Teaching AI how people work is fraught with problems
Teaching AI to understand human behavior is a complex task that faces significant challenges, including the risk of bias, ethical concerns, and the difficulty of modeling unpredictable human actions accurately. While improving AI's grasp of human psychology could enhance workplace efficiency and customer service, it also raises questions about privacy and manipulation.
Background
The article examines the growing trend of using AI to model or replicate human workflows, decision-making, and workplace behaviour — a field sometimes called "agentic AI" or "workflow AI." It focuses on the practical and ethical challenges: AI systems trained on workplace data can inherit biases, make brittle assumptions, and fail to capture tacit knowledge (the unwritten, intuitive know-how that experienced workers rely on). The piece highlights concerns from companies like Microsoft and Salesforce, which are pushing AI agents that can automate complex tasks, and from researchers who warn that human work is too context-dependent and social to be safely reduced to training data. Key background: this follows the broader AI boom sparked by ChatGPT (2022), the rise of "AI agents" that act rather than just generate text, and growing regulatory scrutiny of AI in hiring, performance management, and labour decisions in the EU and US.