Context
6 months project in 2021
Project type
Final major project for my MA
Role
UX Research
Team
Maria Seves and Eda Erdal
The decision-making tool for AI Personality
💭 How does your chatbot represent your brand?
It's time to craft experiences that digital personal assistants embody, moving beyond their flow and dialogue structure. An AI's personality significantly shapes the interactions humans have with it. By developing a detailed backstory and carefully considering behaviours and goals, we can enhance people's engagement with the brand.
🪞 AI Transparency
This research formed part of my final project for my MA in UX Design. It resulted from a design investigation into how we can bring more transparency to human-AI interactions.
🎭 This is… Backstory
Backstory is an online toolkit that helps designers create AI backstories. This pipeline takes users from the ideation of their AI to a materialised and concrete project. Through a series of steps, creators reflect and craft their AI's goals, anatomy, personality, and potential social impact. The journey culminates in the generation of a personalized AI Manifesto.
🔑 Key Insights
Starting this project from a single research question, we had the opportunity to explore different generative research methods in addition to evaluative ones.
The Creative Process
The kick-start for this project was a Research Question
How can we bring more transparency to Human-AI Interactions?
To answer this question, Eda and I partnered to investigate through design to understand some key questions: what does AI mean? How does AI make people feel? And why does Alexa look like a Magic 8 ball?
As all good UX goes, the process took us a couple of months, multiple UX Methods, a few coffee cups, and many iterations to reach our final design. Here is a glimpse of our design process.
Concept Mapping
During our initial meetings, we created a concept map to organize our ideas and make sense of AI, given its abstract nature. In our discussions about AI, we drew an intriguing parallel: we likened chatbots to fortune tellers, operating in mysterious ways to provide glimpses into the future.
Emotion Mapping
We conducted directed storytelling workshops with 8 participants to explore perceptions and emotions about AI. The exercise involved assessing emotions in various AI scenarios. The most common emotions were worried, excited, and acceptance, particularly when reflecting on self, participation, and activities.
Key findings included:
- Participants used metaphors to explain AI concepts
- Those with more AI knowledge expressed greater concerns about data exploitation and control
- Mixed feelings prevailed: satisfaction with AI's benefits but anxiety about future implications
- Limited use of voice assistants among participants
- Concerns about the digital divide and lack of AI literacy
The exercise effectively sparked discussions about AI's impact and the importance of addressing these topics despite their abstract nature.
Artefact Analysis with Alexa
For the next step in our design research, I got Amazon’s Alexa and used the Artefact Analysis methodology to study it and generate insights. My first goal was to understand why Alexa was such a big hit, as it is a device that sold millions in the US in 2019.
Material: Alexa is grey and looks plastic. Its base has a ring of LEDs and a visor that tells the time. Unfortunately, it is impossible to open it without destroying it, so I let it be for now.
Aesthetic: Its spherical format and the ring of light-changing shades remind me of a magic ball from a fortune teller. It looks futuristic but not personal.
Interactive: I used Alexa and sometimes had trouble understanding my accent. While it was helpful for specific questions like finding nearby locations, it still needs to be more flexible and smart to give me personalized information.
Make your own Alexa workshop
For our “Make Your Own Alexa” methods. First, we conducted unstructured interviews to gauge participants’ familiarity with Alexa and their thoughts on the device. Participants then interacted with our Alexa, followed by a workshop where they used Play-Doh to design enhancements. This resulted in five new Alexa prototypes, and here are the key findings from this activity:
- Understanding Preferences: One participant was disappointed when Alexa suggested a recipe they disliked, highlighting potential gaps in Alexa's ability to understand personal preferences.
- Accent Recognition Issues: Participants with accents experienced difficulty as Alexa struggled to understand them, causing frustration.
- Referring to Alexa: Participants often used "her" due to Alexa's female voice, despite some reluctance to anthropomorphize the device.
- Voice Customization Requests: Many participants wanted to change Alexa’s voice, but the device could not accommodate this request.
- Design Enhancements: Participants added human-like features (e.g., mouths, eyebrows) to make Alexa appear more emotional, despite research advising against anthropomorphizing AI.
- Personalization Choices: One participant, an illustrator, chose not to give Alexa eyes, viewing this as a deeply personal decision.
Crazy 8 and Storyboards
After learning about people's relationships with AI, it's time for a brainstorming session. The Crazy 8 exercise yielded diverse ideas. Next, we sketched storyboards of our top ideas. After sharing the storyboards with our lecturers and colleagues in our MA cohort, the winning idea was the AI Toolkit, a toolkit for developing AI projects, especially chatbots and voice assistants, to help define their personas.
User Journey
We used journey mapping to visualize the user experience process, plan the toolkit usage, and understand any missing interactions. Our tools will help prepare this AI project's visual and auditory aspects. We have also planned a "nice to have" third part for the toolkit, which we pursue only if we have extra time. We want to focus on establishing the personality and tone of voice rather than using other tools for conversational flow planning.
Card Sorting
From our previous research methods, we acquired a long list of attributes related to personality and tone of voice. We held an adapted cart sort session to decide which should be part of our project. This technique helps us understand what people see as opposites regarding the tone of voice and how people cluster personality traits.
Wireframing
We iterated using different methods, starting with paper and then evolving our design using various levels of fidelity. We consulted with people for each stage and adjusted our design based on their feedback.
Concept test
After completing the second version of our website, we consulted three experts in Conversational Assistants. Overall, specialists were impressed with the tool.
One positive aspect of the concept was the possibility of using it with large teams and creating a guide to share with the company. Some feedback was provided about missing functionalities they would like to see.
The positive feedback motivates us to continue refining the tool, and we worked on a new iteration before doing our last round of testing and collecting feedback.
User Testing
We invited six people with different levels of knowledge about creating CAs to test our design.
First, we asked how familiar each one was with AI design and whether they ever planned for the personality of the AI. Each participant was then invited to follow the journey on the website. As we only had the interactive prototype on Figma, we required them to take notes of each answer they gave on a piece of paper, manually generating the AI Manifest. In case the participant didn’t have any chatbot solutions to use for the journey, we had two prompts for inspiration:
- Design a chatbot for a University to help students decide which classes to take.
- Design a chatbot for doctors and nurses working at A&E to help them treat incoming patients.
Overall, the people testing were excited about the purpose of this project. These are some of their thoughts:
- One participant saw this as a valuable tool for a follow-up to generating ideas in design thinking sessions. They felt the journey was complete and wanted to explore the topic more.
- Participants liked how it generated a physical artefact that can be used as proof of their ideation and as a document to present to people from other teams.
- Participants enjoyed how the journey was structured. They felt it helped them plan better, as they started by reinforcing the project's goals and target audience.
- Participants also enjoyed learning about personality and tone of voice. Those who work with chatbots expressed how they try to consider personality in their designs. Additionally, it would be crucial to have these Manifest when designing the dialogues and flows for the AI.
- There was some controversy regarding the anatomy stage, where participants chose the appearance of the AI. Further testing and research are necessary to improve this.
Reflecting on our work 💭
This FMP was a long, arduous work of research. We adapted many design methods better to understand AI and people’s relationships with it. We ended up creating a simple platform that was well-received by our target audience.
Reflecting on our research question to bring transparency to human and machine relations is good. By helping designers make conscious choices about how their AIs are portrayed, we are a step closer to making connections more transparent for the end user.
Try out the prototype ✨
Click on the Figma file to open the prototype and navigate our toolkit.
