Artificial intelligence models are increasingly mimicking human behavior—including our biases. A recent episode of the Symposium Podcast explores how psychological prompts, such as telling an AI to "breathe deeply" or adding pressure, can significantly improve model performance. This happens because AI systems are trained on vast amounts of human-generated data, so they inherit our cognitive patterns and shortcuts.
"Psychological prompts work precisely because the AI was trained on our data and reproduces our own biases," explains the podcast host.
The discussion delves into the implications of this phenomenon: as AI becomes more integrated into daily life, understanding its human-like biases is crucial. Developers and users alike must be aware that the biases embedded in AI outputs often reflect societal prejudices. The episode also highlights how prompt engineering—crafting prompts with psychological nudges—can make AI more effective in tasks like reasoning and problem-solving.
With AI adoption accelerating, recognizing these behavioral echoes is key to building fairer and more reliable systems.