Human or Machine: A Subjective Comparison of Piet Mondrian's 'Composition with Lines' and a Computer-Generated Picture

Notes

The Psychological Record, 1966, 16, 1-10 (printed at the head of the first page). The issue number (No. 1) is not printed on the article but appears in Noll's own later bibliographies, agreeing with the supplied citation.. Self-archived by the author at http://noll.uscannenberg.org/Art%20Papers/Mondrian.pdf — Noll posts his own papers on his site at noll.uscannenberg.org, which is where this copy comes from. Hosting here is with his agreement, given to Bre Pettis directly on 6 October 2026. Already in the library with empty notes — summary is to enrich it. Noll reads Mondrian's Composition With Lines (1917) as a near-circular field of black vertical and horizontal bars, cropped at sides, top and bottom, with bars inside a region at the top shortened and lengths and widths otherwise apparently random, and suggests Mondrian may have followed some scheme whose algorithm is unknown. He cites Michel Seuphor's account of the horizontal-vertical theme and Mondrian's own writing on the plastic function of sea, sky and stars. He then programs a semirandom picture of similar composition and tests both on human subjects. The machine is an IBM 7094 driving a General Dynamics SC-4020 microfilm plotter whose cathode-ray tube face is photographed by a 35-mm camera. Bars are built as closely spaced, slightly overlapping parallel line segments and placed at random with uniform probability inside a circle of radius 450 units; vertical and horizontal are equally likely; widths are equiprobable between 7 and 10 lines and lengths between 10 and 60 points; a bar falling inside a parabolic region at the top is shortened in proportion to its distance from the parabola's edge. Trial and error tuned the resemblance. The result is Computer Composition With Lines. Both pictures were copied xerographically so that photographic quality gave no clue, then shown in counterbalanced pairs with three questionnaires (background, identification, preference) to 100 subjects, all but two of them Bell Labs employees, with education from high school to post-doctoral: 69 technical and 31 non-technical, 56 male and 44 female. Fifty-nine per cent preferred the computer picture and only 28 per cent identified it correctly, both differing from chance at the 0.05 level on a binomial test. Chi-square tests showed questionnaire order affected neither answer, and preference and identification were independent. Subjects who liked abstract art preferred the computer picture 76 per cent of the time; 35 per cent of technical but only 13 per cent of non-technical subjects identified it correctly, an association significant at 0.001. Noll reads the non-technical failure as an assumption that randomness means human creativity, insists the test compared order against randomness rather than artistic merit, and notes that merit is not settled by jury. This is the founding empirical study in computational aesthetics and the source of its most-quoted statistic. Points of record: 59 of the 100 subjects preferred the computer-generated picture; only 28 correctly identified which picture the computer made; Both percentages differ from chance at the 0.05 level on a binomial test; The computer picture was made on an IBM 7094 driving a General Dynamics SC-4020 microfilm plotter photographed by a 35-mm camera; Bars were placed with uniform probability inside a circle of radius 450 units; widths equiprobable between 7 and 10 lines, lengths between 10 and 60 points; bars in a parabolic region at the top were shortened; 35% of technical subjects but only 13% of non-technical subjects identified the computer picture correctly (chi-square significant at the 0.001 level); Subjects who strongly liked or liked abstract art preferred the computer picture 76% of the time; those who disliked it, 45%; All but two of the 100 subjects were Bell Telephone Laboratories employees; Noll concludes that the experiment compared order against randomness and does not settle artistic merit, and reports that further experiments on the preferred range of randomness were already in progress. Illustrated: Fig. 1 - Composition With Lines (1917) by Piet Mondrian, reproduced with permission of the Rijksmuseum Kroller-Muller, Otterlo; Fig. 2 - block diagram of the method for producing computer pictures (IBM 7094 and General Dynamics SC-4020 microfilm plotter); Fig. 3 - Computer Composition With Lines (1964) by the author in association with an IBM 7094 and a General Dynamics SC-4020 microfilm plotter, (c) A. Michael Noll 1965; Fig. 4 - the picture pair as presented to subjects on separate sheets, both xerographically copied to identical quality. People named: A. Michael Noll - author, Bell Telephone Laboratories, Incorporated, Murray Hill, New Jersey, Piet Mondrian (1872-1944) - Dutch painter of Composition With Lines (1917), Michel Seuphor - critic, cited (Abstract Painting, 1962; Piet Mondrian, 1957) for calling the painting the most accomplished of Mondrian's horizontal-vertical series, J. Bradley - cited for the 1944 Knickerbocker Weekly article calling Mondrian the greatest Dutch painter of our time, D. Lewis - cited, Mondrian (1957), Helen M. Walker and J. Lev - cited, Statistical Inference (1953), for Yates' correction, M. Weitz - cited, on whether art can be attributed any defining properties. Noted: The citation printed on the paper (The Psychological Record, 1966, 16, 1-10) matches the supplied citation; the issue number is not printed on the article. Fig. 3 dates the computer picture 1964 while its copyright line reads 1965. The museum appears throughout the scan as “Rijkmuseum Kroller-Miiller” and similar; i.e. Rijksmuseum Kroller-Muller. Table 2's columns are mis-set in the scan: the chi-square values and degrees of freedom run together, and the “Abstract art assessment” row is marked significant at p < 0.02 while Table 1's note marks cells at the 0.05 level. The follow-up work announced at the end (preferred range of randomness, statistically identical pictures) became Noll's 1972 Psychological Record paper, which is not in this batch.