Judgments of Time, Aesthetics, and Complexity as a Function of The
Fractal Dimension of Images formed by Chaotic Attractors
1Depatment of Psychology, Silliman University, Dumaguete City, Philippines 6200. and Blueberry Brain Institute, Waterbury Center VT USA 05677
2Correspondence should be sent to: firstname.lastname@example.org
3Department of Physics, University of Wisconsin, Madison, WI 53706
4Department of Psychology, Moscow State University, Moscow, Russia
In this experiment, we obtained judgments of the duration of presentation of 3D images of chaotic attractors, and judgments of their aesthetic value and complexity as a function of their fractal dimension (D2). We used four levels of fractal dimension (four stimuli at each level, mean D2s = .59, 1.07, 1.54, 2.27), with small samples of six students from each of three populations (elementary school students, graduate students, and special education students from ethnic minority groups in residence on our campus). In addition to replicating earlier studies of aesthetic judgments, we additionally asked for judgments of complexity to see if they were also a nonlinear function of the fractal dimensionality of stimuli (they were, increasing to a maximum at D2 = 1.54, and falling off at D2 = 2.27), suggesting the possibility that aesthetics and complexity judgments were at least in part mediated by perceived complexity. Perhaps this non-monotonic result was due to a loss of contrast and detail within the attractors at the highest dimensional complexity. Judgments of duration did not yield a significant F-ratio between groups, but t-tests showed the lowest dimensionality yielded shorter time estimates than those for the higher dimensionalities. If complexity is a determinant of subjective duration of these stimuli, it is saturated at fairly low levels of the fractal dimensionality of the stimuli. Three-way ANOVAs within participants, showed only the fractal dimension as a significant source of variation in aesthetic and complexity judgments. Academic level, gender, or cultural differences were not significant sources of variation in judgments, although there were some interesting individual and cultural findings.
KEY WORDS: psychophysics, time, complexity, aesthetics, chaos, fractal dimension
1.1 Narrative Background
It was originally my intent to study the perception of time as a function of the fractal nature of stimuli, using auditory or musical stimuli, when Elliot Middleton informed me of a program by Julien Clinton Sprott that generated visual strange attractors, which I immediately downloaded from Sprott’s website, and was immersed in a world of beautiful images. We realized that by switching to these visual stimuli we could have a set of images whose fractal dimension was already computed by his program. I had been using his earlier book (Sprott, 1993) with the newly established Chaos Society of Silliman University, and so was familiar with his approach to dynamics, and with his research on aesthetics as a function of fractal dimension (Aks & Sprott, 1996). I later visited him while he adapted his programs to more general psychophysical explorations. While mainly interested in time perception, I had decided to use aesthetic judgments as well, mainly to attenuate the importance on time for my participants. Then I had the idea to add complexity judgments, thinking at the time that there were no precedents for that, which was true for such images, but in fact, I learned later, has a large history in psychophysical research. At any rate, he adapted his programs for me to be flexible to meet the changing needs of future experimental designs. When some undergraduate students at Silliman asked me for an idea for a research project using chaos for an experimental psychology course they were taking, the opportunity for them to do this exploratory study presented itself. About the same time, while giving lectures at a conference in Moscow in 2000 (Sulis & Tromifova, 2001), I shared Sprott’s computer program with Olga Mitina who subsequently did a nice study with it adding the investigation of personality factors (Mitina & Abraham, 2003). Thus began our collaboration.
1.2 Research Background
Nonlinear dynamical systems have recently been explored in cognitive and perceptual systems by the foundational programs of Freeman, (2006; Skarda & Freeman, 1987), Gregson (1996, 2006), Heath, (2000), Kelso and Engstrøm (2006), Turvey (2005), and Ward (2001), among many others. For a brief earlier history, see Abraham (1997a).
Perceptual/neural organizational features of attending complex stimuli may affect both the estimation of complexity and aesthetics, but also, there is the possibility that they could also affect the perception of time. There is likely a nonlinear interaction between the complexity of stimuli and experiencing time and aesthetics. Studies of both the perception of time and of aesthetics have centered on cognitive and biological factors (Anderson & Mandell, 1996; Eisler, A., 2003; Eisler, H.; Eisler, Eisler, & Montgomery, 1996). Many factors of stimuli, cognition, and biology have been investigated. One of the features of stimuli which make both biological and cognitive demands for time estimation is the amount of perceptual/cognitive effort involved, such as can vary with complexity of stimuli (Cupchik & Gebotys, 1988; Stoyanova & Yakimoff et al., 1987). The same could likely be said for aesthetic judgments (Aks & Sprott, 1996, 2003; Mureika, Cupchik, & Dyer, 2004; Sprott, 1993, 2003; Taylor, Spehar, Wise, et al., 2005).
To investigate the possibility of a relationship between complexity and judgments of time and aesthetics it was decided to use abstract stimuli. Chaotic attractors were used for this purpose as they are easy to generate along with objective measurements of their complexity (Aks & Sprott, 1996; Sprott, 1993a,b, 2003). Also, they are relatively free of prior associations. Chaotic attractors as stimuli can be created by integration of three coupled differential equations that produce abstract computer images in two spatial dimensions with the third dimension being represented by color coding, which adds to their aesthetic potential. To check on the assumption that mathematical complexity has a relationship to perceived complexity, it was decided to add an estimate of complexity to the ratings obtained from the participants.
While complexity of stimuli was the principal independent variable, demographic factors as well, namely age and culture (Eisler, Eisler, & Montgomery, 1994; H. Eisler, 1996) were also varied. Therefore, both children and adults, and both urbanized students of Silliman University and rural cultural minority visitors (Ati and Sulod) to a specialized education program to our campus served as participants.
And to enrich the dependent variable side of the experiment, in addition to scales of aesthetics and complexity, and estimates of stimulus duration, some narrative explorations of the features upon which the participants might have been basing their judgments were also attempted.
(1) Six students (1 female, 5 males, from grades 4-5) at Silliman University Elementary School were recruited from a special computer learning project under the direction of the senior author. Their participation in this experiment had direct relevance to their goals of the integration of art, science, and mathematics.
(2) Six graduate Students (4 females, 2 males), Department of Psychology, Silliman University were recruited from a graduate seminar.
(3) Six Ati and Sulod adults (4 females, 2 males) were recruited from a residential special learning program for the current school year hosted by the University’s Department of Education. They were visiting from several Ati and Sulod communities (Negros Oriental Province, where the University is also located).
In accordance with standard practice for undergraduate research, the locus of informed consent was formally placed in the hands of administrators of the educational units involved. Therefore permission to solicit participants was obtained from Ms. Iyoyo, Principal of the Silliman University Elementary School, Dr. Margaret Udarbe-Alvarez, Chair of the Psychology Department, and Dr. Betsy Joy Tan, Dean of the School of Education and administrator for the Ati Educational Project. Individual participants participated by verbal agreement, but that contained the usual elements of informed consent.
The elementary school students were not randomly conscripted from the school, and could be considered as among the high achieving students, being the whole of the members of the experimental enrichment program. The same could be said of the graduate students, who represent a high achieving profile. This achieving profile could work against our finding an age difference, due to the sophistication of both populations. The Ati, also are not a random sample of Ati, being leaders in their communities interested in bringing their Silliman education back to their communities, and from among those, the more adventuresome were among those who volunteered to participate. While these sophistications could have minimized differences in the results, their differences in age, education, life styles, and environment could possibly have influenced the way they responded to the abstract visual stimuli.
Sprott (1993) developed a program for the generation of computer images of chaotic attractors from systems of three nonlinear coupled difference equations (quadratic maps) which were adapted for use in this and similar experiments (Mitina and Abraham, 2003). The equations are:
xn+1 = a + bxn + cxn2 + dxnyn +exnzn +fyn + gyn2 + hynzn + izn + jzn2
yn+1 = xn
zn+1 = yn where x,y,z are variables and a–j are parameters.
