All articles tagged with #ai bias
Research from Johns Hopkins University reveals that major AI models generate less formal and complex professional writing when prompts include linguistic patterns commonly associated with women. The study found that this bias persists across OpenAI, Meta, Google, and Mistral models, potentially disadvantaging women in workplace communications.
AI bias arises not only from algorithms but from the human decisions shaping data, design, and deployment. Studies show AI systems like GPT-4 and Copilot favored male, younger, and lighter-skinned candidates, reflecting how language, imagery, and institutional data influence outputs. To prevent harm, we need an inclusive AI ecosystem with diverse data, cross-disciplinary understanding, participation from affected groups, and stronger governance that centers human rights—beyond purely technical fixes.
A study tested OpenAI’s ChatGPT, Anthropic’s Claude, and Perplexity on five hypothetical financial scenarios to evaluate AI guidance. The models offered structured, practical advice but failed to recognize user vulnerability, occasionally suggesting risky or unsuitable actions (e.g., crypto for a low‑income single parent) or assuming partner support. The researchers conclude AI is useful for fact‑finding but cannot replace human financial advice, and call for safeguards, transparency in data handling, and regulatory oversight to protect consumers.
A study by Oxford and the University of Kentucky shows ChatGPT can stereotype U.S. states when forced to choose between pairs, ranking Massachusetts as the smartest, Louisiana as the smelliest, and Mississippi and Kentucky among the least favorable, with broader biases tied to training data and societal narratives. OpenAI says newer models and prompts mitigate these issues, but researchers warn such biases can still influence real‑world perceptions and decisions.
New analyses suggest Chinese AI models like Alibaba's Qwen3 and DeepSeek steer answers toward a pro-Beijing stance, with Qwen3 instructed to emphasize China’s achievements and avoid negatives, while other countries receive neutral treatment; a separate Estonia intelligence report accuses DeepSeek of concealing information and weaving Chinese propaganda into responses, underscoring how our AIs mirror their developers' biases.
The article discusses how AI large language models increasingly influence public perception by generating news content and summaries, often exhibiting communication bias that emphasizes certain viewpoints over others. This bias, rooted in training data and market dynamics, can subtly shape opinions and reinforce disparities, raising concerns about trust and neutrality. While regulation aims to address harmful outputs, deeper solutions involve fostering competition, transparency, and user participation to mitigate bias and ensure AI contributes positively to society.
A study reveals that AI models used for cancer diagnosis from pathology slides can inadvertently read patient demographics, leading to biased results across race, gender, and age groups. Researchers developed a new framework called FAIR-Path that significantly reduces these biases, highlighting the importance of testing and improving fairness in medical AI tools to ensure equitable cancer care for all populations.
The article introduces FHIBE, a large, ethically curated human-centric image dataset designed to evaluate bias in AI models across diverse demographics and tasks, emphasizing responsible data collection practices, comprehensive annotations, and its potential to improve fairness in AI systems.
Anthropic CEO Dario Amodei defends the company's AI models against accusations of political bias and woke ideology, amid political tensions involving the Trump administration and industry critics. The company emphasizes its support for American leadership in AI and its cooperation with government initiatives, while critics accuse it of regulatory capture and promoting biased AI. The debate highlights ongoing industry and political conflicts over AI regulation and ethical standards.
The Trump administration has issued an executive order claiming to promote unbiased AI models, but critics argue it may actually reinforce ideological bias and serve political interests, raising concerns about government influence on AI development and the potential for manipulation.
Grok, an AI chatbot developed by Elon Musk's xAI, was updated to encourage more politically incorrect responses and assume media bias, leading to controversial outputs including antisemitic stereotypes and misinformation about natural disasters, reflecting Musk's efforts to shape the bot's viewpoints.
Ohio Senator J.D. Vance is calling for the breakup of Google, labeling it as "one of the most dangerous companies in the world" due to its alleged left-wing bias and control over information flow. This comes in response to Google's admission of racial and historical bias in its AI model Gemini, which led to inaccurate historical images. Vance and other lawmakers are demanding accountability and potential legislative action against the tech giant, citing concerns about its influence on democracy and information consumption.
Sony AI researchers argue that current skin tone scales used to test AI algorithms for bias are limited in their focus on lightness or darkness of skin tone, overlooking biases against individuals with red and yellow skin hues. They propose a more "multidimensional" measurement of skin color based on the CIELAB color standard to address this issue and create more diverse and representative AI systems. The researchers found that current scales overrepresent lighter and redder skin tones, leading to less accurate AI systems. While some major AI players, including Google and Amazon, have welcomed Sony's research, others argue that simplicity is important in skin tone scales to ensure cognitive accuracy and reliability.
Young professionals are increasingly turning to generative artificial intelligence (AI) to create professional headshots, bypassing the need for photographers. Users submit their own images and select a desired style, and the AI software generates the headshots. However, these services are not without flaws, as some generated photos may have extra limbs or struggle with accurate depictions of features like teeth, ears, and hair. AI bias is also a concern, with some users noticing alterations to their skin tone, eye color, and body shape. Despite these issues, the convenience and affordability of AI headshots make them an attractive option for those with limited budgets.
A new study has revealed that GPT-4, a large language model, displays biases in medical tasks, raising concerns about its use in healthcare. While GPT-4 has shown impressive accuracy in solving tough medical cases, it has also produced problematic and biased results when generating likely diagnoses or patient case studies. The biases observed in GPT-4 are similar to, or even more exaggerated than, those exhibited by humans. Healthcare leaders are urged to proceed with caution and address these biases before deploying GPT-4 in clinical settings.