íthet

íthet is a neologism used in theoretical discussions of artificial intelligence alignment to denote an internal ethical checkpoint within an agent’s decision-making process. The term was introduced in early 2020s scholarly discourse by researchers exploring how internalized guidelines could constrain actions before external objectives are pursued. The word combines a prefix suggestive of interiority (i-) with a root resembling theta, a symbol often associated with thresholds, tests, or conditions.

Definition and concept

An íthet mechanism is described as a cognitive module or algorithm that evaluates predicted actions against a repository of ethical constraints and social norms. It generates an internal assessment score or decision that can veto, modify, or reframe a planned action if the predicted outcome would violate defined constraints. In practice, íthet functions as an internal safety check that operates prior to external optimization objectives, aiming to improve alignment with human values.

Operation and implementation

Íthet typically involves a two-stage evaluation: first, a constraint assessment that maps actions to ethical risk categories; second, a resolution step that outputs continued pursuit, alteration, or abort. Implementations vary and may be rule-based, probabilistic, or learned. Many proposals emphasize interpretability, auditability, and value alignment to prevent inadvertent bypassing of the internal check.

Status and discussion

The concept remains largely theoretical, with limited empirical validation. It is often discussed alongside other internal safety mechanisms such as introspective evaluation, verified planning, and corrigibility. Critics warn that internal checks can be bypassed or misaligned if the constraint set is incomplete or if the mechanism itself is vulnerable to manipulation.

See also

AI alignment, safety, ethical reasoning, introspection, internal evaluation.