Solutions (trajectories) are displayed on computer monitors in two dimensions (x,y, for the horizontal and vertical axes of the plane of the screen) with the third dimension (z) represented by color coding. These stimuli vary in complexity depending on the parameters used in each integration, and are selected by the program at each integration. These programs develop the trajectory (attractors, plotted by points, not as a continuous line) by iteration of the equations, although they evolve so fast as to be essentially perceived as filling space rapidly over time, rather than as a trajectory growing over time. Thus of the measures of complexity produced for each image by the program, the fractal dimension of the attractor (which measures complexity or density in filling the space) is a more realistic measure of the complexity of the trajectory than the largest Liapunov exponent (which measures aspects of the evolution of the trajectory in time, divergence from nearby starting points). The program was used to generate 168 of these stimuli. A statistics program (STATISTICA) was used to analyze these two measures of the images. Aks and Sprott (1986) used both measures and found aesthetic responses to vary with both measures for some participants. However we chose our stimuli on the basis of having the Liapunov exponents well correlated with the fractal dimension, so it would be irrelevant to try to separate the two aspects of complexity. The fractal dimension, F, employed here is the correlation dimension, usually designated as D2 (introduced to dynamics by Grassberger & Procaccia, 1983; and reviewed in many books and articles, including Sprott, 2003, p. 307-311, and Abraham, 1997b, pp. 17-18; the fast algorithm of Sprott, ibid p. 317, was used for the computations here). Stimuli were chosen with D2s in four ranges, .5–.85, .86–1.4–1.6, and 2.2–2.4 (mean D2s, 0.59, 1.07, 1.54, 2.27). Twenty stimuli, five from each of these four ranges were selected. Four were to be used at the beginning of each run in a fixed order to give a warm-up for the participant, and the remaining 16 were presented in a randomized sequence for the remainder of the run, as described below. These 16 stimuli were used in analyzing the results. These images can be viewed at http://www.blueberry-brain.org/silliman/jemstim.htm
2.3 Computer Implementation and Instructions
A session on the computer started with a set-up phase by the experimenter in which the instruction set and presentation method were selected, and the participant’s identification and data file name were entered. Next, the participant sat at the computer, and the experimenter gave the participant verbal instructions on how the experiment was to be run. Then the stimuli were presented in succession, going through the sequence twice. The following comprised the two runs through the stimuli, the first run to obtain the participant’s ratings of aesthetics and complexity, and the second run to obtain estimation of the duration of presentation of the images. The first run was completed as follows:
Once the experimenter finished setting up the computer to run the experiment, the participant pressed any key on the computer keyboard to begin. Then the first stimulus appeared on the screen. For 1.04 seconds, the stimulus remained on, and then the first question appeared at the top of the screen:
How beautiful was the image [1 (least) — 9 (most)]?
Then the experimenter explained the rating scale with the first stimulus.
The participant responded on the keyboard, with a number and followed with an entering key-press.
The second question then replaced the first question:
How complex was the image [1 (least) - 9 (most)]?
To insure that the participant used a wide range of ratings of complexity, the experimenter provided guidance on numerical values for this and the following three warm-up stimuli.
After that, the participants self-paced themselves for the remainder of the run through the remaining 16 of the 20 images.
The experimenter then reset the computer for the second run that was used to obtain the ratings of perceived duration (time). This second run was completed as follows:
For the estimation of duration of presentation of the images, the same images were used, with the 5th-20th being in a new random order. The participant initiated the run with a key press. Each stimulus was presented for 10.05 seconds, turned off, and then the following question was presented on-screen:
How many seconds was the image on?
(The participant had to hide the face of a watch if worn, and asked not to count.)
Finally, after the runs were completed, the participants were asked to describe which features of the images determined their aesthetic and complexity judgments.
Each run took about 5-8 minutes, the whole session took 20-30 minutes. Some participants were run singly, and some were run simultaneously, using up to the 6 computers in the psychology computer laboratory where the experiment was run. When six participants were run simultaneously, it was found that each of the three experimenters could supervise two participants at a time without difficulty.
The computer program used to run the experiment recorded the participant’s responses automatically. These were then transferred to an Excel spreadsheet program as follows. The responses to the test stimuli for each of the three measures at each of the stimulus complexity levels were averaged for each participant. Also the Pearson product moment correlations were computed for all six pair-wise combinations of the three response measures and the fractal dimension for each participant. Thus a case-by-variables data table was assembled with 24 variables per case: group, gender, the fractal dimension (D2) and the judgments of aesthetics, complexity, and duration at each of the four complexity levels, and the six correlation coefficients.
This table was imported by the STATISTICA data analysis program, where descriptive statistics and tables were completed (box & whisker plots, Figures 1, 2, & 3,). Inferential statistics were also computed, namely 3-way ANOVAs for a within-participant design for each of the three measures (with the complexity level being the repeated variable, and group and gender being the independent between-group variables. T-tests for paired measurements were also used to determine the differences for each of the main measures (complexity, aesthetic, and duration judgments) between each pair of adjacent levels of D2 (low vs low mid, low-mid vs high mid, and high-mid vs high). An alpha level of .05 was used for all statistical tests.
3.1 Judgments of Complexity
The box & whisker plots (Figure 1) show the means, standard error of the mean, and the 95% confidence limits of the complexity judgments. The y-axis represents the means of the judgments of the complexity for each level of complexity of the images. The x-axis represents the four levels of the complexity of the images. The levels of that variable, going from left to right are: low (mean D2 = .59, range: .52–.65), low-mid (mean D2 = 1.07, range:1.05–1.09), high-mid (mean D2 = 1.54, range: 1.46–1.55), and high (mean D2 = 2.27, range: 2.21–2.36).
This figure shows judgments of complexity to be a non-monotonic increasing function for most of the range of D2 as might be expected, but decreasing at the highest values. That is increasing complexity is perceived as such, until the images at the highest D2 values which are reported as less complex. Even within the increasing portion of the curve, the result is decidedly non-Fechnerian: increases in D2 are not met with proportionately increased magnitude of the judgments of complexity. The ANOVA showed significant results for D2, but not for gender or group, nor for any interactions. The t-tests showed all adjacent means to differ significantly from each other. All probabilities were way below the alpha level. Tables can be found at http://blueberry-brain.org/silliman/jemanovacmplx.htm.
3.2 Judgments of Aesthetics
The box & whisker plots (Figure 2) show similar results for the aesthetics judgments. This plot also shows a non-monotonic form nearly identical to that for the complexity judgments, with maximum aesthetics judgments also occurring at the high-mid level of the fractal dimension of the images. That means that for the three lower levels of the fractal dimension, the participants scaled the aesthetics as increasing with increased complexity, but for the highest level of mathematical complexity (D2), they tended to judge the aesthetics as less, quite comparable to the low-middle level. Again, that result is non-Fechnerian even within the monotonic increasing phase of the curve.
The ANOVA results yielded results almost identical to those
for the complexity variable (a table is at http://blueberry-brain.org/silliman/jemanovaaesth.htm).
That is, the two between-group independent variables, group and gender, were
not significant, while the within-participant main independent variable (levels
I will try to find that tape. It will be a blast to see it again. Forgot that I even taped it.
fred2), was significant. The one difference compared to the results for the complexity judgments were that the 3-way interaction was also significant, a result difficult to interpret (see the figure at http://blueberry-brain.org/dynamics/jemint3xaesth.jpg ). The t-tests for adjacent values of D2 also were similar to those for the complexity judgments (see again the table at http://blueberry-brain.org/silliman/jemanovaaesth.htm).
3.3 Judgments of Duration
The results of the judgments of the duration of the presentation of stimuli were not significant. The box & whisker plot (Figure 3) shows (1) no simple relationship to the mathematical complexity (D2) of the images, (2) remarkable accuracy of the means of the time estimates to the actual time of 10 sec. despite (3) great variability in the responses (range: 2-16.25 sec). This accuracy is reflected in the means of the four levels, which were 9.31, 10.06, 9.69, and 10.31 sec. (left to right on the box & whiskers plot), with a grand mean of 9.84 sec.
The ANOVA (the table is at http://blueberry-brain.org/silliman/jemanovatime.htm) shows no significant main effects or interactions, which is not surprising considering the lack of trends and the great variability seen in the box & whisker plot. Despite that, the t-test for one of the comparisons; the ratings for the lo-mid complexity stimuli were significantly greater than the lo level stimuli of the mathematical complexity variable.
While the interaction terms were not significant, the plots of the means for the 3-way interaction (see table at http://blueberry-brain.org/silliman/jemint3xtime.jpg) showed consistently lower time estimates from females compared to males (except for the elementary students which included but one female), and for the Ati/Sulod group compared to the elementary and graduate students.
3.4 Correlations: Overall
Pearson product-moment correlations (see the Table) mirrored the results of the box & whisker plots and the ANOVAs. The mathematical complexity (fractal dimension, D2) was positively correlated to both the complexity ratings (r = .37) and the aesthetic ratings (r = .46), but not the time estimations (r = .03). The highest correlation was between the aesthetic and complexity ratings (r = .64). The remaining two correlations were remarkably unremarkable, the one between time estimation and complexity ratings (r = .07), and the one between time and aesthetics judgments (r = .05). Thus, the three correlations involving relationships among fractal dimension, complexity judgments, and aesthetic judgments were strong, while the three involving time estimates were essentially zero.
3.5 Correlations: Group Comparisons
The students from the elementary school were similar to the overall means of the correlations, with the exception that their correlation between aesthetic and complexity ratings were higher, and in fact were the highest among the groups (r = .84). The graduate students were similar to the overall means of the correlations except for being lower (lowest compared to the other groups) on the correlation between fractal dimension and the aesthetic ratings (r = .24). The students from the Ati and Sulod had the highest correlation compared to the other two groups for the relationship between the fractal dimension and aesthetic ratings (r = .67), and the lowest correlation compared to the other two groups for the relationship between the aesthetic and complexity ratings (r = .44).
Thus the highest correlations were those between the complexity and aesthetic judgments (r = .64), with the elementary students being the highest (r = .84), the Ati/Sulod-adult education students being lowest (r = .44), and the graduate students being in between, right at the mean (r = .63). Another interesting group difference was for the correlation of the fractal dimension to aesthetic ratings, which was highest for the Ati/Sulod students (r = .67), lowest for the graduate students (r = .24), and intermediate, right at the mean for the elementary students (r = .47).
