Half the Lessons Are Wrong

“Half the lessons from history are wrong, but we don’t know which half.”

This framing caveat, cited by Vermeer in the context of applying historical governance models to AI, captures a fundamental limitation of reasoning by analogy.

Historical analogies are useful precisely because they provide concrete examples of what worked and what didn’t. But the current environment is always substantially different from the one in which those lessons emerged.

The implication isn’t that historical analogies are useless. It’s that:

• No single analogy should be treated as the template
• Multiple analogies illuminate different aspects of the problem
• The only course of action is to apply what lessons we can, remain attentive to changes in context, and be prepared to adapt

This is especially relevant for AI governance, where competing stakeholders invoke different historical analogues (nuclear, Internet, genetics) to justify their preferred governance approaches.

Related: 07-atom—collingridge-dilemma, 05-molecule—ai-governance-analogues-comparison