AIXI model

E774590

idealized reinforcement learning agent theoretical model of artificial intelligence uncomputable agent model universal artificial intelligence model

The AIXI model is a theoretical framework for an idealized, maximally intelligent reinforcement learning agent that combines Solomonoff induction with sequential decision theory.

Jump to: Surface forms Statements Referenced by

Observed surface forms (3)

Surface form	Occurrences
AIXI	2
AIXI framework	1
AIXI-tl	1

Statements (50)

Predicate	Object
instanceOf	idealized reinforcement learning agent ⓘ theoretical model of artificial intelligence ⓘ uncomputable agent model ⓘ universal artificial intelligence model ⓘ
assumes	unknown but computable environment ⓘ
basedOn	Bayesian decision theory NERFINISHED ⓘ Solomonoff induction NERFINISHED ⓘ sequential decision theory ⓘ
computabilityStatus	uncomputable ⓘ
decisionCriterion	Bayes-optimality with respect to universal prior ⓘ
definedBy	Marcus Hutter NERFINISHED ⓘ
environmentClass	all lower semicomputable semimeasures ⓘ
formalizedIn	universal artificial intelligence framework ⓘ
formalizes	Legg-Hutter intelligence measure NERFINISHED ⓘ
goal	maximization of cumulative reward ⓘ
hasApproximation	AIXItl NERFINISHED ⓘ MC-AIXI-CTW NERFINISHED ⓘ
hasLimitation	depends on choice of reference universal Turing machine ⓘ not directly implementable on real hardware ⓘ
hasProperty	defined for general reinforcement learning environments ⓘ environment-agnostic ⓘ maximally intelligent in its formal setting ⓘ model-free in the usual RL sense ⓘ not computable in practice ⓘ theoretically optimal ⓘ uses a universal Turing machine prior ⓘ uses algorithmic probability ⓘ
influenced	formal definitions of intelligence ⓘ theoretical research in AGI ⓘ
interactionPattern	agent-environment loop ⓘ
introducedInWork	Universal Artificial Intelligence: Sequential Decisions based on Algorithmic Probability NERFINISHED ⓘ
optimizes	expected discounted reward ⓘ
outputs	actions to environment ⓘ
publicationYear	2005 ⓘ
receives	observations from environment ⓘ rewards from environment ⓘ
relatedTo	Kolmogorov complexity NERFINISHED ⓘ Solomonoff universal distribution NERFINISHED ⓘ algorithmic information theory ⓘ partially observable Markov decision processes ⓘ reinforcement learning ⓘ sequential decision processes ⓘ
requires	enumeration of all programs on a universal Turing machine ⓘ infinite computational resources ⓘ
selects	actions to maximize expected future reward ⓘ
timeStructure	discrete time interaction cycles ⓘ
upperBoundFor	intelligence of computable agents in its setting ⓘ
uses	Solomonoff universal prior NERFINISHED ⓘ expectimax planning over all computable environments ⓘ mixture over all semicomputable probability measures ⓘ

Referenced by (6)

Full triples — surface form annotated when it differs from this entity's canonical label.

Marcus Hutter → developed → AIXI model ⓘ

this entity surface form: AIXI

Marcus Hutter → hasConcept → AIXI model ⓘ

this entity surface form: AIXI-tl

Ray Solomonoff → inspired → AIXI model ⓘ

this entity surface form: AIXI framework

Marcus Hutter → knownFor → AIXI model ⓘ

Universal Intelligence: A Definition of Machine Intelligence → relatedTo → AIXI model ⓘ

universal intelligence measure → relatedTo → AIXI model ⓘ

this entity surface form: AIXI