Prototype: impression checker

Final Major Project / 10

October 30, 2021

Luchen Peng, Tiana Robison, Jinsong (Sylvester) Liu

The first prototype, “your right hand man”, revealed challenges in delivering a comprehensive concept. While still trying to fix it, we want to explore an alternative object that aims to provide a more clear image. So we spent some time reviewing the previous workshop’s video and slides and finding out participants often generate and talks about ideas with facial recognition. It seemed to present a direct association with gender and racial bias and therefore reduce the effort in understanding.

Low-fidelity prototype

Luchen already had serval critical object sketches using facial recognition. This time she also synthesised and created a new product called Impression Checker. We assisted Luchen in quickly building a low-fidelity prototype, using only cardboard, printed paper and sone sponge from Poundland. It shows inspirations from the dynamic background (Siqi) and camera filters (Xander and Jasper), but focus more on how facing an unexplainable AI create pressure and confusion on users. The object lives in ai-supported video interviews where candidates need different strategies to adapt. The Impression Checker is designed to simulate an ai­ interview and give suggestions on improving backgrounds, facial expressions and postures.

Functioning prototype

The low-fidelity prototype shows potential in visualising the context and encouraging reflections on gender and racial bias. Our next step is to realise it in a working model. We first asked for help from Creative Tech Lab at LCC and had the webcam, thermal printer and initial code. Although they suggest we include more buttons to fulfil the complex script, I spent lots of time exploring and finally figured out a way to refine the software and build a consistent interaction: press the same button to continue printing.

In the coming tutorials, we tested the concept with Ai, Wan, Greg and other students. The object worked better than the “right-hand-man” since all participants immediately started curiously following the instructions on receipts. We could recognise them drawing a more apparent association to the theme of data bias through observation and discussion. Moreover, the experience is full of laughter and display great opportunity in modifying the script and props. Greg even suggested we make it random to create more chaos. John pushed us to keep the “AI-suggestions” open: maybe ask people to dress like women and see how they respond.

In review, I compare the Impression Checker to the previous ideas and conclude some reasons that make it successful:
  • Obvious context: it lives at present where Al interview is expanding and somewhat frustrates many candidates. People don’t need to accept a futuristic scenario to be convinced.
  • Simple interaction: I was ambitious to introduce complex interactions in the slapping machine idea, but I have to admit that the “press the button and follow the instruction” design reduces the complexity and leaves more space in perceiving the hidden bias. But I would argue the simplifying also shrink the open interpretation a lot.
  • Visualised data pool: We made a translucent box with loads of images to represent the biased training data. This separate setup assists us in describing the idea and even self-explained.

The following post will focus on context building: making a speculative commercial stand to present our project.