Skip to main content

Resultados relacionados

Resultados totales (Incluyendo duplicados): 5
Encontrada(s) 1 página(s)

RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Artículo científico (JournalArticle)

IQ tests are not for machines, yet

  • Dowe, David L.
  • Hernández-Orallo, José|||0000-0001-9746-7632
[EN] Complex, but specific, tasks¿such as chess or Jeopardy!¿are popularly seen as milestones for artificial intelligence (AI). However, they are not appropriate for evaluating the intelligence of machines or measuring the progress in AI. Aware of this delusion, Detterman has recently raised a challenge prompting AI researchers to evaluate their artefacts against IQ tests. We agree that the philosophy behind (human) IQ tests is a much better approach to machine intelligence evaluation than these specific tasks, and also more practical and informative than the Turing test. However, we have first to recall some work on machine intelligence measurement which has shown that some IQ tests can be passed by relatively simple programs. This suggests that the challenge may not be so demanding and may just work as a sophisticated CAPTCHA, since some types of tests might be easier than others for the current state of AI. Second, we show that an alternative, formal derivation of intelligence tests for machines is possible, grounded in (algorithmic) information theory. In these tests, we have a proper mathematical definition of what is being measured. Third, we re-visit some research done in the past fifteen years for effectively measuring machine intelligence¿since some assumptions about the subjects and their distribution no longer hold., This work was supported by the MEC projects EXPLORAINGENIO TIN 2009-06078-E, CONSOLIDER-INGENIO 26706 and TIN 2010-21062-C02-02, and GVA project PROMETEO/2008/051.
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Capítulo o parte de un libro (BookPart)

Compression and intelligence: social environments and communication

  • Dowe, David L.
  • Das, Paramjit K,
  • Hernández-Orallo, José|||0000-0001-9746-7632
Compression has been advocated as one of the principles which pervades inductive inference and prediction - and, from there, it has also been recurrent in definitions and tests of intelligence. However, this connection is less explicit in new approaches to intelligence. In this paper, we advocate that the notion of compression can appear again in definitions and tests of intelligence through the concepts of `mind-reading¿ and `communication¿ in the context of multi-agent systems and social environments. Our main position is that two-part Minimum Message Length (MML) compression is not only more natural and effective for agents with limited resources, but it is also much more appropriate for agents in (co-operative) social environments than one-part compression schemes - particularly those using a posterior-weighted mixture of all available models following Solomonoff¿s theory of prediction. We think that the realisation of these differences is important to avoid a naive view of `intelligence as compression¿ in favour of a better understanding of how, why and where (one-part or two-part, lossless or lossy) compression is needed., We thank the anonymous reviewers for their helpful comments, and we thank Kurt Kleiner for some challenging and ultimately very helpful questions in the broad area of this work. We also acknowledge the funding from the Spanish MEC and MICINN for projects TIN2009-06078-E/TIN, Consolider-Ingenio CSD2007-00022 and TIN2010-21062-C02, and Generalitat Valenciana for Prometeo/2008/051.
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Capítulo o parte de un libro (BookPart)

On more realistic environment distributions for defining, evaluating and developing intelligence

  • Hernández-Orallo, José|||0000-0001-9746-7632
  • Dowe, David L.
  • España Cubillo, Sergio
  • Hernández-Lloreda, M. Victoria
  • Insa Cabrera, Javier
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, 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 amuch too general distribution, since, e.g., the probability of other agents appearing on the scene or having some social interaction is almost 0 for many 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., We thank the anonymous reviewers for their helpful comments. We also thank the funding from the Spanish MEC and MICINN for projects TIN2009-06078-E/TIN, Consolider-Ingenio CSD2007-00022 and TIN2010-21062- C02, for MEC FPU grant AP2006-02323, and Generalitat Valenciana for Prometeo/2008/051
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Capítulo o parte de un libro (BookPart)

Comparing humans and AI agents

  • Insa Cabrera, Javier
  • Dowe, David L.
  • España Cubillo, Sergio
  • Henánez-Lloreda, M. Victoria
  • Hernández-Orallo, José|||0000-0001-9746-7632
Comparing humans and machines is one important source of information about both machine and human strengths and limitations. Most of these comparisons and competitions are performed in rather specific tasks such as calculus, speech recognition, translation, games, etc. The information conveyed by these experiments is limited, since it portrays that machines are much better than humans at some domains and worse at others. In fact, CAPTCHAs exploit this fact. However, there have only been a few proposals of general intelligence tests in the last two decades, and, to our knowledge, just a couple of implementations and evaluations. In this paper, we implement one of the most recent test proposals, devise an interface for humans and use it to compare the intelligence of humans and Q-learning, a popular reinforcement learning algorithm. The results are highly informative in many ways, raising many questions on the use of a (universal) distribution of environments, on the role of measuring knowledge acquisition, and other issues, such as speed, duration of the test, scalability, etc., We thank the anonymous reviewers for their helpful comments. We also thank José Antonio Martín H. for helping us with several issues about the RL competition, RL-Glue and reinforcement learning in general. We are also grateful to all the subjects who took the test. We also thank the funding from the Spanish MEC and MICINN for projects TIN2009-06078- E/TIN, Consolider-Ingenio CSD2007-00022 and TIN2010-21062-C02, for MEC FPU grant AP2006-02323, and Generalitat Valenciana for Prometeo/2008/051
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Capítulo o parte de un libro (BookPart)

Evaluating a reinforcement learning algorithm with a general intelligence test

  • Insa Cabrera, Javier
  • Dowe, David L.
  • Hernández-Orallo, José|||0000-0001-9746-7632
In this paper we apply the recent notion of anytime universal intelligence tests to the evaluation of a popular reinforcement learning algorithm, Q-learning. We show that a general approach to intelligence evaluation of AI algorithms is feasible. This top-down (theory-derived) approach is based on a generation of environments under a Solomonoff universal distribution instead of using a pre-defined set of specific tasks, such as mazes, problem repositories, etc. This first application of a general intelligence test to a reinforcement learning algorithm brings us to the issue of task-specific vs. general AI agents. This, in turn, suggests new avenues for AI agent evaluation and AI competitions, and also conveys some further insights about the performance of specific algorithms. © 2011 Springer-Verlag., We are grateful for the funding from the Spanish MEC and MICINN for projects TIN2009-06078-E/TIN, Consolider-Ingenio CSD2007-00022 and TIN2010-21062-C02, for MEC FPU grant AP2006-02323, and Generalitat Valenciana for Prometeo/2008/051.