Subjective Randomness and Natural Scene Statistics

Ethan Schreiber, Thomas L. GriffithseScholarship 2007

Abstract

Subjective Randomness and Natural Scene Statistics Ethan Schreiber ([email protected]) Department of Computer Science, 115 Waterman St Providence, RI 02912 USA Thomas L. Griffiths (tom [email protected]) Department of Psychology, 3210 Tolman Hall #1650 Berkeley, CA 94720 USA Abstract 0010101101) more random than arrays in which cells take the same values as their neighbors (such as 0000011111) (Falk & Konold, 1997). These results make it clear that people are sensitive to certain regularities, such as cells having the same values as their neighbors. However, it is difficult to explain why these regularities should be more important than others that seem a priori plausible, such as neighboring cells differ- ing in their values. In this paper, we explore a possible explanation for the ori- gins of the regularities that influence subjective randomness judgments for one class of stimuli: two-dimensional binary arrays. These stimuli are essentially images, consisting of a grid of black and white pixels (see Figure 1), and we might thus expect that the kinds of regularities detected by the visual system should play an important role in determining their per- ceived randomness. A great deal of recent research suggests that the human visual cortex efficiently codes for the structure of natural scenes (Olshausen & Field, 2000; Simoncelli & Ol- shausen, 2001). We consider the possibility that the kinds of regularities that people detect in two-dimensional binary ar- rays are those that are characteristic of natural scenes. We show that the statistics of images of natural scenes provide an explanation for why arrays in which neighboring cells differ in value should be considered more random, and describe an experiment illustrating that a Bayesian model of subjective randomness that evaluates whether a particular array is more likely to be generated by chance or be part of a natural scene provides good predictions of human randomness judgments. The plan of the paper is as follows. The next section presents a formal definition of subjective randomness from the perspective of Bayesian statistics. This establishes the need to define a distribution over stimuli based on a “regular” process, which we approach by introducing a simple statis- tical model for images. We then describe how the parame- ters of this model can be estimated from natural scenes and outline the predictions that result from using this model as part of the Bayesian account of subjective randomness, before going on to present an experiment testing those predictions. We conclude by discussing the implications and limitations of this experiment. We examine factors influencing whether people perceive two- dimensional binary arrays as random. Accounts of subjective randomness often appeal to the idea that people consider stim- uli random when they cannot detect any regularities charac- terizing the structure of those stimuli. We explore the pos- sibility that the kinds of regularities people can detect are shaped by the statistics of their natural environment. For two- dimensional binary arrays – grids of black and white pixels – we can measure these statistics directly from images of natu- ral scenes. By estimating the parameters of a Markov random field from these images, we obtain a potential explanation for why people find arrays in which cells differ in value from their neighbors more random, with the reverse being more char- acteristic of structured images. We present an experiment in which we find a correlation between the extent to which peo- ple consider an array random, and the evidence it provides for having been generated by chance rather than nature. Keywords: subjective randomness; Markov random fields; Boltzmann machine; Gibbs sampling; natural scenes Introduction People are very sensitive to deviations from their expectations about randomness. For example, the game Yahtzee involves repeatedly rolling five six-sided dice. If you were to roll all sixes six times in a row, you would probably be quite sur- prised. The probability of such a sequence arising by chance is 1/6 30 . However, the low probability of such an event is not sufficient to explain its apparent non-randomness, as any other sequence of the same number of dice rolls has the same probability. Consequently, recent accounts of human subjec- tive randomness – our sense of the extent to which an event seems random – have focused on the regularities in an event that suggest a process other than chance might be at work (Griffiths & Tenenbaum, 2003, 2004; Falk & Konold, 1997). The basic idea behind these accounts is that stimuli will ap- pear random when they do not express any regularities. An important challenge for any account of subjective ran- domness based on the presence of regularities is explaining why people should be sensitive to a particular set of regulari- ties. In the example given above, we might be able to appeal to the fact that systematic runs of the same number suggest loaded dice, or some other non-random process influencing the outcomes. However, for other kinds of stimuli, such as the one- or two-dimensional binary arrays used in many sub- jective randomness experiments, explanations are more diffi- cult to come by. A common finding in these experiments is that people consider arrays in which cells take different val- ues from their neighbors (such as the one-dimensional array Subjective randomness as Bayesian inference One explanation for human randomness judgments is to view them as the result of an inference as to whether an observed stimulus, X, was generated by chance, or by some other more
BibTeX
@inproceedings{schreiber2007subjective,
  author = {Ethan Schreiber and Thomas L. Griffiths},
  title = {{Subjective Randomness and Natural Scene Statistics}},
  booktitle = {eScholarship},
  year = {2007},
}

Loading related content...

Figure