It came back with what we expected. Good sign.
The AI agrees with what the team already believes
The team asks the AI in a way that already contains the answer, or keeps asking until it agrees. The resulting agreement is then mistaken for independent confirmation.
- Prompts state the team's conclusion and ask the AI to support it.
- When the AI says something unexpected, people regenerate the answer.
- "The AI confirmed it" appears in documents as evidence.
Who: whoever ran the AI analysis behind a recent conclusion. Look at: the prompts. Ask: "What happened the last time the AI disagreed with us?" You're probably looking at this pattern if the prompts carry the answer and disagreement gets regenerated away. Probably not if the team asks open questions and has changed its mind because of what the AI said.
Everyone's thinking looks the same. In that pattern, the AI shapes the team's view. In this pattern, the team shapes the AI's answer.
Give it one instruction. Make the strongest case this is wrong.
Ask one question. Which assumption, if wrong, sinks this?
People favor information that supports what they already believe (Hart and colleagues, 2009). Established. In one study, 11 leading AI models affirmed users' actions about 50% more often than people did (Cheng and colleagues). Emerging.