Micro UX - Week 5

Brief: Design a way for a person to pass as a ‘generative AI’ in an everyday setting.

Methods: Workshop hosting, Crazy 4's, Storyboarding

Team: Tanya Singh, Akriti Goel, Jakob Prufer, Ruoxi Song, Changlin Hou, myself

---

The primary objective for this week was to successfully host a workshop to test our ideas and receive useful feedback.

However, due to the challenges of securing a room at CSM we were short on time to advertise for participants. We created a flyer to send digitally and also hand out around CSM.

Graphic/flyer to advertise the workshop. Image by Akriti Goel.

After introducing our project to the participants, we outlined the workshop activities, which were a listening exercise, a storyboard activity and role-playing a scenario.

Listening Activity We played 14 clips to the participants and asked them to note whether they sounded like AI, Human or they couldn't tell which. The purpose of this activity was to understand what voices and phrases would more likely be accepted as AI, and how believable AI voices might be.

There was a fair amount on ambiguity or guessing, and a notable amount of AI being presumed to be human. The key findings were:

42% - Average accurate guesses for either human or AI

26% - Average instances where people couldn't make a decision

48% - Instances where AI was identified as human

Storyboard Exercise

Participants were asked to think about where AI exists in their lives currently, and then where they could envisage it becoming involved in our lives in the future.

We had set this up to see which 'everyday settings' AI might be more believable in.

Participants drawing out their storyboards: A speculative use for AI. Image by Akriti Goel.

Participants presenting their storyboards: A speculative use for AI. Image by Akriti Goel.

Findings from running the storyboarding exercise:

• People can imagine strange situations for AI use

• People needed extremely clear descriptions to undertake the task and extra prompts to imagine something speculative

• People worked at different speeds and used the storyboards in different ways resulting in varied outputs.

Scenario Exercise

Of the three people, one was acting as a human, rehearsing to go on a date; the other two were acting in collaboration to give AI-like answers. Participants acting as AI were given a 'cheat sheet' on AI characteristics to assist with their answers, which was derived from all the previous research.

Participants acting out an AI- rehearsal dating scenario. Image by Akriti Goel.

The key feedback notes we received from the scenario were:

"To give AI-like direct answers, how can we steer away from the conflict in our minds?" "To display a lack of emotions is an unnatural state for me personally" "It was obvious that 'AI' was making an effort not to be themselves" "The environment had a big impact on how I felt about saying these things - it wasn't real enough"

Overall, the workshop was successful and gave us insights, not only for our project but also how to host a workshop. We encountered access issues to the room, participants (and hosts!) running late, each activity took much longer than anticipated and anything not scripted in advance was slightly ambiguous to the volunteers.

Thankfully the people taking part were forgiving, plied with cookies and juice, however the hindsight of how unprepared for holding the workshop we were allowed us to be more critical of what we would do differently, if there were to be a next time.

We moved on to developing our conceptual idea around a dating rehearsal AI, which was actually powered by humans (some of the time). We played with some scenario-specific Crazy 4's, and sketched our thoughts over and over again until we had a consensus on the topic.

An afternoon of brainstorming: crazy 4's, storyboards and sketches. Image by Akriti Goel.

Our presentation was a review of the workshop we had held, and then talking through the simple storyboard we had pulled together - seen below:

Storyboard for the final presentation. Images by Changlin Hou

1: A host welcomes you, the audience to the demonstration of the dating AI.

2: The host selects a participant, and asks them about an upcoming date they would like to go on. The host takes details of the person the participant would like to as out, to calibrate the AI.

3: The participant enters the booth, puts on headphones and looks at the screen.

4: With the participant safely inside the booth, the backdrop is wheeled in and set up. The audience are aware, but the participant is not.

5: The participant inside the booth changes the level, looking for a more human-like, calibrated response. The host explains that whilst at the beginning of the experience, the responses they were receiving were AI generated responses, now the participant had 'levelled up', humans would now take over the responses.

6: The human-AI team converse with the participant using text-to-speech software, so the change in response is undetectable for the participant.

The presentation went surprisingly well, and while it was heavily suggested we take some layers out, and simplify the design and build, our concept stood strong on the foundations our research had laid.