The National Portrait Gallery is Jared Soares's favorite museum. It's just a few Metro stops away from the photographer's home in Northeast DC, and he says he's visited dozens of times to admire the works from his favorite artists. But Soares's next visit will be different. The second floor of the gallery now features Soares's award-winning photograph, Misidentified by Artificial Intelligence: Alonzo and Carronne (2023).
One thing that has always fascinated me is how an innocent, dispassionate analysis can still reinforce biases and exacerbate societal problems. Looking at crime rates by district, for example, shows which area has the highest rate. Nothing wrong with that. The issue emerges when that data leads to reallocating police resources from the lowest-crime district to the highest or changing enforcement emphasis in the higher-crime district.
A "discriminatory" artificial intelligence (AI) model used by Sweden's social security agency to flag people for benefit fraud investigations has been suspended, following an intervention by the country's Data Protection Authority (IMY). Starting in June 2025, IMY's involvement was prompted after a joint investigation from Lighthouse Reports and Svenska Dagbladet (SvB) revealed in November 2024 that a machine learning (ML) system being used by Försäkringskassan, Sweden's Social Insurance Agency was disproportionally and wrongly flagging certain groups for further investigation over social benefits fraud.
Its imperfections are human, visible, and correctable. You can see who edited what, when, and why. Grokipedia is its antithesis. It replaces deliberation with automation, transparency with opacity, and pluralism with personality. Its "editors" are algorithms trained under Musk's direction, generating rewritten entries that emphasize his favorite narratives and downplay those he disputes. It is a masterclass in how not to make an encyclopedia, a warning against confusing speed with wisdom.
"We're like, 'Show us a picture of a senior leadership team from Chanel visiting Microsoft'-it is all men in suits," she said.
Researchers have analysed hundreds of thousands of images from places like IMDb and Google Image Search as well as the kinds of online texts used to train large language models. They found a persistent age gap between men and women when it comes to how they are shown and talked about online.
AI is able to build things in minutes or even seconds that would take humans hours to do. But its biggest strength is also its biggest flaw. AI is, at its core, a tool used to imitate human knowledge in an advanced capacity. So while it's able to produce work faster, it also inherits all our biases. It's important to address these biases when building products to ensure that what we're developing can be used by everyone.
Generative AI is moving from drafting emails to shaping labor markets. On platforms like Fiverr, Freelancer.com, and Upwork, millions of workers rely on hourly rates to compete for jobs. As AI increasingly influences pricing recommendations, business leaders face a critical question: Do large language models (LLMs) make these pricing decisions fairly? Or do they perpetuate the same biases and inequities that have long plagued human labor markets?
"I scraped millions of Google Maps restaurant reviews, and gave each reviewer's profile picture to an AI model that rates how hot they are out of 10," says San Francisco-based website creater Riley Walz. "This map shows how attractive each restaurant's clientele is. Red means hot, blue means not."
"The findings, detailed in a preprint paper titled "AI Self-preferencing in Algorithmic Hiring: Empirical Evidence and Insights," further amplify persistent concerns about AI bias. Authors Jiannan Xu, a PhD candidate at the University of Maryland, Gujie Li, assistant professor at the National University of Singapore, and Jane Yi Jiang, assistant professor at Ohio State University, found that "LLMs consistently prefer resumes generated by themselves over those written by humans or produced by alternative models, even when content quality is controlled."
The goal of the Ethical AI Guidebook is to spark a mindset shift-encouraging content creators, agencies, and AI developers to make inclusivity a core part of how they generate and use imagery.
As opposed to investigating deepfake detection models, the study looks more broadly at AI models used for "fake news detection." This highlights systemic biases in technology.