3.6 Correlations: Individual Differences
Half the participants showed a pattern of higher correlations among the fractal dimension and aesthetic and complexity ratings and low correlations involving time judgments (participants 1, 4, 5, 6, 7, 8, 16, 17).
One participant (9), who was also a instructor in art at Silliman University, expressed a preference for visual art that had a lot of open space. However, his aesthetic ratings revealed a zero correlation (r = -.097) rather than a negative relationship, possibly due to the nonlinear relationship between area covered and D2. The zero correlations indicate a constant estimation of time, reflecting his accuracy on this factor. While his claim for a preference for art with open space is not born out, he appears to be a person with strong independent aesthetic convictions, which are certainly not related to mathematical complexity.
Two participants (2, 16) showed very strong correlations among D2 and aesthetic and complexity ratings and some fairly strong positive correlations involving the time estimates.
Another participant (17) also showed very high correlations of D2 to aesthetic and complexity ratings, but his correlation between the two ratings was somewhat lower (0.346) than the mean (0.638) showing independence of aesthetic and complexity judgments. This participant showed the unusual result of moderately strong negative correlation of time estimation to D2 and complexity ratings. A few other participants showed some correlations to time estimation, both positive and negative.
The individual correlations between complexity judgments and aesthetic judgments were the highest of all, with 12 of them being greater than .8, and one was a perfect 1. This is a key finding. Why is it not so for all individuals? Can there be different reasons for different individuals to have high correlations? Can one unravel some of the individual differences by examining the pattern of responses across all measures for particular individuals? For example, participant 15 had identical scores for the two ratings aesthetic and complexity to D2 (0.559). Likewise, the correlations of aesthetics and complexity to time estimates were also identical (.604) for participant 15. Thus correlations of aesthetic ratings were higher to complexity judgments than to actual mathematical complexity. These could reflect response bias or generalization, possibly due to a sense of uncertainty. Two other participants (3, 10) showed this response generalization reflected in high correlations between aesthetic and complexity ratings while the remaining five correlations were negligible, including no relationships to the fractal dimension. Participant (11) was similar to these two participants, except that the aesthetic to complexity correlation was smaller.
Of the remaining three correlations (all for the time judgments), a few were modestly stronger, mainly for participants 2, 7, 14, 15, 16, 17. Thus, there was greater individual variability in the Ati/Sulud group (4 participants compared to 1 each in the other two groups). Of these, two were more strongly negative (Participants 7 & 17), and four (Participants 2, 14, 15, & 16) strongly positive. Three were strongest for the correlations to the mathematical complexity (Participants 2, 7, 15), and three were strongest for the correlations to the complexity judgments (Participants 14, 16, & 17), all three of these in the Ati/Sulud group.
There were no large negative correlations of other variables to aesthetic ratings, the largest being r(A,T) = –0.434 for participant 13. The remaining correlations of aesthetics ratings to time estimates were negligible except for three of the Ati/Sulod group (participants 13, 15, 16) who had correlations of .-0.434, 0.604 and .434 respectively.
The only hint of cultural and gender differences that can be found in the data are swamped by individual differences, and the relationships to mathematical complexity are much stronger and therefore survive individual differences.
Since the judged complexity displayed the classic ∩-shaped function (also known as the inverted U-shaped function; Walker, 1971) of aesthetics to the mathematical complexity of the images as measured by D2, the question arises as to why that should be? Visual inspection of the images gives one possible answer. For stimuli where the mean D2 = .59, the space is quite empty; the attractors fill little of the available 3D Cartesian space for the image. When going to the next higher D2 (mean D2 = 1.07), the attractors fill the state space (the 2D computer screen) more completely, and the space within the attractor is also more completely filled. Going to the next higher (mean D2 = .154), there is a further increase in the filling of space and more visual boundaries and features within the attractor (the third, colored dimension helps a lot with that). At the highest level (mean D2 = 2.27), the attractors again occupy slightly more of the viewing space, but the most apparent difference is the increase in density of points within the attractor, filling the attractor more completely, but with less interior details. The attractors appear more cloud-like, more homogeneous, with less interior definition and boundaries, which could be the basis for their being judged less complex.
Why does the measure, D2 not see all that the eye sees? D2 sees mainly the filling of space and does not enjoy the advantages of the eye, brain, and cognition in giving weight to the boundary effects, and other subtle features, especially those arising from the coloring. And unlike other measures of complexity, such as D0, (also known as the capacity or box-counting dimension; an upper bound on the Hausdorff-Besicovich dimension), which weight the hypercubes of the computational algorithm equally no matter the amount of the occupancy of each hypercube, D2 weights the hypercubes by the extent to which they are each occupied, and thus the more dense images produce higher estimates of dimension.
What else might one investigate concerning judging complexity? Our instructions to participants were quite nonspecific, giving little guidance to participants as to what meaning they might wish to construct or assume for it. Very likely there was a great range in both what they considered complex, and the extent to which they may have been aware of features of the images to which they were responding. We have not made a formal analysis of the participants’ narratives, but many participants corroborated the kinds of details mentioned above. The similarity of the three groups suggests that all participants were rather sophisticated in observing these details. Garner (1962, 1974) speaks of quantifying aspects of spatial dependencies of images within an informational theoretic framework as distributional redundancy and correlational redundancy. Symmetry, a feature studied in many psychophysical studies that tends to reduce perceptual complexity (Atteneve, 1955; Day, 1967, 1968; Perkins, 1932) could not be a factor here, as our images are devoid of this feature along with many other features used in earlier studies. While our images do not contain subsymmetries (Alexander & Carey, 1968) either, their study could be relevant in terms of the nonhomogeneities present in our fractal images.
Randomly constructed polygons were frequently used for studies of complexity during the 1950s to the 1970s. Some studies found that the number of sides of random polygons were a major determinant of perceived complexity (Arnoult, 1960; Attneave, 1957, Attneave & Arnoult, 1956; Day, 1967; Munsinger & Kessen, 1964; Stenson, 1966). Generally, these studies showed complexity judgments as an increasing monotonic function of the number of sides, considered a measure of complexity, of the judged images (Day, 1967). Attneave (1957) also showed that P2/A (P is perimeter/ A area of a polygon) and angular variability also contributed to judgments of complexity.
Another popular basis for image formation for the study of complexity were checkerboard patterns formed of black and white squares (Berlyne, 1958; Chipman, 1977, Chipman & Mendelson, 1979; Dorfman & McKenna, 1966; Houston, Garskof, & Silber, 1965; Karmel, 9166, 1969; Nicki, 1972; Smets, 1973). Some of these showed perceived complexity as an increasing function of the number of elements and as an inverse function of redundancy (information theoretic term for orderliness; Houston et al., 1965; Karmel, 1966, 1969). Karmel (1966, 1969) showed that the amount of contour of such figures was more important than other aspects of pattern complexity. This feature could be a candidate considered analogous to both external and internal contours of our attractors, and to similar boundary-like features within our attractors.
The studies by Chipman (Chipman, 1977; Chipman & Mendelson, 1979) are strongly suggestive of our contention that non-homogeneous features of our attractors are relevant for the subjective complexity of our images. She used 6x6 checker-board-like matrices, 12 of the 36 squares being black, scattered within the matrix producing differing degrees of complexity that was quantified as to number of turns (corners, which is correlated with number of sides), P2/A—a measure of contour complexity; think of Koch or Peano curves, or better, the Koch Island which builds with squares rather than the Koch snowflakes built algorithmically using a Cantorian replacement process from triangular components (Mandelbrot, 1977, Chap. 2). She also used various measures of symmetry (partial and whole) around vertical, horizontal, and diagonal axes, and repetition (missed by symmetry measures). She performed stepwise regression analysis of magnitude estimates to these variables. Separate sections of her study progressively orthoganalized confoundings among these independent variables. Thus she was attempting to determine that “a wide variety of pattern structures may be effective in reducing perceived complexity.” (p. 271.). She points out that for stimuli constructed from particular elements, that there could be arbitrarily different ways of measuring dimensions of complexity, as Brown and Owen (1967) had also shown. In discussing the work of Alexander & Carey (1968) on subsymmetries, she points out that “The particular form in which the pattern variables were computed was determined both by the desire to detect partial organization and by the stipulation that they should be readily extended to patterns of arbitrary size or into a continuous function.” The concept of fractal dimension introduced by Mandelbrot (1977) accomplishes those objectives of a continuous measure independent of size of image, although it fails to determine some the features of internal structure within our images.
Chipman found that complexity judgments increased with increasing number of corners, and that almost all aspects of patterned organization, most especially symmetry features, reduced complexity judgments from those limits. This may seem at odds with our result—the ∩-shaped function—unless we identify our internal organizational features with her corners, i.e., her complexity, rather than with her patterned features like symmetry, which are lacking in our images. Thus when you get to the homogeneous figures of our highest fractal dimension, there is a lack of much internal structure. It might be noted that, while most of her experiments showed complexity judgments as an increasing function of image complexity, in one experiment where range of the number of corners was very high, there is also the hint of the ∩-shaped function, resulting from a drop in complexity judgments at the highest value of the number of corners in her images. That is, when her image complexity gets great enough, perhaps it loses perceptual organization like ours, and appears more homogeneous or its organizational features are too small or too overwhelmed to detect. She did show, with lesser complexity, that partial areas of organization did tend to reduce perceptual complexity, which is akin to Wertheimer’s Prägnanzstuƒen which means “regions of figural stability” (Wertheimer, 1923).
