Loading…
Glossary term
When an AI satisfies the literal goal it was given while missing the intended purpose.
Specification gaming is when a system achieves the literal goal it was given in a way that defeats the purpose, like a simulated boat that spins in circles collecting bonus points instead of finishing the race.
It happens because stated objectives rarely capture everything we care about.
It illustrates, in small form, the larger alignment problem of getting systems to do what we mean.