THINKING THROUGH DRAWING: PRACTICE INTO KNOWLEDGE

104 TEACHERs COLLEGE COLUmbIA UNIvERsITy

Paul the Robot as a Naive Drawer

Background

There are many examples of computerized

systems attempting to draw from reality, e.g., in

computer graphics, a subject referred to as Non-

Photorealistic Rendering or NPR (Gooch & Gooch,

2001; Brennan, 1982; Chen et al., 2004); these sys-

tems produce approximate renderings extrapo-

lated from reality, usually by taking as input digital

images or photographs. NPR systems are meant

to produce figurative paintings, while the scope

of our research is focused on figurative drawings.^1

Recently, Colton explores another facet of NPR by

using an approximate facial expression recogni-

tion AI system to influence the painting style.^2 To

the best of our knowledge, the vast majority of NPR

systems are designed to render drawings in a par-

ticular style by producing output images mimicking

a final result with little attention paid to the creative

steps and feedbacks involved in the artistic gen-

eration itself: i.e., how, as Zeki formulates it (Zeki,

1998), the artist laboriously extracts permanent,

lasting and enduring features from a visual stimulus

forming a novel presentation of the original subject.

The pioneering work of Harold Cohen with

his AARON system (Cohen, 1988) is probably the

most important exception to this (NPR) trend from

Computer Graphics, in which a model of the artist’s

activity whilst drawing/painting from imagination

has been studied, implemented and refined over the

years and successive generations of the system itself.

Since the early days of his work on the AARON

concept in the 1970’s, Cohen has stressed that the

crucial behavior common to any kind of art is the

awareness of the work in progress: therefore some

form of feedback—eventually consiousness—is

essential to art genesis. A program with the ambi-

tion of generating art has to show this capacity of

assessing the result of its past actions to influence

future actions. In contradistinction to Cohen’s

work, which makes no real-time use of computer

vision, we investigate the artististic drawing activity

whilst drawing from reality and memory as well as

from imagination.

There also has been a long tradition of draw-

ing automata or systematic machines which we

can trace back to at least the 18th century, e.g.,

with Maillardet’s automaton which was able to

draw seven sketches and write four poems. Closer

in time to us are important links between the ori-

gin of computational art and drawing machines.

The Algorists who were pioneers of the field made

extensive use of early drawing machines, in particu-

lar pen-plotters. A notable member of this group

is Roman Verostko, who was celebrated in 2009 by

ACM Siggraph with the attribution of a prize hon-

oring achievements in digital art, and who still uses

pen-plotters. Although Harold Cohen has for some

time now used large format ink printers as output

for AARON, until the early nineties he too used

custom drawing/painting machines. Interestingly

the first versions of AARON were using a drawing

robot, a type of mechanical turtle that allowed for

Figure 1. Paul the robot drawing Stella, Tenderpixel Gallery, London, U.K., June 2011.