Her findings also show that the reduction in perceived complexity becomes effective above a threshold of image complexity, when it assumes a power function whose constant quantifies the magnitude of the reduction. Also there was a interaction between the number of corners and degree of symmetry suggesting that symmetry becomes more effective when closer to an “ideal of perfect organization”, Prägnanz (Köhler, 1920), i.e., a perceptual attractor, a result similar to that of Zusne (1971) who found symmetrical forms were perceived as more symmetrical than they actually were, a classic Gestalt result.
Chipman also showed that if different types of organization are combined in an image, they compete rather than summate to produce the reduction in perceived complexity. That suggests there is a self-organizational process going with bifurcations allowing differing basins of attraction to be attempted in the perceptual process of assessing pattern and pattern complexity. As is well known, systems are unstable the closer control parameters are to bifurcation points (Abraham, Abraham, & Shaw, 1990; Aks & Sprott, 2003). It might require complex nonlinear psychophysical models, such as Gregson’s Γ recursion models (Gregson, 1995a,b), to adequately model such a dynamical perceptual process, which we will not attempt here. Aks and Sprott (2003), point out that another modeling process, that of self-organized criticality, may also be employed for bifurcations “from the interaction of the system’s component parts” which could apply to the attentional components within our images, similar to the dynamics within Bak’s sandpiles (Bak, Tang, & Wiesenfeld, 1987).
We might mention some other historical precedents for studying stimulus complexity. Many studies have used complexity as a property of stimulus pattern in a variety of investigations of curiosity, arousal, emotion, and aesthetics, etc, as one of a set of factors which Berlyne calls ‘collative’ which also included ‘novelty, surprisingness, ambiguity, and puzzlingness’ (Berlyne, 1971, p. 69) as well as ‘structural’ and ‘formal ones (Berlyne, 1974, p. 5); Cupchik & Berlyne, 1979). Here we concentrate on those studies simply for their concern with ratings of complexity, and take up those involved in aesthetics and time estimation later. Day (1965) used rank order to relate subjective complexity to objective categories of complexity of stimuli, but not to quantified properties of the physical complexity of stimuli. Most of his stimuli were geometric and curvilinear shapes, but some were not unlike our attractors in some respects. Berlyne, Ogilvie, & Parham (1968) used the same (Day’s) stimuli (Figure 4) and the Shepard-Kruskal multivariate-scaling procedure with principal-axis rotation obtaining results that indicate “that subjective complexity depends primarily on the two principal determinants of information content.” (Berlyne, 1971, p. 201.)
Cupchik’s work on experimental aesthetics goes back to his work with Berlyne and their colleagues at Toronto, with some of his work related the dimensional analysis (factor analytic rather than fractal) of art (Cupchik, 1974). This first paper of his (1974) upon which we wish to focus starts with Wölfflin’s (1915) characterization of five dimensions or stylistic factors of art as linear versus painterly, plane versus recession, closed versus open form, multiplicity versus unity, and absolute versus relative clarity (p. 236). We would view these as in dynamic interplay, creating various basins of perceptual attraction, for both artist and viewer, which would depend upon the interaction of features of the art, personal/cognitive features of the artist/viewer, and cultural/contextual factors. For our study, some of these dimensions are more heavily loaded than others, for example, dimensions of linear-painterly (objective or systematic nomalization versus subjective or systematic distortion in perceptions of art, pp. 237-239) and multiplicity-unity would be more important. Cupchik mentions Zimmermann (1858), Riegl (1893), and Worringer (1908) for the roots of Wölfflin’s distinctions, but when he mentions that “Neither style depicts external reality with complete fidelity.” (p. 237) it suggests the distinction between Parmenides’ world of fixed, eternal, or static ideals and his doxa, the world of the illusory, and his efforts to explain the dynamics between the two. Thus for any given dimension, such as linear-painterly, or multiplicity-unity, there could be two basins of attraction, for linear-painterly, one attractor for external ideals, and one for perceptual creative involvement. The coupling constants could make one predominate with resulting bifurcations, say eventually to a single attractor, perhaps obeying a simple cusp-catastrophe. Or the parameters could allow a continued perceptual trajectory in a multi-basin portrait, or could create bifurcations to cyclic or complex chaotic attractors, all of which requires self-organization and navigation in parameter space.
It might seem something of an excursus to continue a bit more on Wölfflin, but we might point out that Wölfflin was influenced by Dilthey’s historical/scientific hermeneutics, which may be reflected especially in his linear-painterly distinction, which in some ways is like the logocentric-deconstructive distinctions made in post-modern literature (e.g., Derrida), much of which was centered on not only social progress, but also on linguistics, art and architecture (Benjamin, 1936/1979; Baudrillard, 1981/1983). Kristeva (1980), for example makes a linguistic distinction between symbolic (Oedipal, paternal) and semiotic (pre-Oedipal, maternal), which is paralleled by a distinction in personal space as being more metric (Aristotle’s topos) or more amorphous (Plato’s chora). And all of these distinctions are paralleled by the long-running difference of approaches in philosophy and history between the Parmenidean-vs-Heracletian ontological perspectives. Can short-term perception and aesthetics be a fractal zoom on different time scales of self-similar processes that are themselves interdependent? Cupchik notes these parallels also in characterizing the “external factors of the model: democracy-autocracy and consistency-inconsistency of communication and reinforcement.” (p. 244.)
Cupchik performed two experiments using 16 paintings from the Art History Library at the University of Wisconsin that varied on four of Wolffin’s dimensions, as selected by expert judges. In the first experiment, student participants rated the dissimilarity of each of the 120 pairings of the paintings. Nonmetric scaling showed “The dominant dimension distinguished paintings which share an emphasis on outline, and contrasts them with others that appear to emphasize qualities, and whose forms are blurred or indeterminate.” (p. 348.) He compared this to the linear-painterly dimension, thus giving it further definition. Similarly, the scaling procedure showed a dimension close to the tonal-versus-colorful discrimination, one close to the abstract-versus-representational distinction, and the weakest dimension turbulent, active versus simpler, calmer, which seemed to correspond to the multiplicity-versus-unity dimension. (p. 349.) A second experiment was used in which participants rated each of the same paintings on 16 scales, and was analyzed with factor analysis. The first factor emphasized line, clarity, ideas, and coldness, thus being like the linear characteristic, which he called “Classicism”. The second factor de-emphasized accurate reproduction and emphasized color composition and shapes, and thus called “Subjectivism”. (p. 253.) A third was labeled “Complexity”, and the fourth, emphasized emotion, which he called “Expressionism”.
The images in our study may have confounded the complexity dimension, as defined by D2, with the linear-painterly dimension, since not only does the space get filled more completely, but the attractors have more linear features, going to greater homogeneity as D2 increases. It is easy to see how this could affect judgments not only of complexity, but of aesthetics and time as well.
Information theoretic measures of aesthetic patterns have been employed in music (Pinkerton, 1956) and images (Atteneave, 1957, 1959). Some studies related subjective complexity to informational measures of polygons randomly constructed with increasing number of sides (Arnoult, 1960; Atteneave & Arnoult, 1956; Day, 1967; Munsinger & Kessen, 1964), while other used checkerboard patterns (Houston, Garskof, & Silber, 1965; Karmel, 1966, 1969). For some of these studies, subjective complexity was proportional to objective complexity, and inversely proportional to redundancy (Day, 1965; Houston et al., 1965; Karmel, 1966, 1969) as have been several studies of complexity judgments for music (Crozier, 1981). We will consider that when the function is increasing monotonic instead of ∩-shaped in those studies, that the parameter space (independent variable) was not sufficient to show the complete function, that is, the upper limb of the ∩-shaped function.
Walker (1981) offers other explanations for both a rising function of subjective complexity as a function of objective complexity as well as for the ∩-shaped function, depending on the circumstances. For the rising function,
“There can be situations in which the range of complexity values of the physical stimulus can be quite large while the range of psychological complexity values can be considerably restricted. This can occur when the subject fails to process all of the material present in the physical stimulus, especially of the more complex stimuli. A result could be an inverted U when all of the information is processed, but an essentially linear rising pattern when it is not.” (Walker, 1981, p. 47.)
This explanation is consonant with an experiment by Olson (1977) who obtained the monotonic rising result, and with our present experiment with its ∩-shaped function, since our participants may not have been able to perceive or appreciate the mathematical complexity of the images of highest complexity.
We conclude that the ∩-shaped function of subjective complexity as a function of the fractal dimension could be due to a combination of the more amorphous nature of the images with the highest D2, and the lack of time or ability to appreciate the complexity within those images. The obvious nature of the increased complexity of the images with the lowest three ranges of D2 is obvious and responsible for the increasing limb of the ∩-shaped function.
While aesthetics did not provide the principal motivation for the present experiment, it deserves some attention because aesthetics has long been known to be, in part, a function of complexity. The early history of ideas of unity and diversity (chaos depends on the interaction of these as tendencies of convergence and divergence) in aesthetics is summarized by Berlyne(1960) and Gilbert & Kuhn (1953). “The work on the experimental aesthetics of simple visual forms that began with Fechner’s Vorschule der Äesthtik (1876) tends to confirm the view that some intermediate degree of complexity produces the most pleasing effect and the extremes of simplicity or complexity are distasteful (Berlyne, 1960, p. 237).” Fechner mentioned the veining in marble as an example.
