Interview about Wel AI Personal Email Assistant App, winner of the A' Mobile Technologies, Applications and Software Design Award 2024
The concept integrates AI into daily email processes, facilitating analytical reading and offering a conversational approach to seamless email composition, guidance, and drafting. The design stands out for creating a flexible and respectful collaboration between humans and AI, provides assistant that is just right. The design also cultivates a personal writing style and a social relationship library, benefiting both the user and the AI, encouraging growth and improvement like a supportive buddy in life.
View detailed images, specifications, and award details on A' Design Award & Competition website.
View Design DetailsThe key to achieving this balance lies in our deep understanding of how AI processes and transforms human input. Through extensive testing and iterations, we carefully observed the differences between human communication styles and AI-generated content. Our most innovative approach was moving beyond standard neutral tags like "formal" or "polite" to introduce what we call the "Your Style" tag. This tag starts as a blank canvas and learns from both the user's original inputs and their subsequent refinements, creating a personalized style profile.Rather than simply having AI write emails, we're using AI to help users understand and refine their own communication style. Through this long-term learning process, users can improve their expression accuracy while the AI develops a deep understanding of their unique voice. Unlike traditional approaches where everyone gets the same "formal" or "professional" options, each user's style tag evolves differently, reflecting their individual communication patterns.
The design learns from two primary sources: the continuous flow of email communications and user interactions with the platform. By analyzing both outgoing and incoming emails, the AI builds an understanding of relationship dynamics and communication patterns. This is complemented by our unique "Your Style" tag system, which evolves as users draft, refine, and optimize their messages with AI assistance.For social relationship management, we analyze email exchanges to create detailed contact profiles. After each interaction, we update these profiles and present users with refined relationship tags, such as team member or client, along with interaction history summaries. This creates a comprehensive social relationship database.Importantly, we've implemented what we call "progressive assistance." Rather than immediately taking over email writing tasks, we start with subtle suggestions and gradually increase AI involvement as both the user and system mature. If users show hesitation or uncertainty, we scale back the assistance level. This approach ensures that users develop their communication skills alongside the AI, creating a collaborative growth experience rather than dependency.
Our research revealed several fascinating insights about human-AI interaction. First, we discovered that the boundary of acceptable AI assistance isn't fixed but rather dynamic, expanding or contracting based on user trust and AI performance. Users initially prefer minimal AI involvement, but as trust builds through successful interactions, they become comfortable with more comprehensive assistance. However, this comfort zone can contract if the AI makes mistakes, creating a breathing dynamic boundary.Second, we found that maintaining user control over final decisions is crucial, even in cases where users trust the AI completely. This is similar to having a launch button for a rocket – the final control should always rest with the human operator, providing a sense of agency and responsibility.Perhaps most surprisingly, we discovered that users often prefer AI that writes like them rather than perfectly. Even if the AI occasionally makes minor mistakes while matching their style, users showed greater trust in the system compared to one that produced flawless but impersonal content. This insight significantly influenced our design approach, leading to the development of our personalized style tags alongside standard writing options.
The tag-based interface emerged from extensive research into existing email assistance software and understanding how users naturally categorize and process information. We observed that email summary tools were already using categorical tags, which users had grown accustomed to. Additionally, tags proved highly effective for training our language models.Initially, we experimented with traditional summary formats, but found that users struggled with processing large blocks of text, regardless of length. This led us to explore various visualization methods, including graphics, tags, and voice interfaces. We discovered that the same tags used for AI training could serve as excellent quick-reference points for users, instantly communicating key information about sender identity, purpose, and tone.The final design combines relationship tags (e.g., "team member"), purpose tags, and tone indicators to create a comprehensive but quickly scannable overview. For instance, seeing "Urgent feedback from team member" with an appropriate tone tag immediately sets the context before the user even opens the email, reducing communication anxiety and improving efficiency.
