An Approximation of the Universal Intelligence Measure

By Shane Legg, Joel Veness. 2011

71 citations recorded by the source corpus.

The Universal Intelligence Measure is a recently proposed formal definition of intelligence. It is mathematically specified, extremely general, and captures the essence of many informal definitions of intelligence. It is based on Hutter’s Universal Artificial Intelligence theory, an extension of Ray Solomonoff’s pioneering work on universal induction. Since the Universal Intelligence Measure is only asymptotically computable, building a practical intelligence test from it is not straightforward. This paper studies the practical issues involved in developing a real-world UIM-based performance metric. Based on our investigation, we develop a prototype implementation which we use to evaluate a number of different artificial agents.

Subfields: Artificial Intelligence, Computational Theory and Mathematics, Philosophy

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computational Theory and Mathematics

In Philosophy

Citations in the corpus

The Arcade Learning Environment: An Evaluation Platform for General Agents (2012, methodology): “More recently, the approaches of Hernández-Orallo and Dowe (2010) and of Legg and Veness (2011) appear to have some potential.”
Re-evaluating evaluation (2018, background): “Legg and Hutter developed a deﬁnition of intelligence which, informally, states “intelligence measures an agent’s ability to achieve goals in a wide range of environments” [26,27].”
Artificial General Intelligence: Concept, State of the Art, and Future Prospects (2009, background): “10 While the original form of Legg and Hutter’s definition of intelligence is impractical to compute, a more tractable approximation has recently been developed (Legg and Veness, 2013).”
A Neuroevolution Approach to General Atari Game Playing (2014, background): “Specifics of ANN topology are discussed in Section VI.”
Bad Universal Priors and Notions of Optimality (2015, background): “ Υξ ≤ Υξ(π) ≤ Υξ < 1. (5) See Figure 1. It is natural to ﬁx the policy random that takes actions uniformly at random to have an intelligence score of 1/2 by choosing a ‘symmetric’ universal prior (Legg and Veness, 2013). AIXI is not computable (Leike and Hutter, 2015, Thm. 14), hence there is no ”
Evaluation in artificial intelligence: from task-oriented to ability-oriented measurement (2017, methodology): “Taking into account these solutions, some actual tests have been developed (Insa-Cabrera et al, 2011b,a; Legg and Veness, 2013).”