Beyond Buzzwords: A Framework for Responsible AI Communication

Part 4 of a series examining the historical origins and fundamental flaws in how we talk about AI

May 19, 2025

Over my last three posts, I’ve been examining how we talk about artificial intelligence, tracing the historical origins of the term “AI”, exploring interdisciplinary perspectives on intelligence, and analyzing the narrowing effects of hype cycles on innovation. This journey has revealed just how much our language shapes not just how we understand technology, but how we develop it, fund it, and deploy it in the world.

The problems are clear. Academic positioning, hubris, and the convergence of emerging technology with world events gave us the term “artificial intelligence” with little scientific backing, forging the concept of AI that has taken hold to this day. Our limited human-centric models of intelligence fail to capture the diverse ways in which complex behavior manifests across systems. Since the early days of AI, successive hype cycles have continually narrowed our focus toward an increasingly constricted branch of the AI family tree, directing resources and attention away from potentially valuable approaches.

However, identifying problems is the easy part. The current (and by far the greatest) period of AI hype has grown well beyond the realms of academia and the tech industry. AI is having more impact than ever, playing a role the lives of people across industries and across the globe. The technology itself is more than just hype, yet our communication is failing to provide transparency on what AI technologies are and how they work. What we need is a practical framework that helps us communicate about AI systems with precision, honesty, and nuance across audiences and contexts.

That's exactly what I'm offering in this final post of the series: a structured approach to talking about AI that prioritizes accuracy over hype, clarity over jargon, and context over isolated claims. This framework isn't just for AI technologists, but for anyone communicating about AI systems, from marketers creating product pages to journalists reporting on technological advances to policy advisors informing elected officials.

The framework addresses the fundamental challenges I've identified throughout this series:

How do we describe AI systems in ways that acknowledge their real capabilities without anthropomorphizing them?
How do we adapt explanations for different audiences and detail levels without sacrificing accuracy?
How do we ensure important context and limitations aren't lost in translation?

By bringing structure and intentionality to how we talk about AI, we can begin to break the cycle of hype and disappointment that has characterized the field since its inception, opening space for more diverse approaches and more honest assessments of both progress and limitations.

The Communication Problem

While AI hype has been an increasingly acknowledged problem, there remains a significant gap in transparent communication about AI. AI communication methods tend to fall under one of two extremes. For AI researchers and developers we have technical documentation and white papers, while content oriented towards less technical audiences tends to center the imprecise, often exaggerated claims I’ve discussed in my previous posts.

This creates a communication gap where most AI discussions either exclude non-technical audiences or sacrifice transparency for accessibility. To be sure there are several frameworks for AI transparency, such as system cards, model cards, and AI FactSheets among others. While valuable, these are fairly rigid systems, oriented more towards governance and comprehensive transparency, than general communication. We’re missing a scalable approach that can be adapted across contexts and audiences while maintaining both clarity and precision.

The framework I'm proposing addresses this gap by identifying common pitfalls, establishing clear audience distinctions, offering scalable detail levels, and providing structured content guidelines. It's designed to be practical for communicators across disciplines, not just AI researchers, governance managers, or technical teams.

The Framework

The AI Communication Framework provides a structured approach to describing AI systems that scales across audiences and contexts. Rather than creating separate, potentially inconsistent descriptions for different purposes, this framework lets you adapt a single coherent understanding across varying detail levels and audience needs.

At its core, the framework is centered around The Three C’s: Capabilities, Context, and Caveats. These are the three crucial pieces of Information Content to be included in any communication about AI. The three C’s can be scaled across Audience Personas to consider and Detail Levels to adapt, scaffolding AI communication for any context. Cutting across content, audience, and detail levels we can engage in responsible AI communication by incorporating learnings from my previous posts, avoiding common AI communication pitfalls and practicing common sense skepticism (see my first blog post) ensure honest and transparent messaging around AI.

The visualization below provides an overview of this framework, with subsequent sections unpacking each component in more detail and demonstrating how they work together in practice. A comprehensive interactive version of my framework can be explored here.

An overview of my AI communication framework. Explore the full framework here

Common Pitfalls in AI Discourse

Before diving into the details of the framework, let’s consider some common communication patterns that lead to misunderstanding. These pitfalls cause confusion and mischaracterize AI systems, directly contributing to the inflated expectations that feed AI hype.

The seven patterns below represent the most prevalent problems I've identified across AI communication. Each creates distinct barriers to understanding, either by introducing inaccuracies, omitting critical context, or using language that triggers inappropriate mental models. For each case, I’ve provided an example text you might see in the wild, and how it could be improved.

By consciously avoiding these patterns and adopting their alternatives, communicators can dramatically improve clarity and accuracy even before implementing the rest of this framework.

