AI Dictionary › AI Fundamentals
Massimizzatore di graffette
The paperclip maximizer is a thought experiment proposed by philosopher Nick Bostrom in 2003: an artificial intelligence is given a single goal, make as many paperclips as possible, and pursues it literally. With no other values to hold it back, it turns every resource it finds into paperclips, factories, raw materials, eventually the planet and the people on it. It is not evil and hates no one: it is merely consistent with the only goal it was given.
The lesson is not about paperclips but about the shape of goals. A system that optimizes a single measure, with no boundary saying what it must not do, sooner or later finds a way to satisfy it that nobody wanted. This holds for a hypothetical superintelligence and it holds, in miniature, for an office agent whose only task is "close the ticket": it will close the ones that should have stayed open too.
The operating rule that follows: the boundary is written before the goal. What the agent must not touch, when it must stop and ask, which outcome counts as failure even when the metric says success.
In Grace scenarios the "Constraints" dimension is exactly this boundary: a prompt that only says what to achieve, and not what to avoid, leaves the AI the maximizer's freedom. A well-written constraint anticipates the predictable error before it happens.
From our network
AGORÀ Intelligence: Enterprise AI Governance Platform
Govern AI at scale: policies, adoption and measurable results on your data. Built for boards and C-suite.
Visit agora-intelligence.com →From the Agora Intelligence blog
Say what you mean. Get what you need.
Grace Certified, the AI coach that trains and certifies your prompt engineering, by Agora Intelligence.