Birkhoff, an early developer of mathematical dynamics (Birkhoff, 1927, 1932) formulated a mathematical theory of aesthetics in which complexity was a factor (Birhoff, 1932). Aesthetic value (M) was a function of complexity C of the image (diversity or numerosity) upon which attention and tension depended, and order (unity, due to properties such as symmetry) of the image, upon which (as updated by Graves, 1951) resolution of the tension depended. Birkoff’s formula was M = O/C. That is, aesthetic value should be proportional to order, and inversely proportional to complexity. Very shortly, there were attempts to test this theory (Davis, 1936; Eysenck, 1941) which found, similar to our result, that there was a maximum of aesthetic judgment at intermediate values of Birkhoff’s M. Our measure of complexity, D2, treats complexity as a single dimension stretching between order and complexity, rather than assuming a composite function, although much dynamical thinking looks at oppositional forces along dimensions of the state space, and our generative equations utilize three variables. The curvilinear relationship of all our judgments suggests that these judgments are not unidimensional.
Rashevsky (1938) suggested a possibility based on mathematical assumptions of excitation and inhibition in cortical neurons which gave a “measure of total excitatory effect, which is identified with aesthetic value. Rashevsky’s measure sounds very much like a measure of complexity, but it actually bears a curvilinear relation to Birkhoff’s M, reaching a sharply delineated maximum when M is at an intermediate value (quoted in Berlyne, 1960, p. 239).
In the 1950’s, there was a dramatic development in ‘intrinsic’ motivation, moving beyond the ‘extrinsic’, interoceptively driven drives such as hunger and thirst, touched off primarily by the work of Harry Harlow and his students at Wisconsin (Harlow, 1950). These included studies of manipulatory and exploratory behavior, arousal, and curiosity. Some studies of exploratory behavior had already appeared (Dashiell, 1925; Mote & Finger, 1942; Nissen, 1930). Daniel Berlyne also developed a major program of research at Toronto (1960). Visual stimuli are important in such studies, and Berlyne (1958, 1969) was thereby apparently drawn to art, beginning as evident in a note on the golden mean, which Fechner (1876) had also studied. One of the extraordinary features of his program in addition to the innovative experimental methods was his development of a coherent theoretical framework throughout all these directions of exploration, curiosity, play, humor, and aesthetics which brought together elements of neurophysiological arousal theory and behavior conflict theory (Berlyne, 1971). His two-factor approach began with Wundt’s arousal theory (Wundt, 1874), which showed a ∩-shaped arousal curve resulting from the summation of positive and aversive hedonic tendencies (Figure 5; from Berlyne, 1971).
Wundt proposed the relevance of this conflict-arousal theory to aesthetics: “. . . the simplest cases of the pleasing and displeasing have been almost entirely lost sight of, although they constitute a necessary basis for psychological theory including the explanation of the most complicated aesthetic effects.” (Wundt, 1874, p. 222; quoted in Berlyne, 1971, p. 28).
Berlyne also used the linear dynamical approach-avoidance conflict model of Miller (Miller, 1937, 1959) to explain ∩-shaped aesthetic curves as a function of positive and negative affect, including the influence of novelty and complexity on response strength (Berlyne, 1971, p. 166). Lewin, known for his dynamical field theories, also had a model of approach-avoidance conflict (Lewin, 1951). There have been some simple nonlinearizations of Miller’s model (Figure 6; Abraham, 1995; Abraham et al., 1990; Townsend & Bussmeyer, 1989) and of Lewin’s model (Abraham, 1997a).
Most of the stimuli used in earlier experiments were not particularly aesthetic (Figure 4), being mostly simple black white line drawings varying in complexity, informational content, dissonance, novelty, etc, they thus did not cover a broad range of aesthetic possibilities. It is easy to imagine the stronger aversive affect toward them dominating the weaker pleasing affect toward them, thus generating the ∩-shaped function of aesthetics to complexity, but that involves some assumptions about the homogeneous figures in our highest fractal images being either aversive compared to the expectation of the more aesthetic figures included, or at least being more boring. To further conjecture on such possibilities, it may first be worthwhile to look at some of the more recent work based on dynamics and more aesthetic images.
There are two principal lineages of contemporary studies, those based on aesthetic judgments of fractal images, inaugurated by Sprott and his colleagues (the Sprott Squad); and those using images based on art, such as those of Taylor and his colleagues (Team Taylor).
Sprott first announced his study of aesthetic evaluations of chaotic attractors in his earlier book (Sprott, 1993). He used monochromatic (black-on-white) two-dimensional quadratic maps displayed on computer screen and ratings on a five-point scale, and obtained the ∩-shaped function with preferred D2s in the 1.1 to 1.5 range. Aks and Sprott (1996) replicated this result using a slightly different psychophysical method, that of choosing the most preferred image from four varying in D2 and the principal Lyapunov exponent. While they got a ∩-shaped function for both measures, we did not investigate the Lyapunov exponent as our images were drawn so fast on the computer screen it did not seem to offer any possibility for visual evaluation of the divergence of trajectories. That possibly may have been premature, both on the basis of the positive results of both the Sprott (1993) report and the Aks & Sprott (1996) study, and it could be that the Lyapunov, λ, may have contributed to the internal detail at mid-range dimension, and to the homogeneity of the for the high dimensional images, with D2 mainly measuring the result. That is, D2 may be more a result of the dynamics in filling the space, but λ may be a better reflection of the flow of the dynamics that determines the clumping or internal detail, the tendencies to converge as well as diverge. Aks & Sprott these properties, that D2 measures the filling of space and λs measure the sequential properties of the divergence and convergence of trajectories.
We think that our study shows a preference for images with higher D2s not merely because the Cartesian embedding space is 3D, making three the upper limit of possible D2s compared to the upper limit of two for the 2D space of Aks and Sprott’s images, but possibly the greater aesthetic value that color affords and its emphasis on the internal heterogeneity of the images. We conclude that the preference for slightly higher D2s in this study is real and due to the changes in images, mainly in the use of color and context making our images a bit more appealing.
Scott Draves developed an ingenious system of obtaining voting from thousands of internet users for his animated art deployed as screen-savers which he called electric sheep (Draves, 2005a,b; based on Dick, 1986). These were brief animations using 2D nonlinear iterated function systems to create fractal attractors of considerable aesthetic value. The population of deployed screen-savers evolved over time based on voting, and were generated by (a) random seeding of computational parameters, (b) genetic algorithms, and (c) users’ contributions. Sprott’s fast algorithm (Sprott, 1994, 2003) was later applied to the images to determine their D2s (Draves et al., in press). They obtained a positively skewed ∩-shaped function of voting to D2 with an average D2 = 1.52 +/- 0.23 for the most highly rated sheep, possibly an underestimate of central tendency due to the skewing. While the frequency distributions of the images were also extremely highly skewed, and not evenly distributed as in the Sprott, Mitina, and Abraham studies, it appears that the voting did not simply follow these underlying distributions, but represented real preferences for the mid-dimensional animated images, especially in rejection of the very highest D2s and the randomly seeded animations. Thus there was not only a preference for the mid-dimensional images, but an evolution toward those preferences (Draves et al., in press, figure 3). There was a similar evolution of Pollack’s drip paintings from D0 = 1.1 to 1.7 over the years 1944 to 1954 (Taylor; 2002, 2005; Taylor et al., 2003), representing a tendency to increase toward greater complexity, while the Draves’ evolution seemed to be more downward, resulting from the extinction of higher dimensional images.
Taylor began by establishing that Pollock’s paintings satisfied the mathematical criteria as fractal images from resolving the measurements of D0 (“box counting” or “capacity” dimension, whose values are close to D2 although D2 weights denser areas of the attractor more (Taylor et al., 1999). Taylor found that D0 increased for Pollock, both rapidly in the early stages of crating one painting (1950), from about .5 to 1.9, as well as over the 10 year period mentioned above; presumably representing Pollock’s own aesthetic preferences for D0 (Taylor, Micolich, & Jonas, 2002).
Many have speculated that aesthetic appreciation of fractal images and environment could be due to nature (evolution of neural properties) or nurture (experience in natural and social environments). The former would suggest the possibility of universal preferences for an ideal value of fractal dimension, and to test this concept, Taylor, using the method of paired comparisons, and images from nature, art, and computer-mathematical generation (Taylor, 2001, 2005; Spehar, Clifford, Newell, & Taylor, 2003) and found a maximum preference for mid-value D0s = 1.3-1.5 “irrespective of their origin”. They note that our failure to obtain gender and cultural differences supports this finding of universality, it should be obvious that our small N and cultural and educational conditions were not optimal for finding such differences, and more careful work in that direction would still be worth doing despite exponential growth of global cultural homogenation.
