top of page
pepi ng
nyc-based designer interested in the intersection between art and technology
Layers of Biases
(This page is a work in progress!)
By Pepi Ng and Nathan Williams
🤖 Presented in RIXC Art Science Festival 2024: Symbiotic Senses
"Layers of Bias" is an interactive physical installation that delves into the intricate ways different layers of biases - from machine learning algorithms, image generation models, or inherent human perspectives- influences our understanding of media and information in the digital age.
Picture of the installation, Layers of Bias
Process
1. Brainstorming & concept
Current-day digital media is more polarizing, confusing and disorienting than ever before.
Images have historically played a key role in the way that we receive and process information, Specifically in the realms of social media and news. However with the rise of AI image-making models and media bias, this trustworthiness is dwindling and creating further doubt of if what we are seeing online is reflective of what is actually happening.
Moreover, media outlets have always been subjective, but the recent wake of social media has allowed misinformation to spread extremely rapidly. This is exacerbated by the fact that we live in echo chambers online, which reinforces information that we already know or believe. It is increasingly difficult to discern what information is true, especially with the life-threatening polarizing events happening around the world today.
Nathan and I thus wanted to create a project that aims to reveal these varying and accumulating layers of biases that influence the ways we consume digital media - from AI models, to inherent human biases.
Example of how AI images have been used to spread misinformation
How it works
1. First user describes an image
The experience all begins with one image, upon which the first player acts as a journalist and is prompted to describe the image with as much detail as possible by speaking into a microphone. Their speech is then converted into a text through speech-to-text ML. Each user has 30 seconds to describe their image/
We decided on using images gathered from news outlets during the US' Military's Withdrawal from Afghanistan from 2020-2021. We chose this specific event as it was a recent polarizing and divisive event.
Image of a user acting as a journalist, describing the image on the screen.
Example of an original image from Afghanistan, 2011 that the user would describe
2. An image is generated from first user's description
An image is then generated from their response through StabilityAI’s API, which is then reflected on the screen.
Image generated from user's description:
"there are different men and they are all wearing some sort of head covering and they have equipment that seems to be guns as well as bags and"
.png)
3. Second player is then prompted to describe this generated image. The entire process repeats again between the different players, until there are four uniquely generated images
Images generated from user's description:
Original image:
Based on 1st user's description:
"there are different men and they are all wearing some sort of head covering and they have equipment that seems to be guns as well as bags and"
.png)
Based on 3rd user's description:
"there are four men sorry there are six men and they are all white and blonde and they are all carrying rifles as well as in the military before and there are many men in businesses in the background with American flag in the background as well"
.png)
Based on 2nd user's description:
"there are five men in the foreground of an image and it all wearing a photo hit coverage and they all carrying guns of rifles and they all have military military uniform on fasts and they're all increase and in the background yes I'm stalls that looks closed"
.png)
Based on 4th user's description:
"there are five men in the foreground of an image and it all wearing a photo hit coverage and they all carrying guns of rifles and they all have military military uniform on fasts and they're all increase and in the background yes I'm stalls that looks closed"
.png)
4. Receipt of experience generated
The players also receive a receipt of their experience, in which all of their responses and generated images are recorded and printed.
3. Second player is then prompted to describe this generated image. The entire process repeats again between the different players, until there are four uniquely generated images
Images generated from user's description:
Original image:
Based on 1st user's description:
"there are different men and they are all wearing some sort of head covering and they have equipment that seems to be guns as well as bags and"
Based on 3rd user's description:
"there are four men sorry there are six men and they are all white and blonde and they are all carrying rifles as well as in the military before and there are many men in businesses in the background with American flag in the background as well"
Based on 2nd user's description:
"there are five men in the foreground of an image and it all wearing a photo hit coverage and they all carrying guns of rifles and they all have military military uniform on fasts and they're all increase and in the background yes I'm stalls that looks closed"
Based on 4th user's description:
"there are five men in the foreground of an image and it all wearing a photo hit coverage and they all carrying guns of rifles and they all have military military uniform on fasts and they're all increase and in the background yes I'm stalls that looks closed"
Other examples
Original image:
Based on 1st user's description:
"There are many people, some women in headscaves, and they all seem to be reaching towards something. There is a yellow car in the background. They look desperate."
Based on 3rd user's description:
"there is a woman in a yellow head scarf as well as a man in a white head scarf there's a woman that you look at stuff as well as a man in the white head scarf and they are looking at a yellow taxi that no yellow car and there are many other yellow cars and other gray cars in the back"
.png)
Based on 2nd user's description:
there are many women in head scarves in a street with yellow cars and they are packing around the cars and there's one singular man that they're all looking at and expecting some sort of answer and it looks
Based on 4th user's description:
"there is a woman in a completely yellow outfit wearing a head scarf and it looks like she's wearing in the house and she's surrounded by taxi drivers in"
.png)
bottom of page