Gray, C., & Bhattacharjee, I. (2023). Mind over machine? The psychological barriers to working effectively with AI. Humanity Redefined.
Claude summary of the piece:
- The article explores the psychological barriers that prevent humans from working effectively with AI models, even as these models become more capable.
- Studies show that humans are not good at choosing when to delegate tasks to AI models. They tend to under-delegate, even when the AI would perform better.
- Humans also struggle to objectively evaluate AI-generated outputs. When they collaborate in creating the outputs, they rate them higher than external evaluators.
- Accountability is tricky with AI collaboration. Humans get most of the blame if things go wrong, even if expectations of the AI were high.
- Training helps less than expected. Short classes do not improve delegation decisions or evaluation. More concrete process changes like guardrails are more effective.
- Many people have positive attitudes towards AI advances and are excited about productivity gains. But skills need development through hands-on learning.
- The article recommends explicit guidance on AI use, practical training focused on skills, and independent reviews of AI-generated work. Overcoming biases is key to unlocking the potential of AI.