Study: GPT-4V can evaluate social situations similar to humans (www.utu.fi)

A recent study from the University of Turku reveals that GPT-4V, an advanced version of OpenAI’s language model, can evaluate social situations from images and videos with human-like accuracy. Researchers tested GPT-4V on 138 social features—from facial expressions to cooperative or hostile interactions—and found its assessments closely matched those of over 2,000 human evaluators. Notably, the AI’s consistency outperformed individual humans, though collective human insight remains superior. This marks a significant leap in AI’s ability to understand complex social dynamics beyond mere object recognition. The implications extend strongly into neuroscience, where interpreting social contexts in brain imaging studies is labor-intensive. Using GPT-4V’s rapid and reliable social evaluations, researchers successfully mapped brain networks related to social perception, achieving results comparable to human-derived data. By automating this process, AI can dramatically reduce the time and cost of large-scale neuroscience experiments, processing data in hours rather than thousands of human work hours. Beyond research, the study highlights practical applications in healthcare, marketing, and security. AI’s round-the-clock monitoring could help medical professionals track patient well-being, gauge audience reactions to advertisements, or identify unusual activity on surveillance cameras. As AI becomes adept at interpreting nuanced social cues, it promises to augment human decision-making by handling continuous observation and leaving humans to focus on critical judgments.