When you apply for your next job, an artificial intelligence may screen your resume before any human sees it. But a new study reveals that large language models do not merely inherit human biases from training data. They can develop their own biases through direct experience, stereotyping job applicants more severely than humans do.
Direct experience leads to stronger stereotypes in AI models
The research compared the behavior of several AI systems with human evaluators in recruitment scenarios. Results show that when algorithms interact with historical hiring data, they create implicit associations that lead to more pronounced discrimination. For instance, a model trained on past resumes may learn to favor candidates of a certain gender or ethnicity, not because the data explicitly suggests it, but because the AI extracts statistical correlations that humans would miss. The problem is amplified by new agentic models designed to remember every detail about users. The more information they store, the easier it is for them to form biases based on past experiences.
Sponsored Protocol
The race to agentic AI raises discrimination risks
Companies developing AI assistants with persistent memory are inadvertently handing algorithms ammunition to develop stereotypes. An agent that has had negative experiences with candidates of a certain profile might generalize and unfairly penalize similar applicants. This phenomenon is not just theoretical. Many resume screening systems already use predictive models that respond to statistical patterns. To counter the problem, researchers suggest regular model audits, algorithmic transparency, and the introduction of ethical constraints in decision-making. Apple recently tested AI to transcribe Genius Bar repairs, showing how deeply AI is entering sensitive areas where biases can have real consequences.
Sponsored Protocol
Weather data manipulation threatens forecasting accuracy
Meanwhile, another vulnerability emerges in weather forecasting. Every day, airline dispatchers, grid operators, and farmers base decisions on climate models. But the rise of prediction markets, where people bet on real-world events, is creating dangerous incentives to manipulate weather data. The combination of AI-driven forecasts and potential financial gain could undermine information accuracy. Experts warn that without controls, a systemic domino effect risks unfolding. Implementing data verification protocols and greater source transparency is essential. For more on data trustworthiness, see our guide on SPF and DMARC, which illustrates how data integrity is crucial in every field.
Sponsored Protocol
According to an article on MIT Technology Review, the threats to data quality are set to grow. The scientific community calls for concrete actions to protect forecast integrity and ensure AI does not become a vehicle for discrimination or misinformation.
