When Designing Useful AI Interfaces, The Metaphor Matters
Few tools have made such a profound impact as quickly as ChatGPT. Are we using the right metaphors to interact with it?
On November 30th, ChatGPT turned one. Few tools have made such a profound impact so quickly. As many have noted, the public response “shocked” ChatGPT’s creators at OpenAI. Initially considered a ‘low-key research preview’ the night before it was released, OpenAI employees were betting on how many users would access it within the first week. According to The Atlantic’s Charlie Warzel, the highest guess was 100,000 users. In reality, the chatbot reached 1 million within the first five days, setting the record for the fastest-growing user base in history. Warzel writes, “The product was intended to be the software equivalent of a concept car. Instead, it became one of the most popular applications in the history of the internet.”
In the automotive industry, concept cars are designed to showcase new styles, technologies, and features. Not intended for immediate mass production, they are used to test out ideas, gauge public interest, and demonstrate a company's capabilities and vision for the future. These cars often feature innovative designs and technologies that may or may not make it into production models. If and when a concept car transitions to mass production, it undergoes significant changes to make it practical, affordable, and compliant with various regulations. This transformation often involves scaling back on some of the more ambitious or experimental features to meet market demands and manufacturing capabilities (case in point: the new Tesla Cybertruck). The mass-produced version is a more polished, realistic, and user-friendly vehicle, tailored to the needs and expectations of the mass market. Embedded within Warzal’s use of the concept car metaphor is a warning. He suggests that just like a concept car, ChatGPT has not undergone its transformation into its more polished, user-friendly, safe, version that meets the needs and expectations of the mass market.
I appreciate Warzal’s use of a metaphor to describe ChatGPT because I have been thinking a lot about metaphors lately. Metaphors are powerful. They are not just stylistic embellishments but foundational tools that play a critical role in shaping our understanding, perception, and communication about the world around us. They help us to map familiar knowledge onto unfamiliar or complex concepts, and in the process make those concepts more accessible. When Warzal uses the concept car metaphor, I don’t have to understand much about LLMs or ChatGPT for the matter to understand what he is trying to convey.
Metaphors are particularly helpful when it comes to design. They can make new technologies and experiences understandable and accessible to users by connecting familiar experiences with novel functionalities and providing a common language for designers and their users.
A classic e-commerce metaphor that we all have experienced is that of the “Shopping Cart.” Nearly all online shopping websites employ the metaphor of a physical shopping cart to help shoppers understand that they can "add" items to a virtual cart as they continue browsing, just as they would in a physical store. When they're ready to purchase, they can "view their cart" and "proceed to checkout," mimicking the real-world shopping experience.
But, as useful as metaphors are in making abstract notions tangible, they're not without their pitfalls. They risk oversimplifying, limiting, and even distorting, our understanding by confining it within the boundaries of the metaphor. Metaphors are also deeply rooted in the historical and cultural contexts from which they emerge, reflecting the prevailing beliefs, technologies, and paradigms of their times. Understanding the power, limitations, and risks of metaphors is crucial because they not only describe reality but also shape it.
It’s not hard to find examples of publicly-embraced, yet poorly chosen metaphors that have limited our imagination and led to a litany of bad decisions, policies, and consequences. Look at the metaphor, “The Economic Pie.” It suggests that a fixed amount of wealth exists, so if one person gets a bigger piece of the economy, someone else gets less. This oversimplifies economic growth, implying that it is a zero-sum game. It ignores that economies can grow, shrink, and evolve and that wealth can be created, not just redistributed. The metaphor also masks issues of systemic inequality by ignoring how the pie was even “baked” in the first place.
When it comes to interacting with generative AI, the metaphor that we most frequently use is that of the “Prompt Engineer.” What’s a Prompt Engineer? Hard Fork’s Kevin Roose recently said on an episode of the podcast that prompt engineering is, “one of the most interesting new jobs in the AI economy,“ and went on to define it as, “basically people whose job it is to figure out how to talk to AI language models, how to get the most out of them, how to generate outputs that are useful for your work or your personal life, and also to find the ways that they are broken and bad.” The Wall Street Journal’s Joanna Stern defines it even more simply as a person who, “fine-tunes the prompts that go into a generative-AI large-language model—aka LLM—to extract valuable but buried information for an employer or its clients.” When Stern searched for the term “Prompt Engineer” on LinkedIn and the job-search site Indeed, thousands of listings came up, with salaries ranging from $100,000 to over $200,000.
But is “Prompt Engineer” the correct metaphor to describe someone whose primary responsibility is “to figure out how to talk to AI language models” in order “to extract valuable but buried information for an employer or its clients”? I would like to suggest that it is not. For starters, it overemphasizes the role of science/technology in the act of prompting and deemphasizes the role of human psychology and communications. In addition, it implies that there is a rigidity to the process of prompting, especially around its methodologies and processes. It also ignores the critical ethical considerations that go into interacting with generative AI. Most importantly, if businesses think of their future knowledge workers as “AI Prompt Engineers” it will incorrectly shape their understanding and perceptions of generative AI and in turn, distort the interfaces that are designed for them to interact with it.
As a creative exercise, I’ve been brainstorming potential alternatives to the “Prompt Engineer” metaphor identifying different ways of describing a person who works closely with generative AI platforms to ensure their output is accurate, coherent, logical, ethical, and conveys the true intent of the humans seeking answers from it. I explored a wide variety of names, some were literal, and others were more whimsical and creative. I even created a few neologisms. Here are my favorites.
