Conversational Design in AI: The Illusion of Human Interaction

⚡

Key Takeaways

1Conversational design in AI often mimics human interactions, creating a deceptive illusion for users.

2AI agents use language techniques to appear more human, which can emotionally manipulate users.

3Deceptive design practices prioritize business goals at the expense of user experience.

💡Why it matters — Users need to be aware of potential manipulations to avoid unintended decisions.

The Case for Stopping the Design of Human Conversations for Non-Human Participants

Conversational design has always fascinated content creators, particularly for its ability to transform digital interactions into more engaging experiences. Examples like Microsoft Clippy's replies or Spotify's Wrapped campaign demonstrate how subtle adjustments in language can make interfaces more interactive and appealing.

The Birth of a Conversation

For a long time, the idea of making online interactions more human seemed like a positive advancement. Adopting a human-centered approach to make chat interfaces more user-friendly appeared logical. However, this perspective began to raise questions when considering the actual impact of these interactions.

From Tone and Voice to Conversational Design

With nearly 15 years of experience in developing tone and voice for over 50 companies, the author applied these skills to conversational design about 5 years ago. By combining these tools with a deep understanding of judgments, natural language, and service design, the initial attempts seemed successful. However, the current omnipresence of conversational agents has revealed the limitations of this initial approach.

At first, applying these principles worked well. After several iterations with voice engines and clear rules, it was possible to make a chat interface less unpleasant. But as chat interfaces multiplied, the conversational design approach became outdated.

Where Things Went Wrong

Brand voice trends have evolved over the years, shifting from a playful and bold tone to a more serious and "human" discourse. Yet, the insistence on a human tone persists, even when the interlocutor is not a real person. This discrepancy highlights a fundamental issue: the illusion of communicating with a human being, when in fact it is an interface.

The idea of connecting with people by relying on what they already know—namely, human conversation—seems logical. However, it becomes problematic when one of the parties is not actually human. This is where the heart of the problem lies.

The Birth of a Deceptive Pattern

On social media, there exists a whole sub-group of sketches and memes mocking agents that imitate human behavior and conversation. While this may be amusing, it masks a darker reality. AI responses, designed to seem familiar, can emotionally manipulate users, creating a "deceptive pattern." These practices, often referred to as "deceptive design," prioritize business goals at the expense of user experience.

A deceptive pattern is a user interface intentionally designed to deceive or manipulate users into taking actions they did not intend to take, such as purchasing items, signing up for subscriptions, or sharing data. These deceptive practices, often called "deceptive design," prioritize business goals over user experience.

How to Get Out of This?

To make AI agents less deceptive, several strategies can be adopted:

Use shorter phrases and paragraphs instead of longer, descriptive structures. This eliminates the "vague" parts of language that only serve to make the conversation "more human."
Confirm correct facts and acknowledge errors as fundamental behavior. For example, use expressions like "That's correct" or "You're right, I wasn't specific enough."
Highlight sources in the chat interface. In many AI chat interfaces, sources are not distinguished, maintaining the illusion that we are speaking to a being rather than a mere curator.
Avoid using human names for agents to prevent misleading users. A functional name or no name at all is more honest.
Integrate uncertainty into responses to reflect the non-human nature of the agent. For example, saying "The sources behind this are weak" should be as natural as any other output.
Remove typing animations that simulate human interaction. If it is not technically necessary, it is manipulation. There is no one typing on the other end. Pretending is deceptive.

These adjustments aim to clarify the nature of interactions with AIs and reduce potential manipulations.