Designing A Delightful AI Personality – UX Planet
🎯 Problem Statement (aka Oops I Forgot Something Big)
A few months later, I was chatting with Diane Kim who has done pioneering work on AI interaction design at x.ai and told her about my plans for building an AI assistant for conferences. She asked me,
“Who would your AI be if they were a real or fictional person?”
I didn’t have an answer to her question. I was so focused on just building an AI chatbot that I completely forgot about my design roots. In this crowded marketplace of AI assistants, it’s the personality that will be the differentiating factor that brings delight to the user. This was the ‘Aha’ moment for me and I wrote down the problem statement:
How Might We design a delightful AI personality to reduce the workload on the conference organizers?
I reached out to the MidwestUX Conference team with my proposal of building an AI chatbot for them pro bono. Not only did they agree but the organizers even provided me a dedicated team to help launch this idea off the ground.
🎤 Creating a voice for my AI (aka Voice Style Guide)
My first order of business was to create a voice style guide for the chatbot. Just like websites have digital style guides that define the color, layout, fonts, etc., a voice style guide describes the personality of a chatbot. What are the bot’s frequently used words? What adjectives describe the bot? Who is (and is not) your bot? Without a voice style guide, your bot will come off as a confused piece of code with 23 split personalities hidden in it. At that point, your frustrated end user will stop talking to your bot and instead watch M. Night Shyamalan’s movie Twitch.
The bot’s voice must reinforce the brand-as though the brand itself were speaking to you. I asked Anja Harris, the conference organizer, for guidance here. She and her team provided me the conference brand guidelines as a starting point.
Using the conference brand guidelines, I created a voice style guide for the MidwestUX chatbot.
Once the voice style guide was created, came the hard part which is ensuring that all the bot content copy aligned to the voice style guide. For example, consider the scenario below:
QUESTION: Conference schedule
BOT REPLY: Visit this link to see the conference schedule
How do I make this bland answer light-hearted and fun? What emotional state will this question trigger in the mind of the attendee? These are some of the questions I asked myself as I iterated on the content copy to add some pun.
☀️ Designing For Delight
The key to shipping Minimal Delightful Products over Minimal Viable Products is paying attention to the small things that may not seem important at first, like emojis 🤔.
Take a look at an early prototype of the chatbot below that had no emojis. There’s nothing wrong with the prototype — it works, but it’s bland. No personality.
After talking with the folks at Donut 🍩, I realized the power of emojis to bring raw content to life. I started adding relevant emojis ( 🎫, 🏨, 💻, etc.) to the action buttons. Lo and behold, the prototype came to life!
Many attendees would be interacting with a slackbot for the first time at the conference. With that in mind, an easy to use help was an early design consideration. Generally, the first conversational exchange between two unknown human beings is to exchange greetings. So, I programmed the chatbot such that the moment you greeted AI sage, there was an option to get help by clicking on the
Get help action button.
At any point in the conversation, if you typed the trigger word
help you would be redirected to the help command.
Command list would basically list the types of commands recognized by the chatbot with examples. I did this deliberately to prevent attendees from trolling the chatbot.
What if the attendee’s question wasn’t listed in the command list? To tackle this problem I built a live agent functionality similar to an interactive voice response system you encounter when calling a big company’s helpline number. Once the attendee would select
help category they would be connected to the appropriate person on the organizing team via a direct message on slack.
While conducting user research for this project, I found that searching for a particular speaker was one of the most frequently completed actions by conference attendees. Traditionally, an attendee would go to the conference website, scroll till the end of time, find their speaker’s talk details and then sprint! I split the speaker details into a two column layout with bold headlines for easier readability and save space (mobile users don’t want to be bombarded with long ass passages of text). Last but not least, I also coded an action button
Connect which would redirect to the speaker’s twitter page and save the attendee time from having to google the speaker.
If the speaker was giving a workshop, I even added the option to buy tickets via the chatbot.
And yes, I color coded the workshop and talk details differently on purpose to avoid confusion.
⚙️ Machine Learning With IBM Watson
I used IBM Watson on the backend to make AI Sage actually smart, not just sound smart. For example, attendees can ask the basic question, ‘Where is the conference taking place?’ in a thousand different ways. It’s not feasible for me to program all those scenarios into the chatbot. When an attendee would ask that question in a way not understandable by AI Sage, I would go into the Botkit console and train the system to recognize that input.
Over time with more examples and training, AI Sage became wise. So when an attendee typed
mwux venues instead of typing out the whole command ‘Where is the conference taking place’, AI Sage replied with delight:
The conference took place over four different venues. Once the attendee would select a venue from the dropdown, AI Sage would provide details related to that location.
I shared my journey of building AI sage with the MidwestUX conference attendees. After my talk, I wandered through the conference hallways. I saw attendees delighted and laughing when they chatted with AI Sage. So, why did I spend all those sleepless nights building AI Sage? The answer lay in a brown colored card I held in my hands. This small thank you note from the organizers meant more to me than any form of monetary payment or hockey shaped usage growth. It reaffirmed my belief that anyone can make an impact in their community regardless of connections/credentials/expertise as long as they are willing to hustle!
After the conference success, I took a Greyhound bus to Austin and met Ben. He was kind enough to take me out for dinner. I vented to him about my numerous failed attempts at breaking into the startup world. I asked him point blank if there was light at the end of this tunnel of frustration. As he paid the bill, Ben looked into my eyes and said,
“I have been in the startup game for ten plus years. People will notice when you keep making cool shit. Are you patient enough to keep hustling until they do?”
I thanked him for the dinner and took the midnight Greyhound bus back to Dallas. Next morning, a news article caught my attention and made me smile. A chatbot entrepreneur from Texas called Ben Brown had sold his company to Microsoft.