All articles tagged with #data bias
A new analysis challenges the long-standing claim that brains of children with ADHD mature more slowly than those of their peers. By reexamining the data with more robust methods, researchers conclude that the landmark finding was likely a mirage caused by artifacts or biases in the data, suggesting ADHD brain maturation may proceed on a timeline similar to neurotypical development.
Claims that AI will soon surpass human intelligence rest on comparing machines to isolated cognitive benchmarks. The authors argue human intelligence is social, embodied, and cumulative—emerging from language, cooperation, and culture—while AI remains narrow, data-biased, and lacking genuine understanding or shared intentionality. Seen this way, AI is a powerful tool, not a mind superior to humanity, and hype distracts from governance, bias, and labor implications.
A new preprint suggests that daylight savings time can indirectly affect gravitational wave detection by altering human activity patterns, which in turn impacts the sensitivity and data quality of observatories like LIGO, potentially introducing systemic biases in gravitational wave astronomy.
Executives at the World Economic Forum in Davos discussed the potential of using blockchain to prevent bias in the data used to train artificial intelligence models, which could be a "killer use case" for the technology. By storing training data on the blockchain, developers can track and verify the data used by AI systems, allowing for the rollback of any false information or biases. This approach could provide a crucial check and balance system for AI, potentially revolutionizing the way AI models are trained and used across various industries.
Social media companies have been focusing their automatic content detection and removal efforts more on content in English than the world’s 7,000 other languages, leading to poor local language content moderation and contributing to human rights abuses. Multilingual large language models have been developed to address this issue, but they face challenges such as poor quality data, disparities in the amount of data they train on in each language, and the curse of multilinguality. Social media companies need to offer more transparency and accountability, prioritize individual language performance over scalability, and consult with language communities to improve content moderation.
A new study by Stanford University has found that both physical specimens and digital field observations are flawed and biased, with the degree of coverage gaps and biases depending on the type of dataset. The study revealed that observation-only records are biased and favor certain regions, time periods, and organisms, while preserved specimens are becoming scarce. However, both flawed datasets can still be useful for detecting shifts in the number and abundance of species in an area, and understanding their deficiencies can help improve future biodiversity data collection.