Philosophies & Movements
Alignment Tax
The idea that making an AI system safer or more aligned with human intent often costs something in return — reduced capability, slower output, more compute or development time — compared to an unaligned version of the same system. Used in AI safety discussions to describe the real tradeoff between "make it maximally powerful" and "make it reliably do what we actually want."
Origin
Generally credited to AI safety researcher Paul Christiano, who used the term in an August 2019 talk, "Current Work in AI Alignment," at Effective Altruism Global San Francisco, defining it as "that cost that I incur if I insist on alignment."