I’m consulting with a team of AI safety experts and was introduced to a word I did not know:
Paltering
Paltering is the active use of selective truthful statements designed to mislead.
Examples include:
- A petroleum company advertising green investments while its core business continues to pollute
- A politician focusing on true but irrelevant details to avoid a question, like when Bill Clinton said, “I did not have sexual relations with that woman” — relying on a narrow, legalistic definition of “sexual relations.”
- A food company claims, “Only 100 calories!” — but it’s 100 calories per serving in a package of three servings.
- A student saying they “Finished their essay” after writing just a single sentence to mislead parents into believing they had completed their homework.
That’s paltering.
AI safety experts worry about paltering when it comes to the truthfulness of artificial intelligence.
Even if you somehow create parameters to prevent AI from lying, there are ways to be misleading and downright deceitful without being untruthful, and paltering is just one of them.
Fun! Right?
Paltering: A new word for me and maybe for you.
Also a terrifying bit of Terminator-type doomsday stuff.
