Beyond Commands: The Art of Communicating with AI

Every day, technology seamlessly intertwines with our daily lives. Our conversations with machines have evolved beyond simple commands to dialogues. Generative AI, the latest frontier in this evolution, challenges us to rethink our approach to interacting with technology. No longer can we rely on the straightforward directive; instead, we find ourselves engaging in a dance of dialogue, leading and being led by AI in a collaborative journey. This new mindset demands not just technical skill, but a nuanced mastery of conversation, empathy, and guidance. A skill everyone thinks they have, but questions when ChatGPT doesn’t give them the answer they want.

From Directive to Dialogic

To give a clear example, let’s say traditional software functions like a child who hasn’t yet developed the theory of mind, operating strictly within the parameters of its programming and not understanding that other people don’t have the same knowledge about situations as themselves. The Sally-Anne experiment is a classic psychological test used to assess a child’s ability to attribute false beliefs to others. In this experiment, a child watches one doll, Sally, hide a marble in a basket and then leave the scene. Another doll, Anne, moves the marble to a box. Upon her return, the child is then asked where Sally will look for the marble. The test evaluates whether the child understands that Sally still believes the marble is in the basket.

This concept becomes very relevant again with conversational AI. This AI navigates conversation through context, inference, and the data it has been trained on. For users, this means engaging with AI not through rigid commands but as participants in a dialogue, understanding that the AI’s “perspective” is shaped by its training data and algorithms.

Just as a child learns to consider Sally’s belief about the marble’s location, users must guide the conversation with an awareness of the AI’s processing mechanisms.

Understanding AI’s “Thought” Process

The Sally-Anne test underscores the importance of recognizing that others operate based on their own set of information. Similarly, understanding how AI models generate responses—recognizing their reliance on vast datasets, patterns, and algorithms—enables users to frame questions and prompts more effectively. This understanding allows for adjustments in the way questions or commands are structured, anticipating how the AI might interpret them based on its “knowledge” of the context.

An issue that frequently occurs is that people forget to add context because they usually don’t have to. A practical example of this is letting ChatGPT write a poem, and getting back a poem that does not rhyme. For instance, in a conversation between real people, it is understood that a poem follows an AABBCC rhyming scheme unless stated otherwise. But now imagine if this conversation were held between 2 poets, they would further specify what kind of poem they want, should it rhyme? Do you want a Shakespearean sonnet?

Empathy with AI Capabilities

Developing empathy towards AI involves recognizing its capabilities and limitations, akin to understanding that Sally does not have the same information about the marble’s location. By empathizing with the AI’s “perspective”, users can better navigate interactions to achieve desired outcomes. This doesn’t mean anthropomorphizing AI but rather acknowledging the framework within which it operates, enabling more effective and fruitful interactions, and leading to more meaningful insights.

Overcoming Challenges in Human-AI Communication

Interacting with generative AI presents unique challenges, but understanding and strategic approaches can lead to more effective and satisfying exchanges. Here are some tips for your conversations:

• Misinterpretation of Instructions: Ensure clarity by providing detailed and specific instructions. For example, instead of asking for “a report on market trends,” specify the exact type of report needed, including subject, time frame, and focus areas.
• Limited Context Understanding: Keep the AI aligned with your objectives by summarizing or restating the context in ongoing dialogues, helping it maintain focus and relevance.
• Unexpected Responses: When AI goes off-topic, direct the conversation back with refined, targeted prompts and feedback. This helps maintain the focus and encourages more relevant responses.
• Creativity Limitations: Guide the AI towards your creative vision by providing examples or descriptions of the desired style or outcome. This can significantly improve the alignment between AI-generated content and user expectations.

By applying these strategies, users can enhance their interactions with AI, turning challenges into opportunities for more meaningful and productive dialogues.

Blog by Dennis Hulsebos