Bias and No Free Lunch in Formal Measures of Intelligence

Hibbard, Bill

doi:10.2478/v10229-011-0004-6

Cited by 27 publications

(35 citation statements)

References 5 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Note that a meaningful choice of each environment is much more effective than choosing them arbitrarily. This can replace or complement those evaluations based on repositories and it is certainly more effective than choosing all the environments or selecting them by a (universal) distribution, as in [46] (including some improvements over the distribution, such as the one introduced by [39]). …”

Section: Discussionmentioning

confidence: 99%

“…Although not explicitly used as an estimation of difficulty (but rather as a way to derive a distribution), this approach becomes much more cumbersome [39] than it might seem, because the relation between complexity and difficulty is intricate. In [34], another variant of Kolmogorov complexity (Kt max ) was suggested as a measure of complexity for the environments, but it was already stated that this approximation was unidirectional: some very complex environments might be easy, i.e., high rewards could be obtained by very simple policies ( [34, sec.…”

Section: Looking At the Problemmentioning

confidence: 99%

See 1 more Smart Citation

On environment difficulty and discriminating power

Hernández-Orallo

2014

Auton Agent Multi-Agent Syst

View full text Add to dashboard Cite

This paper presents a way to estimate the difficulty and discriminating power of any task instance. We focus on a very general setting for tasks: interactive (possibly multi-agent) environments where an agent acts upon observations and rewards. Instead of analysing the complexity of the environment, the state space or the actions that are performed by the agent, we analyse the performance of a population of agent policies against the task, leading to a distribution that is examined in terms of policy complexity. This distribution is then sliced by the algorithmic complexity of the policy and analysed through several diagrams and indicators. The notion of environment response curve is also introduced, by inverting the performance results into an ability scale. We apply all these concepts, diagrams and indicators to two illustrative problems: a class of agent-populated elementary cellular automata, showing how the difficulty and discriminating power may vary for several environments, and a multi-agent system, where agents can become predators or preys, and may need to coordinate. Finally, we discuss how these tools can be applied to characterise (interactive) tasks and (multi-agent) environments. These characterisations can then be used to get more insight about agent performance and to facilitate the development of adaptive tests for the evaluation of agent abilities.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Looking At the Problemmentioning

confidence: 99%

On environment difficulty and discriminating power

Hernández-Orallo

2014

Auton Agent Multi-Agent Syst

View full text Add to dashboard Cite

show abstract

“…These weights are such that an agent's intelligence lies between zero and one. The choice of PUTM must be constrained to avoid bias in this measure [3]. There are at least two ways in which this measure is inconsistent with our intuitions about measuring human intelligence:…”

Section: Introductionmentioning

confidence: 99%

“…It sums rewards from the first time step, with no time to learn. AIXI always makes optimal actions [4] (as long as it is defined using the same universal Turing machine used to define the measure [3]), but AIXI is not computable. We allow humans time to learn before judging their intelligence.…”

Section: Introductionmentioning

confidence: 99%

Measuring Agent Intelligence via Hierarchies of Environments

Hibbard

2011

Artificial General Intelligence

Self Cite

View full text Add to dashboard Cite

Abstract. Under Legg's and Hutter's formal measure [1], performance in easy environments counts more toward an agent's intelligence than does performance in difficult environments. An alternate measure of intelligence is proposed based on a hierarchy of sets of increasingly difficult environments, in a reinforcement learning framework. An agent's intelligence is measured as the ordinal of the most difficult set of environments it can pass. This measure is defined in both Turing machine and finite state machine models of computing. In the finite model the measure includes the number of time steps required to pass the test.

show abstract

“…Thus, aggregating results using this distribution would assign higher intelligence to simple programs. One possible solution is to set a minimum complexity value [13], but this is clearly an arbitrary choice. An alternative option would be to define intelligence as the maximum complexity level where a system can score 'significantly better' than random.…”

Section: Introductionmentioning

confidence: 99%

On More Realistic Environment Distributions for Defining, Evaluating and Developing Intelligence

Hernández-Orallo

Dowe

España

et al. 2011

Artificial General Intelligence

View full text Add to dashboard Cite

Abstract. One insightful view of the notion of intelligence is the ability to perform well in a diverse set of tasks, problems or environments. One of the key issues is therefore the choice of this set and the probability of each individual, which can be formalised as a 'distribution'. Formalising and properly defining this distribution is an important challenge to understand what intelligence is and to achieve artificial general intelligence (AGI). In this paper, we agree with previous criticisms that a universal distribution using a reference universal Turing machine (UTM) over tasks, environments, etc., is perhaps much too general, since, e.g., the probability of other agents appearing on the scene or having some social interaction is almost 0 for most reference UTMs. Instead, we propose the notion of Darwin-Wallace distribution for environments, which is inspired by biological evolution, artificial life and evolutionary computation. However, although enlightening about where and how intelligence should excel, this distribution has so many options and is uncomputable in so many ways that we certainly need a more practical alternative. We propose the use of intelligence tests over multi-agent systems, in such a way that agents with a certified level of intelligence at a certain degree are used to construct the tests for the next degree. This constructive methodology can then be used as a more realistic intelligence test and also as a testbed for developing and evaluating AGI systems.

show abstract

Bias and No Free Lunch in Formal Measures of Intelligence

Cited by 27 publications

References 5 publications

On environment difficulty and discriminating power

On environment difficulty and discriminating power

Measuring Agent Intelligence via Hierarchies of Environments

On More Realistic Environment Distributions for Defining, Evaluating and Developing Intelligence

Contact Info

Product

Resources

About