Alternative Metaphors to "Prompt Engineer"
A list of alternative metaphors to the concept of "prompt engineering" along with the pros and cons of each.
AI Ambassador:
Pros: Evokes a role of liaison and representation, emphasizing communication and diplomacy.
Cons: May not suggest the hands-on, detailed involvement in prompt creation and refinement.
AI Communication Analyst:
Pros: Suggests a focus on understanding and analyzing AI communications, aligning with the traditional roles of analysts as interpreters of information.
Cons: Analyst might imply a more data-focused role than one centered on language and communication.
AI Communications Manager:
Pros: Implies oversight and coordination of AI communications, suggesting a blend of strategy and content management.
Cons: Could be perceived as more administrative, potentially overlooking the creative aspects of prompt and shaping.
AI Content Moderator:
Pros: Implies a role that involves refining and moderating AI content, suggesting involvement in content quality, accuracy, and appropriateness.
Cons: The title doesn’t convey the interactive nature of the role.
AI Content Strategist:
Pros: Indicates specialized and strategic expertise in planning and overseeing AI content development, straightforward and professional.
Cons: Somewhat generic; doesn't explicitly convey the interactive, communicative, or empathetic aspects required.
AIFluent:
Pros: Implies ease and fluency in interacting with AI, suggesting adaptability and understanding.
Cons: Lacks specificity about the role's functions; more about proficiency than content management.
AI-Glot:
Pros: Suggests versatility in AI languages and systems, akin to an editor's adaptability in handling diverse content.
Cons: May not immediately convey the full range of editorial responsibilities.
AI Linguist:
Pros: Highlights expertise in the language aspects of AI, implying a deep understanding of communication between AI and humans.
Cons: The title might be perceived as overly academic or specialized, potentially not conveying the broader scope of interacting with and shaping AI content or output.
Human-AI Alignment Specialist:
Pros: Emphasizes expertise in aligning human and AI interactions, suggesting a focus on effective collaboration and mutual understanding.
Cons: The role might be too specific or narrow in scope, not clearly indicating the range of potential responsibilities. Human-AI alignment is not a widely understood concept.
Human-AI Interaction Designer:
Pros: Clearly focuses on designing interactions, which could include the careful shaping of language and content for AI-human interfaces.
Cons: Designer might suggest more of a creative planning role than the nuanced, adaptive work of iterating and refining generative AI output.
Prompt Designer:
Pros: Implies a creative role in crafting AI prompts, requiring an understanding of language and effective communication.
Cons: Focuses more on the creation aspect, potentially underplaying the role of editing and refinement.
Prompt Editor:
Pros: Clear and concise, directly relates to refining and working with AI prompts, making them accessible and understandable, and ensuring content quality.
Cons: While it emphasizes language and communication, it might be seen as narrowly focused on prompt refinement rather than the creation or design of a prompt.
While my personal favorites are “AI-Glot” and “AI Linguist” they lack the universal appeal or understanding required to be effective. Ultimately, the metaphor that makes the role the most understandable and accessible without oversimplifying, limiting, or distorting it is that of “Prompt Editor.” Prompt editing is a better metaphor for describing how humans ideally interact with generative AI. Editors are known for their expertise in language, clarity of expression, critical thinking, adaptability, and attention to detail, all of which are crucial skills for crafting effective prompts and refining AI-generated content. Editors have always functioned as gatekeepers and translators, ensuring the quality, accuracy, and appropriateness of content. To do their jobs, they must understand context, nuance, and relevance, and they must be collaborative, flexible, and empathetic.
In conclusion, the choice of the metaphor we agree to use to describe human interaction with generative AI is not just a semantic exercise; it's a reflection of our understanding, expectations, and the direction we intend to steer these interactions through design. The metaphor of a "Prompt Editor" transcends the technicalities of communicating with AI to capture the essence of human-AI collaboration. It emphasizes the critical role humans play in guiding, refining, and contextualizing AI outputs, ensuring they align with our ethical standards, cultural sensibilities, and practical needs.
This metaphor of “Prompt Engineer” also hints at a future where AI is not just a tool operated by experts, but a collaborative partner that requires human insight and oversight. It suggests a dynamic where AI extends and enhances human capabilities, rather than replacing them. By envisioning the interaction as editing, we acknowledge the iterative, nuanced, and sometimes imperfect nature of AI communication, and the ongoing human effort needed to shape and improve it.
Moreover, the "Prompt Editor" metaphor democratically opens the door to a wider range of participants in the AI field. It invites individuals with diverse skills in language, communication, and critical thinking to play a pivotal role in shaping our AI-driven future. This inclusivity is vital for ensuring that generative AI technologies serve a broad spectrum of needs and perspectives, making them more equitable and effective.
As we continue to navigate the evolving landscape of generative AI, our metaphors will undoubtedly evolve too, reflecting new understandings and aspirations. For now, the "Prompt Editor" serves as a fitting representation of our current stage in this journey – one that recognizes the importance of human judgment, creativity, and ethical responsibility in the realm of artificial intelligence. By embracing this metaphor, we not only define our current role in the generative AI revolution but also shape the path it will take, ensuring that it aligns with our highest ideals and serves the greater good.