Our approach differs fundamentally from chatbots because we designed this system before the ChatGPT era, focusing purely on user needs rather than following current AI interaction patterns. Instead of forcing users into a chat interface, we integrated AI capabilities directly into the familiar email interface.The assistant is available on-demand rather than dominating the interaction. Users can activate different levels of assistance based on their needs – from simple tag summaries to phrase suggestions, paragraph rewrites, or full email responses. This graduated approach to AI assistance maintains efficiency without requiring users to engage in constant chat-style interactions for every task.The design focused on providing the right level of help at the right moment, scaling from minimal to comprehensive assistance based on user preference and context. This progressive integration approach sets us apart from chatbot-style interfaces that often force users into a rigid interaction pattern for every task.
Interestingly, in the years following our project, we've seen many AI tools adopt similar color schemes. Our choice was driven by specific psychological and practical considerations. We wanted to convey the fluid, dynamic nature of AI assistance while clearly distinguishing between AI-generated content and objective information.For AI-generated content and optimized responses, we used gradient blues to suggest fluidity and ongoing evolution. For objective information like email text and fixed tags, we maintained a clean white background to represent unchanging, factual content.The choice of blue specifically came after testing various colors. For a daily-use email tool, we wanted to avoid colors that might trigger strong emotional responses (like red or yellow) or imply definitive judgments (like green). The combination of blue and white creates a neutral, calming environment that doesn't interfere with the emotional content of the messages themselves. This allows the AI's outputs to carry their own emotional weight without being influenced by the interface design.
Our design provides a vision for future AI-assisted software that acknowledges and adapts to different levels of user needs. Whether users need quick tag summaries, word substitutions, sentence rewrites, or full response generation, the system flexibly adapts to these varying requirements based on context, content, and user confidence levels.While current AI software often provides a one-size-fits-all approach, we believe AI should be intelligent not just in its capabilities but in how it integrates into user workflows. It's gratifying to see some of our design concepts, such as clear visual distinction between AI-generated and objective content, being adopted by current AI tools.However, the key innovation we hope to see more widely adopted is the intelligent scaling of AI involvement. The future of AI assistance isn't just about making AI smarter, but about making it more contextually aware of when and how to assist users, providing a spectrum of support rather than a binary all-or-nothing approach.
These distinct modes emerged from our research showing that users approach email composition and relationship management with fundamentally different mindsets. The Assistant mode focuses on immediate communication needs, analyzing content and providing real-time suggestions for improvement, while the Librarian mode helps users organize and understand their broader network of relationships.The Librarian mode acts as a knowledge repository, helping users maintain a comprehensive view of their professional relationships and communication patterns. It provides historical context and relationship insights that inform future interactions.These modes work symbiotically – the Librarian mode's accumulated knowledge enhances the Assistant's ability to provide contextually appropriate suggestions, while the Assistant mode's interactions contribute to the Librarian's growing understanding of relationship dynamics. This dual approach creates a more comprehensive and nuanced email management experience.
My background in autonomous vehicle design significantly influenced my understanding of human-AI collaboration. In self-driving vehicles, AI assists but doesn't completely take over, with humans remaining ultimately responsible – similar to our email assistant's approach.The automotive industry taught me three crucial aspects of human-AI interaction: the importance of split-second timing, the potential for serious consequences from mistakes, and the risk of over-reliance on automation. While email communication isn't as time-critical as driving, these principles helped shape our approach to user engagement and safety mechanisms.This experience also provided valuable insights into testing methodologies for AI systems. The rigorous testing requirements in automotive applications helped us develop robust verification processes for our email assistant, ensuring reliability while maintaining user agency.
Our award-winning approach demonstrates that ethical AI integration isn't just about capability limits – it's about thoughtful implementation that respects user agency and promotes growth. We've shown that AI can enhance rather than replace human communication skills through our progressive assistance model.The app contributes to the ethical AI conversation in three key ways. First, it demonstrates how AI can preserve and enhance individual communication styles rather than standardizing them. Second, our transparent approach to AI involvement, clearly indicated through visual design, sets a standard for honest AI integration. Finally, our focus on user growth rather than dependency shows how AI tools can empower rather than replace human capabilities.We believe this approach can serve as a model for future AI tools, showing how to balance automation with human development, and how to integrate AI assistance while maintaining user autonomy and authenticity in communication.
Dive into a world of design excellence with our curated highlights. Each feature showcases outstanding creativity, innovation, and impact from the design world. Discover inspiration and learn more about these incredible achievements.