In addition to their confirmation of the ∩-shaped function, they report that the galvanic skin response (skin conductance, a measure of sympathetic activity related to arousal) exhibits a U-shaped function, that is, arousal is minimal when aesthetics is high (Taylor, 2005, 2006; Taylor, Spehar, Wise, Clifford, Newell, & Martin, 2005; Wise & Taylor, 2003). It would be nice to see this work replicated with a wider range of D0 and with GSR measures to the images rather than only arising from their attenuation to the stress of cognitive tasks. Their finding, nonetheless, could be important to the Miller-Dollard-Berlyne-Abraham conflict theory, showing minimal aversion and maximal positive affect at mid-dimensional values. The generic ∩-shaped function is similar the gradients postulated by Miller for his rats staying or oscillating near a midway point in a runway where they have been both shocked and given water at the goal end (Miller, 1937a,b; 1959; see also Abraham, 1995; Abraham et al., 1990; Berlyne, ). Their discussion of evolutionary or ‘neuro-aesthetics’ (see especially Taylor, Spehar, et al., 2005; Zeki, 1999) addresses a broad history of speculation on that subject (Barrow, 2003; Mandlebrot, 1979, 1989; Zeki, 1999). While there is often a confounding of phylogenetic and ontogenetic factors involved in such speculations, the possibility is consistent both with the universality mentioned previously, and the limbic aspects of conflict theory.
Summary & Conclusions
We thank Debbie Aks for sharing her expertise in conducting psychophysical experiments and Anna Lourd Villaneuva, a friend and graduate student, for assisting at making her friends among the Ati and Sulod feel comfortable participating in our experiment. We also thank Ms. Iyoyo, Principal of the Silliman University Elementary School, Dr. Margaret Udarbe-Alvarez, Chair of the Department of Psychology, and Dr. Betsy Joy Tan, then Dean of the School of Education, and now Vice President of Academic Affairs, for their support, Dr. Christian Schales for setting up our computer laboratory, and Dr. Agustin Pulido, President of Silliman University, for establishing that laboratory for the Department. Thanks also to Robert Gregson, dynamical psychophysicist extraordinaire, for comments on our manuscript, and on the contributions of Berlyne, Birkhoff, Davis, Eysenck, and Graves. Gregson also provided analyses of eigenvalues of matrices from our results giving sophisticated support to the obvious limitations of the ANOVA’s and of the averaging of correlation coefficients due to heterogeneity of variances, large variations in individual differences, nonlinear features of psychophysical processes (his series of books on that subject being leading proponents of that field), and possible correlations in variances. I pass some of these on in an appendix, to show not only his sophistication, but the extent to which he went to assist us in our analyses. A true scholar in our field, and one who much deserved the award given him by the Society for Chaos Theory in Psychology and the Life Sciences.
Abraham, F.D. (1995). Dynamics, bifurcation, self-organization, chaos, mind, conflict, insensitivity to initial conditions, time, unification, diversity, free will, and social responsibility. In R. Robertson & A. Combs, (Eds.), Chaos theory in psychology and the life sciences (pp. 155-173). Mawah: Earlbaum.
Abraham, F.D. (1997a). An holistic thread in the dynamical fabric of psychology. In Montouri, A. (Ed.), The Dialectic of Evolution: Essays in Honor of David Loye. World Futures, 49, 159-201.
Abraham, F.D. (1997b). Nonlinear coherence in multivariate research: Invariants and the reconstruction of attractors. Nonlinear Dynamics, Psychology, and Life Sciences, 1, 7-33.
Abraham, F.D., Abraham, R.H., & Shaw, C.D. (1990). A visual introduction to dynamical systems theory for psychology. Santa Cruz: Aerial.
Abraham, R.H., & Shaw, C.D. (1982/1992). Dynamics: The geometry of behavior. Santa Cruz: Aerial/Redwood City: Addison-Wesley.
Aks, D.J., & Sprott, J.C. (1996). Quantifying aesthetic preference for chaotic patterns. Empirical Studies of the Arts, 1(1), 1-16.
Aks, D.J., & Sprott, J.C. (2003). The role of depth and 1/f dynamic in perceiving reversible figures. Nonlinear Dynamics in Psychology and Life Sciences, 6, 159-178.
Alexander, C., & Carey, S. (1968). Subsymmetries. Perception & Psychophysics, 4, 73-77.
Anderson, C.M., & Mandell, A.J. (1996). Fractal time and the foundations of consciousness. In E.R. MacCormac & M. I. Stamenov (Eds.), Fractals of brain, fractals of mind (pp. 75-126). Philadelphia: Benjamins.
Arnoult, M.D. (1960). Prediction of perceptual responses from structural characteristics of the stimulus. Perceptual and Motor Skills, 11, 261-268.
Attneave, F. (1954). Some informational aspects of visual perception. Psychological Review, 61, 183-193.
Attneave, F. (1955). Symmetry, information, and memory for patterns. American Journal of Psychology, 68, 209-222.
Attneve, F. (1957). Physical determinants of the judged complexity of shapes. Journal of Experimental Psychology, 53, 221-227.
Attneave, F. (1959). Application of information theory to psychology. New York: Holt.
Attneave, F., & Arnoult, M.D. (1956). The quantitative study of shape and pattern perception. Psychological Bulletin, 53, 452-471.
Avital, T., & Cupchik, G.C. (1998). Perceiving hierarchical structure in nonrepresentational paintings. Empirical Studies of the Arts, 16(1), 59-70.
Bak, P., Tang, C., & Wiesenfeld, K. (1987). Self-organized criticality: An explanation of 1/f noise. Physical Review of Letters, 59, 381-384.
Barnsley, M.P., & Hurd, L.P. (1992), Fractal image compression. Wellesley: AK Peters.
Barrow, J.D. (2003). Art and science — Les liaisons dangerouses? In J. Casti & A. Karlqvist (Eds.), Art and complexity, pp. 1-20. Amsterdam: Elsevier.
Baudrillard, J. (1981/1983). Simularcres et Simulation. Paris: Galilee. P. Foss, P. Patton, & P. Beitchman (Trans.). Simulations. New York: Semiotest(e).
Behrman, B.W., & Brown, D.R. (1968). Multidimensional scaling of form: A psychophysical analysis. Perception & Psychophysics, 4, 19-25.
Benjamin, W. (1936/1969). The work of art in the age of mechanical reproduction. In H. Arendt (Ed.), H. Zohn (Trans.), Illuminations.. New York: Schocken.
Berlyne, D.E. (1958). The influence of albedo and complexity of stimuli on visual fixation in the human infant. British Journal of Psychology, 49, 315-318.
Berlyne, D.E. (1960). Conflict, arousal, and curiosity. New York: McGraw-Hill.
Berlyne, D.E. (1969). La section d’or et la composition pictural occidentale et orientale. Sciences d l’Art, 6, 1-5.
Berlyne, D.E. (1971) Aesthetics and psychobiology, New York: Appleton-Century-Crofts.
Berlyne, D.E. (1974). (Ed.), Studies in the new experimental aesthetics: Steps toward an objective psychology of aesthetic appreciation. Washington, D.C.: Hemisphere.
Berlyne, D.E., & Olgivie, J. (1974). Dimensions of perception of paintings. In D.E. Berlyne (Ed.), Studies in the new experimental aesthetics: Steps toward an objective psychology of aesthetic appreciation (pp. 181-226). Washington, D.C.: Hemisphere.
Berlyne, D.E., Olgivie, J.C., & Parham, L.C.C. (1968) The dimensionality of visual complexity, interestingness and pleasingness. Canadian Journal of Psychology, 22, 376-387.
Birkhoff, G.D. (1927). On the periodic motions of dynamical systems. Acta Mathematica, 50, 359-379.
Birkhoff, G.D. (1932). Sur quelques courbes formées remarquables. Bulletin Société Mathématique de France, 60, 1-26.
Birkhoff, G.D. (1933). Aesthetic measure. Cambridge: Harvard.
Brown. D.R., & Andrews, M.H. (1968). Visual form discrimination: Multidimensional analyses. Perception & Psychophysics, 3, 401-406.
Brown, D.R., & Owen, D.H. (1967). The metrics of visual form: Methodological dyspepsia. Psychological Bulletin, 68, 243-259.
Chipman, S. (1977). Complexity and structure in visual patterns. Journal of Experimental Psychology: General, 106, 269-301.
Chipman, S., &. Mendelson, M. (1979). Influence of six types of visual structure on complexity judgments in children and adults. Journal of Experimental Psychology: Human Perception and Performance, 1979, 5, 365-378.
Crozier, J.B. (1981). Information theory and melodic perception. In H. I. Day (Ed.), Advances in intrinsic motivation and aesthetics (pp. 433-461). New York: Plenum.
Cupchik, G.C. (1974). An experimental investigation of perceptu8al and stylistic dimensions of paintings suggested by art history. In D.E. Berlyne (Ed.), Studies in the new experimental aesthetics: Steps toward an objective psychology of aesthetic appreciation (pp. 235-257). Washington, D.C.: Hemisphere.
Cupchik, G.C., & Berlyne, D.E. (1979). The perception of collative properties in visual stimuli. Scandinavian Journal of Psychology, 20, 93—104.
Cupchik, G.C., & Gebotys, R.J. (1988). The experience of time, pleasure, and interest during aesthetic episodes. Empirical Studies of the Arts, 6(1), 1-12.
