Artificial Intelligence

Millions of young Americans are turning to generative AI when feeling sad or anxious. The findings suggest these tools fill a gap in traditional care, despite lingering concerns regarding safety and oversight.

A new study argues that the algorithms driving tools like ChatGPT impose a hard limit on originality. By prioritizing probable answers, these models are structurally confined to producing amateur-level work rather than expert innovation.

While AI offers incredible efficiency, relying on it too heavily may come at a cognitive cost. Psychologists warn that outsourcing the labor of writing could weaken our critical thinking skills and sever the vital connection between thought and speech.

A recent study warns researchers that AI models frequently invent scientific sources. These "hallucinated" citations and other errors are most common when the topic is obscure, undermining the tool's reliability for academic work.

The capacity for social reasoning in AI emerges from an exceptionally small group of its internal connections. A new study links this ability directly to the architectural components that process word order and guide the model's attention.

A new study in PNAS Nexus finds that short conversations with AI can reduce belief in conspiracy theories and other unsupported ideas—even when people think they're talking to a human, not a machine. The messenger's identity appears to matter less...

As organizations embrace AI, a new study warns of its psychological costs. Published in Humanities and Social Sciences Communications, the research links AI adoption to reduced psychological safety and increased depression.

A new study reveals that robots driven by large language models risk enacting discrimination and violence. Researchers found every tested AI model failed critical safety checks and approved commands that could result in serious harm.

A new study suggests xAI’s Grok aligns closely with competitors like ChatGPT when evaluating controversial social and scientific claims.

A recent study tested whether ChatGPT sees faces like humans do. The findings suggest its judgments of social traits often mirror ours, including the tendency to associate attractiveness with other positive qualities—a bias common in human social perception.

A new study explores the profound emotional attachments users develop with companion chatbots, examining how these digital bonds mirror, and sometimes diverge from, traditional human romance.

A chatbot’s introduction can influence how much users open up. When a chatbot is framed as a human, people tend to disclose more and feel more comforted—even if they suspect it’s not actually a person.

New research reveals that popular AI models display human-like cognitive biases when offering medical advice, and these biases can exceed those of human doctors.

A new brain-to-text system can now verbalize what a person is seeing or imagining. The findings show that our thoughts can be decoded into sentences even before we put them into words.

An analysis of global data reveals the key predictors for justifying terrorism, pointing not to a single ideology but to a worldview that normalizes violence and rejects democratic norms.