• Home
• About
• Contact Us
• Activities
• Resources
  • By Discipline
  • By Method
  • By Type
• Publications
• Get Involved
• e-Science

The Drawbots Project: Computational Intelligence, Creativity and Cognition: a multidisciplinary investigation

The Drawbots project is using Evolutionary Robotics to attempt to understand the nature of creativity. The key question is 'what is the simplest robot that can be produced (using Evolutionary Robotics) whose behavior might be described as creative?'

Running from 1 May 2005 to 31 October 2008 this AHRC-funded project has brought together an international team of artists, scientists, philosophers and critical theorists to investigate the relationship between contemporary theories of creativity and the arts and of those of artificial life and artificial intelligence.

The project aims to enhance understanding of creativity and cognition and to stimulate dialogue in this area. It applies contemporary evolutionary and adaptive systems methodology to the field of creative behaviour and the visual arts and intends to create a series of 'artworks'. The University of Sussex and Lancaster University are partners in the project, and international collaborators and contributors include the Creativity and Cognition Studios at the University of Technology, Sydney.

The Project

Aims and Objectives

The objectives of this program are: to investigate the existing body of knowledge concerning the relationship between creative behaviour and evolutionary and adaptive systems; to create autonomous systems both in virtual environments and embodied in real world automata that are capable of manifesting creative behaviour; the development of autonomous agents engaged in creating artworks; to consider the theoretical and historical implications of automatically-generated creative output in the context of theories of aesthetics, creativity and cognition as well as histories of art.

Throughout the twentieth century many artists commented on the autonomy of the artworks they created. Constructivism provided a framework for ‘machine art’ and this became formalized in the 1960s as systems or process art. By 1967 it was possible for the systems/conceptual artist Sol LeWitt to suggest that: ‘The idea becomes a machine that makes the art’. Collaborations between artists and scientists became more common and many turned to computing machines and computational methodologies in order to pursue their work. One of the key questions emerging from such work was whether a defined system could produce (or be) an artefact with the requisite variety to be acceptable to a peer group as a work of art?. There is now no doubt that the answer to this question was affirmative and a large body of relevant work has been adopted as a part of the cultural record.

In the early 1970s a group of artists (many of whom were associated with the Slade School of Art's Computing and Experimental Department) who were influenced by the systems group began to develop ideas that, a decade later, became adopted by science and called ‘artificial life’ or a-life. The techniques of a-life (like evolutionary and adaptive computing) have enabled the investigation of a wide range of problems that do not have defined quantitative outcomes. A computational entity is trained to find a solution to a problem or task that the trainer themselves may not know how to solve. This provides a method for applying computational methodology to qualitative (in contrast to quantitative) problems.

One hypothesis of this research project is that these techniques may be applied to the area of knowledge bounded by the fields of aesthetics, cognition, art and philosophy. The project is attempting to create drawn artworks (not schematics or diagrams) using a-life automata developed with evolutionary and adaptive computing techniques. This will be achieved through a close creative collaboration between a scientist (Phil Husbands) and an artist (Paul Brown). Simultaneously investigators from the fields of philosophy, artificial intelligence and cognitive and computational science and art history, cultural and critical theory are exploring the historical and critical foundations for this experiment. As the work progresses they are developing a framework to contextualize its outcomes. The project addresses practice as well as a more general hypothesis of quantitative or computational aesthetics. It is attempting to discover and demonstrate a robust theoretical foundation for the field.

Two artists who were both associated with the developments at the Slade in the 1970s pioneered the main computational methodologies underlying these investigations. One of these artists was Harold Cohen who adopted a ‘top-down’ approach inspired by the work of the Stanford Artificial Intelligence Laboratory and their development of ‘Expert Systems’. Cohen analysed his creative behaviour and made a rule-based system (AARON, 1970 - present) which can now autonomously produce ‘drawings’. The other artist referred to is Edward Ihnatowicz who was inspired by Piaget’s theories of childhood-learning to create adaptive robotic works (Sam, 1968; The Senster, 1970) These applied a ‘bottom-up’ approach that has since become the dominant methodology of contemporary a-life research. The essence of this approach is the creation of simple systems that learn and adapt by interacting with their environment. The resurgence of interest in neural networks in the late 1980s and the development of genetic algorithms during the 1990s have provided two of the fundamental techniques of this evolutionary approach.