Cupchik, G.C., & Heinrichs, R.W. (1981). Toward an integrated theory of aesthetic perception in the visual arts. In H. I. Day (Ed.), Advances in intrinsic motivation and aesthetics (pp. 463-485). New York: Plenum.
Dashiell, J.F. (1925). A quantitative demonstration of animal drive. Journal of Comparative Psychology, 5, 205-208.
Davis, R.C. (1936). An evaluation and test of Birkhoff’s aesthetic measure and formula. Journal of General Psychology, 15, 231-240.
Day, H.I. (1965). Exploratory behavior as a function of individual differences and level of arousal. Unpublished Ph.D. thesis, University of Toronto.
Day, H.I. (1967). Evaluation of subjective complexity, pleasingness, and interstingness for a series of random polygons varying in complexity. Perception & Psychophysics, 2, 281-286.
Day, H.I. (1968). The importance of symmetry and complexity in the evaluation of complexity, interest and pleasingness. Psychonomic Science, 10, 339-340.
Derrida, J. (1967a). Of Grammatology. Baltimore and London: John Hopkins.
Derrida, J. (1967b). Speech and Phenomena. Evanston: Northwestern.
Derrida, J. (1967c, 1978). Writing and Difference. London: Routledge & Kegan Paul.
Dick, P.K. (1968). Do androids dream of electric sheep? New York: Ballantine. (Adpted for Scott’s film, Bladerunner.
Dorfman, D., & McKenna, H. (1966). Pattern peference as a function of pattern uncertainty. Canadian Journal of Psychology, 20, 143-153.
Draves, S. (2005a). The Electric Sheep Screen-Saver: A case
study in aesthetic evolution. In: Rothlauf, F., Branke, J.; Cagnoni, S., Corne,
D.W., Drechsler, R., Jin, Y., Machado, P., Marchiori, E., Romero, J., Smith,
G.D., Squillero, G. (Eds.), EvoWorkshops. LNCS, 3449, 458-467. Heidelberg:
Draves, S. (2005a). The electric sheep screen saver: A case study in aesthetic evolution. Applications of Evolutionary Computing, Lecture Notes in Computer Science, 3449.
Draves, S. (2005b) The Fractal Flame Algorithm. http://flam3.com/flame.pdf.
Draves, S., Abraham, R., Viotti, P., Abraham, F.D., & Sprott, J.C. (in press). The aesthetics and fractal dimension of electric sheep. International Journal of Bifurcation and Chaos. http://www.ralph-abraham.org/articles/full-list.html (MS#120)
Eisler, A. D. (2003). The human sense of time: Biological, cognitive, and cultural considerations. In R. Buccheri & M. Saniga (Eds.), The nature of time: Geometry, Physics and Perception (pp. 5-18). Dordrecht: Kluwer Academic Publishers.
Eisler, H. (1996). Time perception from a psychophysicist’s perspective. In H. Helfrich (Ed.), Time and mind (pp. 65-86). Göttingen: Hogrefe & Huber.G.
Eisler, A.D., Eisler, H., & Montgomery, H. (1996). Prospective and retrospective time perception: Cognitive and biological approaches. In S. Masin (Ed.), Cognitive models of psychological time (pp. 251-256). Hillsdale: Erlbaum.
Eysenck, H.J. (1941). The empirical determination of an aesthetic formula. Psychological Review, 48, 83-92.
Fechner, G.T. (1876). Vorschule der Äesthtik. Leipzig: Breitkopf & Härtel.
Freeman, W.J. (2006). Neurodynamics: An exploration in mesoscopic brain dynamics. http://sulcus.berkeley.edu/
Garner, W.R. (1962). Uncertainty and structure as psychological concepts. New York: Wiley.
Garner, W.R. (1974). The processing of information and structure. New York: Halsted.
Gilbert, K.E., & Kuhn, H. (1953). A history of aesthetics. Bloomington: Indiana.
Gilden, D. L., Schmuckler, M. A., & Clayton, K. (1993). The perception of natural contour. Psychological Review, 100, 460-478.
Grassberger, P., & Procaccia, I. (1983). Measuring the strangeness of strange attractors. Physica D, 9, 189-208.
Graves, E.M. (1951). The art of color and design. New York: McGraw-Hill.
Gregson, R.A.M. (1996). N-Dimensional nonlinear psychophysics. In E.R. MacCormac & M. I. Stamenov (Eds.), Fractals of brain, fractals of mind (pp. 155-178). Philadelphia: Benjamins.
Gregson, R.A.M. (2006). Informative psychometric filters. Canberra: ANU E Press.
Harlow, H.F. (1950). Learning and satiation of response in intrinsically motivated complex puzzle performance by monkeys. Journal of Comparative and Physiological Psychology, 43, 289-294.
Heath, R.A. (2000). Nonlinear dynamics: Techniques and applications in psychology. Mahwah: Lawrance Erlbaum Associates.
Houston, J.P., Garskof, B.E., & Silber, D.E. (1965). The informational basis of judged complexity. Journal of General Psychology, 72, 277-284.
Karmel, B.Z. (1966). The effect of complexity, amount of contour, elementst size, and element arrangement of visual preference in the hooded rat, domestic chick, and human infant. Unpublished Ph.D. thesis, George Washington University.
Karmel, B.Z. (1969). The effects of age, complexity, and amount of contour on pattern preferences in human infants. Journal of Experimental Child Psychology, 7, 339-354.
Kelso, J.A.S., & Engstrom, D.A. (2006). The complementary nature. Cambridge: MIT.
Köhler, K. (1920). Die physischen Gestalten. in Ruhe und im stationären Zustand, Ejine naturphilosophische Untersuchung, Erlangen. Note: Köhler credits Wertheimer with coining the “law of Prägnanz”, the tendency towards simple Gestalten” (form).
Kristeva, J. (1980). Desire in Language. L. S.
Roudiez, ed.; T. Gora, A. Jardine, & L. S. Roudiez, Trans.). New York: Columbia.
(A collection of ten papers spanning from 1966 to 1976, with an excellent
introduction by Roudiez, including a guide and discussion of terminology. See
also, Abraham, 1998 at
Kuhl, J. (1986). Motivational chaos: A simple model. In A.R. Brown & J. Veroff (Eds.), Frontiers of motivational psychology. Berlin: Springer-Verlag.
Laffal, J. (1955). Response faults in word association as a function of response entropy. Journal of Abnormal and Social psychology, 50, 265-270.
Lee, P.L. Ambiguity and preference in Cubist paintings. Unpublished master’s dissertation, University of New Brunswick.
Macar, F. (1996). Temporal judgments on intervals containing stimuli of varying quantity, complexity and periodicity, Acta Psychologica, 92 (3), 297-308.
Macar, F., Pouthas, V., & Friedman, W.J. (1992). Time, action and cognition. Dordrecht: Kluwer.
Mandelbrot, B.B. (1977). The fractal geometry of nature. New York: Freeman.
Mandelbrot, B.B. (1989). Fractals and an art for the sake of art. Leonardo; Supplement: 21-24.
Miller, N.E. (1937a). The analysis of the form of conflict reactions. Psychological Bulletin, 34, 720.
Miller, N.E. (1937b). Reaction formation in rats: An experimental analog for a Freudian phenomenon. Psychological Bulletin, 34, 724.
Miller, N.E. (1959). Libealizaton of basic s-r concepts: Extensions to conflict behavior, motivation, and social learning. In S. Koch (Ed.), Psychology: A study of a science, Vol. 2, pp. 196-292. New York: McGraw-Hill.
Mitina, O.V., & Abraham, F.D. (2003). The use of fractals for the study of the psychology of perception: Psychophysics and personality factors, a brief report. International Journal of Modern Physics C, 14(8), 1-14.
Mote, F.A., & Finger, F.W. Exploratory drive and secondary reinforcement in the aduisition and extinction of a simple running response. Journal of Experiment Psychology, 31, 57-69.
Munsinger, H.L., & Kessen, W. (1964). Uncertainty, structure and preference. Psychological Monographs, 78(9) (Whole No. 586).
Mureika, J.R., Cupchik, G.C., & Dyer, C.C. (2004). Multifractal fingerprints in the visual arts. Leonardo, 37(1), 53-56. Also: 17 May 2005, http://arxiv.org/abs/physics/0505117.
Nicki, R.M. (1970). The reinforcing effect of uncertainty reduction on a human operant. Canadian Journal of Psychology, 24, 389-400.
Nicki, R.M. (1972). Arousal increment and degree of complexity as incentive. British Journal of Psycholgy, 63, 165-171.
Nicki, R.M. (1981). Ambiguity, complexity, and preference for works of art. In H. I. Day (Ed.), Advances in intrinsic motivation and aesthetics. New York: Plenum.
Nicki, R.M., & Lee, (1977). Ambiguity (uncertainty) and preference for Cubist works of art. Paper given at the Annual Meeting of the Eastern Psychological Association, Boston, April.
Nicki, R.M., & Shea, J.F. (1971). Subjective uncertainty, the orientation reaction, and the reinforcement of an instrumental response. Perception and Psychophysics, 7, 374-376.
Nissen, H.W. A study of exploratory behavior in the white rat by means of the obstruction method. Journal of Genetic Psychology, 37, 361-376.
