As the demand for intelligent agents grows, having a human-centric user experience is more important than ever. We’re still early in the life cycle of these AI tools, and the best practices for designing and developing AI tools aren’t well known. At Architech we’ve been fortunate enough to have been asked to help our partners develop AI agents in a way that navigates the delicate balance between efficiency and humanity. To learn more about our Collections AI solution that streamlines your debt recovery process by speaking with your customers through fluent voice, text and email, check it out here.
Based on some of our conclusions building AI agents, below I've compiled 10 tips that we’ve found useful in transforming the journey for design and engineering teams grappling with the task of making an AI agent not only valuable but genuinely comfortable for human users.
1. Humanizing is Not the Same as Imitating
It’s a common misconception that making an AI agent human-centric means imitating human behavior, even going so far as to have your agent pretend to be human, passing itself off as a human employee, and claiming to be human when asked. Here’s the thing - no matter how good your agent, it will carry some element of the uncanny. Users will notice, and you’ll lose trust. There’s nothing to be gained by lying about being an AI agent.
Conversely, you gain a lot of trust in the tool by having it be forthright about its nature. An AI agent is an interface; to use it effectively, a user needs to understand what kind of interface it’s interacting with. Focus on creating an interface that aligns with user needs, expectations, and emotions without mimicking human behavior artificially. Also – users aren’t dumb. They’ll figure out that they’re talking to an AI agent. And they won’t appreciate being tricked.
2. Expect Curiosity from Your Users
As users engage with your AI agent, they will - naturally - be curious about its capabilities and limitations. This is an opportunity to build trust and a positive user experience. By anticipating common questions and incorporating answers to into your design, you not only provide valuable information but also demonstrate transparency - which is key to building trust.
A transparent agent reassures users about its capabilities and boundaries. In turn, it also encourages your users to explore the full potential of the agent, inviting them to make the most of its features and functionalities from a place of curiosity rather than apprehension.
3. Expect Distrust from Your Users
Conversely, users may be carrying preconceptions, prejudice, and/or had previous negative experiences with AI agents, and may approach your tool from a position of distrust.
Acknowledging user skepticism is crucial. Incorporate features like explainability, offering insights into the decision-making process and limitations of the AI agent, to alleviate concerns. Have a backup solution, such as a human agent, available for users who are completely unwilling or unable to use your AI tool.
4. Build a Persona for Your Agent
You needn’t go so far as to give your agent a name, face, and history, but your team should all understand the character, tone, and personality of the agent you’re building. A persona is a great way to make sure everyone is on the same page. Before getting too deep into prompt engineering, consider bringing the whole team together to collaborate on a persona – that way everyone can contribute to, understand, and agree on what the agent will say and do, and work together to develop consistent prompts and outputs.
5. Foster Trust in the Organization, Not in the Agent
While users' interactions with the AI agent may be brief, their engagement with the organization behind the AI agent could be long-lasting. Therefore, it's more valuable to focus on establishing trust in the organization offering the AI tool rather than solely in the AI agent itself. Trust in AI is closely linked to the reputation of the organization backing it.
6. Model the Language You Want the User to Use
Generally, users have better conversations with AI agents when the exchanges are short and focused. Long, technical explanations and info-dumps are frustrating, especially over the phone. In many cases, some LLMs have trouble extracting useful information from user responses that are long and unfocused.
Model the types of inputs you need your agent to receive by having it speak in the same way: e.g. in short sentences, focusing on key phrases. Users will naturally begin to mimic this style.
7. Think in Turn Pairs
In the context of conversation design, the concept of "turn pairs" refer to the sequence of exchanges between a user and an AI agent, where each turn consists of a statement from one party followed by a response from the other. This concept acknowledges that conversations typically unfold in a back-and-forth manner, with each participant taking turns to speak or respond. Understanding turn pairs helps in designing AI agents that can engage in coherent and contextually aware interactions, ensuring a smoother and more intuitive conversation for users.
When designing statements, think about the types of response the statement might generate, and think about the types of statements your agent might get back, and what kinds of responses it might prompt, and so on. This helps in creating a more coherent and contextually aware AI agent and contributes to a smoother and more intuitive conversation.
8. Lean on Human Conversation Practices & Conventions
Leverage established human conversation practices and conventions to make the interaction between your users and your AI agent feel familiar.
Design the user interface to resemble conversations with existing frameworks that a user might recognize, and keep to typical conversational norms. After all, we already know how to have a conversation. For example, when designing an agent for answering the phone, expect that users will have interacted with call centre agents before and leverage the standards and practices of a call centre worker when designing for your AI agent.
9. Pay Attention to When Human Conversation Practices & Conventions Don’t Make for Good Experiences
While following human conversation practices is a good tip, be aware of situations where they might hinder user experience. In short, some real-life practices when speaking with humans lead to negative experiences when speaking with AI agents.
Using tip #8 as an example, call centres often train their workers to be apologetic and apologize profusely for inconveniences. This is actually a worse experience when interacting with an AI agent, as it tends to make the conversation drag on without offering anything useful.
Be mindful of the realities of the conversation’s context and user needs, as well as the realities of conversing with AI, to avoid awkward or frustrating experiences.
10. Don't Let the LLM Run Wild
Frankly, LLMs on their own can’t be trusted. They are imitation machines, designed to generate a plausible imitation of a response to a prompt – but they have no critical or logical faculties.
Instead, build your agent as you would a technology stack, with the LLM as the topmost layer between user and system. Ensure there are careful logic, gates, and code behind the scenes to ensure the LLM stays on track and minimize embarrassing or harmful hallucinations and mistakes. Regularly review and update the model to align with evolving ethical standards.
In summary
Humanizing AI agents is an ongoing process that involves a combination of thoughtful design, ethical considerations, and user-centric practices. By implementing these 10 tips, we hope you can benefit from some of our conclusions in creating AI agents that are efficient but also resonate with users on a deeply human level.
Explore the Art of possible with a FREE AI ideation workshop
Engage with our team of AI, Data, Engineering, and Design experts to co-create customized AI solutions that target your enterprise challenges and align with your 2024 objectives.
In this 3-hour session, we'll work together to transform innovative concepts into actionable strategies. By the end, you'll walk away with a comprehensive blueprint for AI integration, including a prototype plan outlining the optimal technologies, tools, and methodologies to swiftly develop a proof-of-concept. Register now to kickstart your AI journey!
Connect with our team today to secure your free session!
