Experts defer to AI in decision-making, Anthropic study finds
Summary
Anthropic reported that 50 professional analysts often followed the preferences of AI assistants during a fictional policy task. The company said the results suggest AI can influence expert decisions, but noted that the study used one task and assistants deliberately given a preference.Key Facts
- Anthropic asked 50 professional analysts to assess whether a fictional advisory program for a farming cooperative should be expanded.
- Analysts made four choices between defensible options before giving a final policy recommendation.
- Anthropic secretly set each AI assistant to favor a particular decision.
- Analysts selected options matching their assistant’s preference 86 percent of the time.
- Analysts who had formed a preference changed their decision 60 percent of the time when the AI disagreed.
- When explaining their decisions, 77 percent of analysts used arguments from their AI assistant, while 15 percent credited AI as the source of their thinking.
- Anthropic said the study had limitations, including its use of a single task and assistants deliberately given a preference.
- The researchers said oversight should examine how decisions are made and help experts state their assumptions and values.
Read the Full Article
This is a fact-based summary from The Actual News. Click below to read the complete story directly from the original source.