Olson, M.H. (1977). Complexity and preference and information processing rate. Unpublished doctoral dissertation, University of Michigan.
Peckham, M. (1965). Man's rage for chaos: Biology, behavior and the arts. New York: Chilton.
Peitgen, H.-O., & Richter, P.H. (1996). The beauty of fractals: Images of complex dynamical systems. Berlin: Springer.
Perkins, F.T. (1932). Symmetry in visual recall. American Journal of Psychology, 44, 473-490.
Pickover, C.A., (1990). Computers, pattern, chaos, and beauty. New York: St. Martins.
Pickover, C.A. (1995). Keys to infinity. New York: Wiley.
Pinkerton, R.C. (1956). Information theory and melody. Scientific American, 194(2), 77-86.
Rashevsky, N. (1938). Contribution to the mathematical biophysics of visual perception with special reference to the theory of aesthetic values of geometrical patterns. Psychometrika, 3, 253-271.
Riegl, A. (1893/1992). Stilfragen. Berlin. Tr. E. Kain, Problems of style. Princeton: Princeton.
Short, L. (1991). The aesthetic value of fractal images. British Journal of Aesthetics, 31, 342-355.
Skarda, C.A., & Freeman, W.J. (1987). How brain makes chaos in order to make sense of the world. Behavioral and Brain Sciences, 10:161-195.
Smets, G. (1973). Aesthetic judgments and arousal. Leuven: Leuven.
Spehar, B., Clifford, C., Newell, B., & Taylor, R.P. (2003). Universal aesthetic of fractals. Computers & Graphics, 27, 813–820.
Sprott, J.C. (1993a). Strange attractors: Creating patterns in chaos. New York: M&T.
Sprott, J.C. (1993b). Automatic generation of strange attractors. Computers and Graphics, 17, 325-332.
Sprott, J.C. (1994). Automatic generation of iterated function systems. Computers and Graphics, 18, 417-425.
Sprott, J.C. (2003). Chaos and Time-Series Analysis. New York/Oxford:Oxford.
Stamps, A.E., III (2002). Entropy, visual diversity, and preference. Journal of General Psychology,
Stenson, H.H. (1966). The physical factor structure of random forms and their judged complexity. Perception and Psychophysics, 1, 303-310.
Stoyanova, Y., & Yakimoff, N., et al., (1987). In search of painting’s influence on time judgment. Acta Neurobiologiae Experimentalis, 47(2-3), 103-109.
Sulis, W, & Trofimova, I. (2001). Nonlinear Dynamics in the Life and Social Sciences. Amsterdam: IOS Press.
Taylor, R.P. (1998). Splashdown. New Scientist, 2144, 30-31.
Taylor, R.P. (2001). Architect reaches for the clouds: How fractals may figure in our appreciation of a proposed new building. Nature, 410, 18.
Taylor, R.P. (2002, December) Order In Pollock's chaos. Scientific American, pp. 116-119.
Taylor, R.P. (2005). Pollock, Mondrian and nature: Recent scientific Investigations. Chaos and complexity letters, 1(3), 265–277.
Taylor, R.P. (2006). Reduction of physiological stress using fractal art and architecture. Leonardo, 39, 245-251.
Taylor, R.P., Micolich, A., & Jonas, D. (1999). Fractal analysis of Pollock’s drip paintings. Nature, 399, 422.
Taylor, R.P., Micolich, A., & Jonas, D. (2002). The construction of Pollock's fractal drip paintings. Leonardo, 35, 203-207.
Taylor, R.P., Spehar, B., Wise, J.A., Clifford, C.W.G., Newell, B.R., & Martin, T.P., (2005). Perceptual and Physiological Responses to the Visual Complexity of Pollock’s Dripped Fractal Patterns. Journal of Non-linear Dynamics, Psychology, and Life Sciences, 9, 89-114. http://materialsscience.uroregon.edu/taylor/art/Boston.pdf
Turvey, M. T. (2005). Theory of Brain and Behavior in the 21st Century: No Ghost, No Machine. Japanese Journal of Ecological Psychology, 2, 69-79.
Walker, E.L. (1981). The quest for the inverted U. In H. I. Day (Ed.), Advances in intrinsic motivation and aesthetics. New York: Plenum.
Ward, L.M. (2001). Dynamical cognitive science. Cambridge: MIT Press.
Wertheimer, M. (1923). Untersuchungen zur Lehre von der Gestalt, II. Psycholog. Forsch, 4, 301-350. Also as "Laws of organization In perceptual forms", in W.D. Ellis (Eng. Trans), A source book of Gestalt psychology. New York: Humanities, pp. 71-88, see especially p. 79.
Wise, J.A., & Taylor, R.P. (2002). Fractal Design Strategies for Enhancement of Knowledge Work Environments. Proceedings of the 46th Meeting of the Human Factors and Ergonomics Society. Baltimore, MD. PP 854-859.
Wölfflin, H. (1915/1950). Kunstgeschichtiche Grundbegriffe. Munich: Bruckmann. Principles of art history. New York: Dover.
Worringer, W. (1908/1953). Abstraktion und Einguhlung. Munich: Piper.. Abstraction and empathy. New York: International Universities Press.
Wundt, W.M. (1874). Grundgzüge der physiologischen Psychologie. Leipzig: Engelmann.
Yevin, I. (2001). Synergetic of brain and synergetic of art (Sinergetika mozga i sinergetika Iskussiva). Moscow: Geos. (In Russian.)
Zausner, T. (2007). Process and meaning: Nonlinear dynamics and psychology in visual art. Nonlinear Dynamics, Psychology, and Life Sciences, 11, 149-165.
Zausner, T. (2007). Process and meaning: Nonlinear dynamics
Notes for further development, incorporation, deletion (memory aids):
Updates since February 24, 2006. Main updates since last version are in section 4.1 and references.
, and Cupchik and his colleagues (The Cupchik Cabel). Cupchik was a protégé of Berlyne (Cupchik, 1981; Cupchik & Berlyne; 1979.
Eisler, H., Eisler, A. D., & Montgomery, H. (1998). Speed of time and the psychophysical function for subjective duration. In S. Grondin & Y. Lacouture (Eds.), Fechner Day ´98 (pp. 83-88). Québec: International Society for Psychophysics.
Eisler, A. D. (2003). The human sense of time: Biological, cognitive, and cultural considerations. In R. Buccheri & M. Saniga (Eds.), The nature of time: Geometry, Physics and Perception (pp. 5-18). Dordrecht: Kluwer Academic Publishers. [abstract]
Notes for further development of the paper:
Imagine prey-pred or other handy-dandy tool for model, bifurcations and portraits, response diagrams. In our strange attractor art, is the viewer tending to find some ideal form in the internal boundaries of our art, and thus finds the attraction in this attractor, or is the viewer made happier from freedom from form found in a mid-dimensional attractor, as a better less static ideal form?
His dissertation on the psychological aspects to architectural appreciation is a synthesis of the major intellectual influences in his life: Burckhardt's broad view of what constitutes an historical document, Wilhelm Dilthey's psychological approach to historical hermeneutics, Geisteswissenschaften, and Giovanni Morelli's (q.v.) technique of visual comparison, to name but three
Aks, Mitina, Richards
Gregson, Kocic-Stevanovska, others? Taylor Zausner2007bifurcations
The gsr, implication for conflict model, implication for stability/instability
Mitina result similar; Draves result, Berlyne/Yevin dynamics, inverted U, aversion at top end, preference for intermediate complexity with some detail. How do the personality studies (Aks; Mitina; Richards) bear on the Berlyne 2-factor theory; aversion or homogeneity: both probably contribute.
We have noted the high correlation between complexity and aesthetic ratings, which could be attributable to any or all of three factors: (1) response generalization (non independence of response tendency independent of the rating category, (2) both ratings are responding to the same dynamical interaction between stimuli and participant, or (3) subjective complexity mediates aesthetics, linear rather than an inverted U.
2nd lineage: real art
time function: from Mitina,paper of Macar
fragment removed from section 4.1: disposition?
Another approach to the dynamics involved would be to see if people can learn to discriminate dimensionality by viewing images and training on them with feedback or labeling based on their dimensionality. Can the mathematical aspects of images be learned through experience judging them? Besides such learning being an obvious possibility, there was a series of experiments in the 1970’s by Nicki and his associates showing that learning could be influenced by the degree of blurring of images of everyday objects (Nicki, 1970, Nicki & Shea, 1970) and cubist paintings (mainly Braque and Picasso) for which information theoretic uncertainty values (Laffal, 1955; the methodology depending not on objective measurement of stimulus properties but on participants free associations to the paintings) had been established (Lee, 1972; Nicki & Lee, 1977). These studies showed a ∩-shaped function of subjective uncertainty to degree of blurredness (Nicki, 1970; subjective uncertainty of intermediate blurredness at just over .5 bits). They also indicated that learning could depend on a reduction of uncertainty and conflict. Nicki (1981) also showed that increases in ambiguity could increase reward value. These findings raise the issue, that while we often consider complexity as one determinant of aesthetic preference, might it be that that aesthetics can also influence subjective uncertainty (ratings of complexity), that they are involved in a complex system of interactive factors? We take this up later after considering aesthetics.