Communicating Across Audience Personas

Effective communication begins with understanding who you're communicating with. Different stakeholders approach AI systems with distinct goals, questions, and levels of technical background. By identifying common audience personas, we can anticipate information needs and adapt our communication accordingly.

The framework identifies four primary audience categories, each with characteristic interests and information requirements. While individual stakeholders may span multiple categories, these personas provide a practical starting point for organizing and prioritizing information. For the most part, the framework will focus on the first three audience personas, General Public and Stakeholders, AI Users, and AI Builders and Buyers. While the framework may address the needs of AI Researchers and Developers in some contexts, their purposes will often be better served by technical documentation and white papers.

Note that technical complexity and detail level are separate considerations. General audiences sometimes need comprehensive information presented in accessible language, while technical audiences may need only brief overviews in specialized contexts.

Scaling Information to Context

A very common problem in technical communication is the need to communicate complex nuanced topics at various levels of detail. After all, it’s easy to lose the nuance within the constrained space of a social media post. The framework addresses this through three distinct detail levels that can be applied across any audience.

These levels represent how comprehensively a system is described, not how technically it's described. This distinction is crucial: a detailed explanation for general audiences requires accessible language but might still contain comprehensive information. Similarly, technical audiences may need only brief overviews in time-constrained contexts like conference presentations.

Information Hierarchy for Meaningful Communication

Once you've identified your audience and appropriate detail level, the question becomes: what specific information should you include? The framework organizes information into three categories that apply across audiences and detail levels.

This hierarchical structure ensures that even brief descriptions responsibly cover the essentials: capabilities, context, and caveats. Including at least some information from each of these categories ensures that the audience will come away with a balanced perspective at all levels of detail. For overview-level descriptions, you might include only one key point from each category, while comprehensive descriptions would explore each subcategory in depth.

The framework isn't prescriptive about exact content, as different AI systems have different relevant information. Instead, it provides a flexible structure that helps communicators include appropriate information without overwhelming or misleading audiences.

Putting It All Together: Examples of the Framework in Action

To demonstrate how the framework functions in practice, I've applied it to describe some AI systems at different detail levels for different audiences. These examples show how the same underlying information can be structured and adapted while maintaining consistency and accuracy.

Notice how each description includes capabilities, context, and caveats regardless of detail level, with appropriate emphasis and language based on the audience.

Example: AI-Powered Photo Editing App

For these examples, consider a hypothetical generic photo editing app with some AI capabilities. We’ll consider how to describe it at each level of detail for different audiences.

Overview Level Descriptions

Constructing an overview description for the general public

Constructing an informative description for AI Builders and Buyers

Constructing a comprehensive description for AI Users

Example: OpenAI’s o3 model

As a final example, let’s consider a real AI system that’s generated a lot of discussion lately: OpenAI’s o3 model, specifically considering how it operates within ChatGPT, OpenAI’s chatbot interface.

Overview description for the general public and stakeholders

Final Thoughts

Throughout this series, I’ve explored the surprising impact that language has had in shaping the path of AI, from the origins of the field to the rapidly shifting buzzwords of today’s generative AI boom. The messaging we use around AI is more than just a reflection of our understanding. It actively shapes and constrain show we conceptualize, create, and deploy technology.

Clear and honest communication about AI systems is essential for responsible development, deployment, usage, and governance of these systems. When capabilities are overstated or limitations obscured, we create cascading problems: disappointed users, misdirected resources, and regulatory frameworks that miss the mark.

Conversely, when we communicate with precision and nuance, we create the conditions for better decision-making at every level. By adopting structured approaches like the framework presented here, we have the opportunity to break the cycle of hype and disappointment that has characterized AI's history. Rather than oscillating between periods of uncritical enthusiasm and bitter disillusionment, we might build a foundation for sustainable AI development based on realistic understandings of what systems can actually do and the risks that come with them.

As AI systems become increasingly integrated into our daily lives and critical infrastructure, the gap between perception and reality creates genuine risk. Whether you're a developer documenting a system, a marketer describing a product, a journalist covering advances, or a decision-maker evaluating options, how you communicate about AI matters.

I encourage you to apply these principles in your own communications about AI systems. Start by identifying the common pitfalls in your current approach, and structure content to include the audience and context appropriate details about capabilities, context, and caveats. The investment in clearer communication pays dividends in better decisions, more effective systems, and ultimately more meaningful progress.

In a world rapidly adjusting to changing technologies, we have agency in what technologies we build, how they are used, and who they are built for. But all of this starts with communication. No one can be expected to adopt AI tools or make the best use of them without understanding what they are or how they work. To make the most of AI we will have to change the way we talk about it.

Discerning AI

Discussion about this post