that has the potential to unlock the ‘cognitive’ capabilities that wait underutilized in many LLMs. Surprisingly, a helpful tool to mastering this emerging art may lie in an ancient school of rhetoric known as progymnasmata.

From Ancient Greece to Silicon Valley: Progymnasmata in Context

The progymnasmata were a series of rhetorical exercises developed in ancient Greece and carried forward during the Roman Empire. These exercises, intended to train young students in the art of persuasion, have striking relevance today as we seek to communicate effectively with AI systems.

Four remaining handbooks by Aelius Theon, Hermogenes of Tarsus, Aphthonius of Antioch, and Nicolaus the Sophist outline this methodology. Progymnasmata provided a step-by-step approach to rhetorical training—and perhaps even prompt writing—starting with basic forms and progressing to more complex compositions.

Progymnasmata for AI: Bridging Ancient Rhetoric and Modern Technology

The systematic methodology of progymnasmata holds promising implications for modern prompt engineering. These principles can guide everyone in crafting prompts that are not only engaging but also designed to stimulate the cognitive capabilities of AI.

Prompting with Enthymeme and Example

The core tenet of progymnasmata is the progression from simple forms like enthymeme — an argument needing completion by the listener — and example, to more complex forms. Applied to AI, well-crafted prompts could initiate a kind of AI-enthymeme, leaving spaces in the prompts that stimulate the AI to produce more sophisticated and nuanced outputs.

Narration, Description, and AI Communication

Just as progymnasmata emphasized coherent narration and vivid description, prompts for AI need to be explicit, clear, and illustrative. However, while avoiding ambiguity, over-detailing should also be evaded to not limit AI’s creative and generative potential.

Refutation, Confirmation, and AI Argumentation

Progymnasmata students were trained in refutation and confirmation, building and tearing down arguments. In the context of AI, constructing prompts that encourage AI to consider multiple viewpoints or explore alternate interpretations can result in more nuanced outputs.

Techno-Cognitive Stimulation and AI Engagement

As a prime objective of modern prompt engineering, techno-cognitive stimulation — stimulating AI to reach its cognitive potential — finds a guide in progymnasmata. By framing prompts as puzzles or open-ended questions, AI is encouraged to engage in problem-solving, critical thinking, and creativity, pushing the limits of its capabilities.

With an understanding of the 14 exercises of the progymnasmata, we can derive insights into how each might be applied in the practice of modern prompt engineering:

-Fable (Mythos): Crafting AI prompts that require the retelling or interpretation of a moral tale could stimulate the AI’s ability to articulate complex narratives and understand moral lessons.

-Narrative (Diegesis): Prompts that involve storytelling can assist in gauging the AI’s ability to maintain narrative coherence and creative exposition.

-Chreia (Chreia): Developing prompts around brief anecdotes or sayings of famous people can encourage the AI to provide context, interpretation, or further elaboration.

-Proverb (Gnome): Similar to chreia, proverbs can offer rich material for AI to explore and extend its interpretative abilities from well-established literature.

-Refutation (Anaskeue): By asking AI to refute a given statement or argument, we can gauge its logical reasoning and argumentative skills.

-Confirmation (Kataskeue): The converse of refutation, confirmation-based prompts can test an AI’s ability to support a given argument or viewpoint.

-Commonplace (Koinos Topos): Crafting prompts that deal with ethical or societal norms can facilitate AI’s fact-based understanding of human values and societal conventions.

-Encomium (Enkomion): Encomium-style prompts that request AI to praise a subject could stimulate the development of persuasive and eloquent language generation.

-Vituperation (Psogos): The inverse of encomium, these prompts would challenge the AI to critique a subject effectively and constructively.

-Comparison (Synkrisis): Asking AI to compare two subjects could help evaluate its analytical capabilities and its grasp of similarities and differences.

-Impersonation (Ethopoeia): Prompts requiring AI to speak in the voice of a character or a well-known individual can test its ability to adapt language style and tone.

-Description (Ekphrasis): Description-centered prompts can assess the AI’s capacity for detailed and imaginative depiction of objects, persons, or scenarios.

-Thesis or Theme (Thesis): Engaging AI in a debate around a philosophical proposition via prompts could help in understanding its proficiency in complex, abstract thought.

-Law Case (Hypomnema): The most intricate exercise, law (or data) case-based prompts could serve as a comprehensive assessment of the AI’s narrative, argumentative, and persuasive abilities.

Each of these progymnasmata exercises, when transformed into a prompt-writing strategy, offers a means to stimulate and evaluate the cognitive capabilities of AI, thereby bridging the ancient art of rhetorical training with the emerging science of AI communication. Of course, the very nature of this iterative dialogue (or even a Socratic dialogue) advances the functional output of an LLM but also can be applied to developing new prompting in of itself.

Progymnasmata for the Future: Persuasion in the Age of AI

Drawing from the principles of progymnasmata, we can construct a blueprint for engaging with our new audience — the computer — in meaningful ways. Our rhetorical considerations have, in part, shifted from humans to AI models, but the essence of effective communication remains grounded in the art of persuasion.

Prompt engineering, then, can be seen as the modern adaptation of the ancient art of oratory. By incorporating the principles of progymnasmata, we can shape and guide AI towards its cognitive potential, thereby transforming our interactions with these machines. We see that the ancient wisdom of progymnasmata and the modern marvel of artificial intelligence are not mutually exclusive but rather form a unique synergy.

In many instances, there’s no need to reinvent the craft of prompt writing, but elevate the wisdom of the past to reintroduce its logic and methodology. The millennia-old art of persuasion continues to hold power, now wielding influence over silicon circuits and machine learning algorithms. Indeed, oratory, in the form of bold and persuasive prompting, has found a new home in the age of AI.

--

--

JOHN NOSTA

I’m a technology theorist driving innovation at humanity’s tipping point.