Who prompts the Prompter? Understanding human-AI interactions

I’m bullish on good UX. In a world where most people have access to best-in-class LLMs like Claude 3.5 Sonnet and GPT-4o, it’s not enough to just connect your data through an API.?

If moats still exist, your UX will be the strongest.?

In other words: how easy is it for users to achieve their desired result with your product?

Forcing users to be “Prompt Engineers” is not a viable business model. The average human needs to be prompted just as much (if not more) than the platform they are using.?

Good outputs demand a symbiotic relationship where both the human and the AI are treated as promptable systems. I call this "double-sided prompting".

The thing is, it’s not a new concept. It’s just basic UX principles operating in the context of chats.

Human prompting

As users, we are being prompted all the time. So much so, that it’s easy to forget that it’s happening.?

From “mind the gap” on train platforms to “please have your boarding pass ready” at the airport, simple phrases guide our actions and our experiences daily.?

What’s more, prompts affect our habits. For example, Google guides searches through the “People also ask” feature. Instead of keeping follow up questions open-ended, it gives us a choose-your-own adventure with limited options.?

Of course, humans can veer off the path of prompting. We have free will. We can start another search query or intentionally dip our toes into the gap.?

But these “prompts” make it easier to navigate the world around us - whether it’s in real life or on the web. Having preset options and alerts are good for the user experience. And it’s what most gen AI tools are lacking today.?

Double-Sided Prompting

Double-sided prompting is the idea that in any interaction between a human and an LLM, both parties are simultaneously guiding and being guided.?

The human prompts the AI with questions, tasks, or data, and in return, the AI prompts the human with suggestions, clarifications, and additional questions that help refine the input and output.

This two-way interaction ensures that the AI doesn’t just generate responses in a vacuum but actively engages the user to improve the quality of the interaction. It’s a dance where both partners are leading and following, constantly adjusting to achieve the best possible outcome.

In other words, chats shouldn’t be one-sided. They should be conversations.?

Principles of double-sided prompting

These principles for double-sided prompting are nothing new. They are a combination of best practices from UX, HMI and conversational design (shout out to Erika Hall)!?

While the user’s goals and wishes remain centered,?

1. Human prompting comes first.?

It takes expertise to look at a blank canvas and create art. Most of us need a paint-by-numbers to create something significant.?

Just like the early web necessitated HTML knowledge to create personal websites, we saw genAI tools that required a lot of user knowledge to maneuver. The fact that “prompt engineering” became a common phrase points to how difficult it has been for the average user to get what they want out of genAI systems.?

The web evolved. Now we have tons of no-code websites and apps that people don’t have to learn an additional skill to manuveur.?

More and more, I’ve been seeing genAI chats support users through the prompting process.

Pi includes pre-written prompts and showcases them in a newsfeed style.?

Claude often asks clarifying questions to understand user intent.?

1. Preventing bad outputs > Fixing bad outputs

Understanding and responding to the user’s context is vital. Whether it’s a website interface or a chatbot, systems should be aware of the user's needs, preferences, and environment. In double-sided prompting, context awareness is critical for the AI to ask the right questions and for the user to provide relevant information. This ensures a seamless interaction where both parties are aligned.

Exa does this well. Rather than waiting for user to craft the perfect prompt, it interprets the intent of a search and autoprompts it for the best results.?

2. Feedback Loops Effective UX design incorporates feedback loops that allow users to see the immediate impact of their actions. In conversational design, this principle is reflected in how users receive confirmations, updates, or clarifications in real-time.?

For machine-human interfaces, feedback loops ensure that interactions are not only responsive but also adaptive to user input. In double-sided prompting, feedback loops enable both the human and the AI to adjust their prompts and responses, refining the conversation iteratively.

Going back to the exa example, showing the user the autoprompt allows for further iterations. I searched for “early web” and it was prompted as “Here is a link to explore the early days of the web”.?

When the results weren’t quite what I needed, I understood why. I took the structure of the autoprompt and adjusted it to fit my needs. The system then autoprompted again for a refined search that yielded exactly what I was looking for.?

3. Clarity and Simplicity Good UX design strives for simplicity, ensuring that interfaces are intuitive and easy to navigate. Conversational design shares this goal by making interactions straightforward and avoiding unnecessary complexity. In machine-human interfaces, clarity is essential to prevent confusion and ensure that users understand the system’s responses. Double-sided prompting leverages this principle by prompting both the AI and the user to keep their exchanges clear and focused, enhancing mutual understanding.

4. Cooperation and Goal Alignment Effective conversations require cooperation and a shared understanding of goals. This principle is also central to UX, where user journeys are designed to help users achieve their objectives efficiently. In machine-human interfaces, cooperation is about creating systems that assist users in achieving their goals, not just processing inputs. In double-sided prompting, both the human and the AI are aligned towards a common goal, making the interaction more collaborative and productive.

The Benefits of Double-Sided Prompting:

1. Enhanced Collaboration: Double-sided prompting fosters a collaborative environment where both the human and AI contribute to the outcome. This leads to richer, more nuanced outputs that are better aligned with the user’s goals.
2. Increased Accuracy: By allowing the AI to prompt the human, potential misunderstandings or ambiguities can be addressed in real-time. This reduces the risk of errors and ensures that the AI’s outputs are more accurate and relevant.
3. Empowered Users: When users are treated as promptable systems, they become more engaged in the interaction. They are encouraged to think critically, ask better questions, and ultimately take more ownership of the AI’s outputs.
4. Continuous Learning: Double-sided prompting creates a feedback loop where both the AI and the user learn from each interaction. The AI refines its understanding of the user’s needs, and the user becomes more adept at guiding the AI.

Designing for Double-Sided Prompting:

To implement double-sided prompting effectively, AI designers must focus on creating interfaces that facilitate this two-way interaction. This involves:

• Context-Aware Prompting: The AI should be able to ask relevant questions based on the context of the interaction, guiding the user to provide the most useful information.
• Iterative Refinement: Encourage users to refine their inputs through AI-generated suggestions, ensuring that the final output is as accurate and relevant as possible.
• Transparency: Clearly communicate the AI’s thought process and reasoning, so users understand why certain prompts or suggestions are being made.
• Feedback Mechanisms: Incorporate feedback loops where users can easily correct or adjust the AI’s prompts, leading to continuous improvement on both sides.

Conclusion:

By treating both the human and the AI as promptable systems, we move towards a more collaborative, accurate, and empowering interaction model. As AI continues to evolve, embracing this paradigm will be key to guiding users to their desired outputs.

Rather on relying on users to learn “prompt engineering”, system should be as user-friendly as possible. This means that a key team member to hire is actual UX designers. In particular, content designers with strong skills in information architecture and conversational design will be key to a future where chats actually feel like assistants instead of tools.