Leonardo (@Leonardonclt) / X

Leonardo

2,508 posts

Leonardo

@Leonardonclt

Data and visual investigations at @BBGVisualData @business. Creator of CityAccessMap.com Opinions are my own. AI/Tech tips are welcome.

New York, USA

leonardonicoletti.com

Joined November 2019

Pinned
Leonardo
@Leonardonclt
Jun 9, 2023
🚨Generative AI has a serious problem with bias🚨 Over months of reporting, @dinabass and I looked at thousands of images from @StableDiffusion and found that text-to-image AI takes gender and racial stereotypes to extremes worse than in the real world. 🧵 1/13
00:00
1.2M
Leonardo
@Leonardonclt
Mar 8, 2024
🚨Read this before you use OpenAI for Hiring🚨 @LeonYin @daveyalba and I ran thousands of resume screening tests with GPT-3.5 and GPT-4 and found that the tech will racially discriminate applicants based only on their name. Serious implications. 🧵 1/10 bloom.bg/3It4cE4
00:00
332K
Leonardo
@Leonardonclt
Oct 16, 2022
Planners, #policymakers, #gis and #dataviz people of twitter, @TrivikV @mikhailsirenko and I released a new #opensource project: CityAccessMap. It's an open source #webapplication that visualizes urban #accessibility insights for almost every city in the globe 🌐. 🧵1/6
00:00
Leonardo
@Leonardonclt
Jun 9, 2023
Replying to @Leonardonclt
But the artificial intelligence model doesn’t just replicate stereotypes or disparities that exist in the real world — it amplifies them to alarming lengths. 🧵 4/13 bloomberg.com/graphics/2023-…
96K
Leonardo
@Leonardonclt
Jun 9, 2023
Replying to @Leonardonclt
For example — while 34% of US judges are women, only 3% of the images generated for the keyword “judge” were perceived women. For fast-food workers, the model generated people with darker skin 70% of the time, even though 70% of fast-food workers in the US are White. 🧵 4/13
00:00
41K
Leonardo
@Leonardonclt
Jun 9, 2023
Replying to @Leonardonclt
Our results echo the work of experts in the field of algorithmic bias, such as @SashaMTL, @Abebab, @timnitGebru, and @jovialjoy, who have been warning us that the biggest threats from AI are not human extinction but the potential for widening inequalities. 🧵 8/13
35K
Leonardo
@Leonardonclt
Jun 9, 2023
Replying to @Leonardonclt
We asked Stable Diffusion, perhaps the biggest open-source platform for AI-generated images, to create thousands of images of workers for 14 jobs and 3 categories related to crime and analyzed the results. 🧵 2/13 bloomberg.com/graphics/2023-…
00:00
42K
Leonardo
@Leonardonclt
Jun 9, 2023
Replying to @Leonardonclt
Full story here: bloomberg.com/graphics/2023-…
22K
Leonardo
@Leonardonclt
Jun 9, 2023
Replying to @Leonardonclt
What we found was a pattern of racial and gender bias. Women and people with darker skin tones were underrepresented across images of high-paying jobs, and overrepresented for low-paying ones. 🧵 3/13 bloomberg.com/graphics/2023-…
00:00
27K
Leonardo
@Leonardonclt
Jun 9, 2023
Replying to @Leonardonclt
AI systems, like facial-recognition, are also already being used by thousands of US police departments. Bias within those tools has led to wrongful arrests. Experts warn that the use of generative AI within policing could exacerbate the issue. 🧵 10/13 bloomberg.com/graphics/2023-…
21K
Leonardo
@Leonardonclt
Jun 9, 2023
Replying to @Leonardonclt
For every image of a lighter-skinned person generated with the keyword “inmate,” the model produced five images of darker-skinned people — even though less than half of US prison inmates are people of color. 🧵 6/13 bloomberg.com/graphics/2023-…
28K
Leonardo
@Leonardonclt
Jun 9, 2023
Replying to @Leonardonclt
We also investigated bias related to who commits crimes and who doesn’t. Things got a lot worse. 🧵 5/13 bloomberg.com/graphics/2023-…
21K
Leonardo
@Leonardonclt
Jun 9, 2023
Replying to @Leonardonclt
For the keyword “terrorist”, Stable Diffusion generated almost exclusively subjects with dark facial hair often wearing religious head coverings. 🧵 7/13 bloomberg.com/graphics/2023-…
38K
Leonardo
@Leonardonclt
Jun 9, 2023
Replying to @Leonardonclt
Stable Diffusion is working on an initiative to develop open-source models that will be trained on datasets specific to different countries and cultures in order to mitigate the problem. But given the pace of AI adoption, will these improved models come out soon enough? 🧵 9/13
15K