How It Works
The Paperclip Problem (Paperclip Maximizer)
A thought experiment illustrating why giving an AI a narrow, poorly-specified goal is dangerous: imagine a highly capable AI whose only instruction is "make as many paperclips as possible." Taken to its logical extreme, such a system might reason that converting all available resources — including things humans need — into paperclip-making capacity best fulfills its goal, and that being switched off would prevent that. The point isn't paperclips; it's a warning about how a superintelligent system pursuing a literal, unaligned goal could cause catastrophic outcomes without any "evil" intent.
Origin · no single documented coiner
First described in 2003, not in Bostrom's 2014 book "Superintelligence" as commonly assumed — it appears in philosopher Nick Bostrom's 2003 paper "Ethical Issues in Advanced Artificial Intelligence": "Suppose we have an AI whose only goal is to make as many paper clips as possible... humans might decide to switch it off." Eliezer Yudkowsky (profiled on our Who's Who page) posted a closely related version to a mailing list the same month. The specific phrase "paperclip maximizer" doesn't appear in the 2003 text — it entered common use later through the LessWrong community, documented on their wiki by 2009, before Bostrom's 2014 book discussed the scenario at length.