Part of the theoretical motivation for the project is Margaret Boden's threefold theory of creativity. She defines these as: (i) combinational, (ii) exploratory, and (iii) transformational. The two latter involve a structured conceptual space, or style of thinking/ acting. In drawing, these would be styles of drawing. One question behind this research project is whether evolutionary techniques, where an entity’s capabilities are developed through mutations (changes in properties) and selection, can enable a robot to explore and/or to transform a given conceptual space or style. Another is whether a stable style can emerge from evolutionary processing, without predefined constraints being provided (e.g. as explicit functions that test the fitness of the automaton - in the Darwinian sense, or as constraints on just what type of mutations are allowed). In human creativity the conceptual space is provided by the culture, is this also true for the automaton? A third research question concerns the distinction between creative exploration and transformation, or between different degrees of change ? whether in a single mutation or in the resulting style/space.

Boden's theory distinguishes ‘tweaking’ from ‘transforming’, and between ‘superficial’ and ‘fundamental’ transformations. Currently these distinctions are informal. The first type (combinational creativity) may involve random combination, but more usually involves combinations that are constrained by expectations about what might be the most promising types. Using these expectations as fitness functions, the project will have the potential to illuminate combinational creativity too. According to Boden creativity is multi- dimensional because conceptual spaces are multi-dimensional. An artwork may be ’creative‘ (in sense i. ii. or iii) with respect to one dimension (aspect), but not others; or it may be very creative in one way, and only mildly creative in another.

Evaluation plays an essential part in the definition, and therefore the recognition, of creativity and evaluative criteria should be made as explicit as possible. This is, in general, a very difficult matter, whether for art historians, critical theorists, or workers in computer-based art.

Evaluation will form a significant element of the Drawbots project at two major points: (1) in the choice, and justification, of fitness criteria which determine the automata's behaviour; (2) when the aesthetic worth (and creativity) of the automata's output is evaluated. It is intended that the working methodologies developed and adopted by the project will determine and illuminate these evaluative criteria. That is to say that it will be possible to define or suggest means towards defining, some currently qualitative aspects of aesthetics, creativity and critical theory in a more robust and quantitative form.

Three inter-dependent groups work on the project: the ‘cognitive' group - concerned with creativity and cognition; artificial intelligence; philosophy and aesthetics; the ‘robot art’ group - concerned with development of the evolutionary and adaptive systems implementation; and the 'art theory' group, concerned with context and implications for the fields of history of art and critical and cultural theory.

Project Outcomes

During year one the ‘robot art’ group created a working interactive simulator which could be configured for different control and evaluation parameters. The ‘cognitive’ and ‘art history’ groups investigated the history and context of the research and considered the determining parameters of the project.

In year two a working robot simulator will allow the exploration of the various parameters that feed into the direction and outcome of the research. Each of the research groups plan to run experiments using the simulator.

In year three the working robot will be completed and the research will continue in this embodied ‘real world' format. Explorations of environmental interaction will also be made. Various reference and peer groups will be introduced to the robot to provide input and help evaluate performance. The ‘robot art’ group will have created a small community of robots that will be co-evolved and which will have the facility for communication and interaction with each other and their environment. The overall evaluation of the project will also be undertaken and the research output prepared. At the conclusion of the project in October 2008 an exhibition of the robot community will be held in the UK. A international symposium is also planned, in order to showcase the project outcomes and to focus further international effort in this area. Major workshops are also planned overseas.

Materials, Tools and Methods

Subject Domain

Robotics, Visual Art

Artificial Life, Artificial Intelligence Specific Methods - Evolutionary and Adaptive Systems

Tools

Computers Systems, Custom Robotics

Method Categories

Practice-led Research

Staff and Advisors

Principal Staff

• Professor Margaret Boden, Informatics, University of Sussex.
• Professor Phil Husbands, Informatics, University of Sussex.
• Dr Charlie Gere, Institute for Cultural Research, Lancaster University.
• Paul Brown (artist), Informatics, University of Sussex.
• Professor Ernest Edmonds, University of Technology, Sydney, Australia.

Other staff members

• Jon Bird, Informatics, University of Sussex.
• Dustin Stokes, Informatics, University of Sussex.
• Bill Bigge, Informatics, University of Sussex.
• Doctoral Candidate Simone Gristwood, Institute for Cultural Research, Lancaster University.

External Expertise

• Professor Sue Gollifer, Brighton University.
• Dr. Mitchell Whitelaw, University of Canberra, Australia.
• Dr. Jon McCormack, Monash University, Australia.
• Dr. Rob Saunders, University of Sydney, Australia.
• Professor Tony Longson, CalState University, USA.
• Linc Smith, Informatics, University of Sussex.