This paper reflects the research and thoughts of a student at the time the paper was written for a course at Bryn Mawr College. Like other materials on Serendip, it is not intended to be "authoritative" but rather to help others further develop their own explorations. Web links were active as of the time the paper was posted but are not updated.
2006 First Web Paper
Before Stephen Thaler perhaps it was the words of Poe, Dickenson, or Delillo that kept men and women shivering in their insomnia as they pondered the questions of identity, questions of freedom, human creativity, and death; but now a new set of words has the potential to strike such fear in the hearts of humans—the dying words of a hallucinating computer program. In 1989 Thaler took a giant step towards the possibility of artificial intelligence as he witnessed a machine begin to take on aspects of human creativity. On Christmas Eve Thaler was wondering if he could learn anything about near death experiences by observing the 'death' of a computer program, so he designed a simple program to spit out the words of familiar Christmas Carols. This program was not extraordinary in any way—it utilized a switch-based, Boolian-logic configuration where every action follows deterministically from other actions. Its structure prohibited it from deviating from the information or rules it had been given, but Thaler made things interesting when he introduced a "Grim Reaper" program into the Christmas Carol program. This second program was designed to gradually and systematically destroy the connective switches by disassembling the interlocking patterns that defined the Caroling program's integral organization. As the program 'died' it began to run through all the combinations of songs that it knew—it's life "flashed before its eyes." This phenomenon was not the end of the discoveries however as the program's destruction became a catalyst for unexpected creative processes:
"...it began to hallucinate. The network wove its remaining strands of memory together, producing what someone else might interpret as damaged memories, but what Thaler recognized as new ideas. In its death spiral, the program dreamed up new carols, each created from shards of its shattered memories. "'Its last dying gasp was, 'All men go to good earth in one eternal silent night,'" Thaler said" (7).
Although the meaning of this sentence is eerily self-reflective, the significance of the program's last statement is not in the words' natural meaning. It is rather the phenomenon of an un-programmed Christmas Carol line that was created in the liminal space between established memory and random deviation. Human consciousness and creativity have remained sacred ideas even into the twenty-first century, but this unexpected output exemplifies how a controlled amount of chaos in a parameterized system can create results that are neither irrelevant nor expected. Thaler saw the potential for computer systems to literally think outside of boxes that had originally been closed, and it was this basic principle that inspired the invention of a creativity machine (4). A creativity machine is one whose architecture is changing and adapting in very specific ways to produce more imaginative outputs than closed computer programs are capable. This does not imply that a computer system can now have an imagination, but that by disrupting the established weights and thresholds of certain internal switches with fractal algorithms, a program will create new information that is born from probabilistic rules rather than deterministic ones. Before a set group of interactions might always output five ones and five zeros, but by imposing the very specific fractal equation that Thaler began to use, the output was not restricted to the same gate openings and closings all the time (6). Probabilistic programs force the system to create irrational results so that they can be figured into the map of conceptual space that that a program explores. These conceptual spaces can be anything from songwriting, durability of plastics, mixed drink combinations, or chess.
When an 'un-creative' computer chess program, for example, sets out to make the best move possible in a certain situation, it begins to sort through its memory. This memory contains the objectives and rules of the game and also contains a rich and diverse history of millions of grandmaster games. Obviously, if a computer program were given enough time it could run through every combination of possible moves, but this type of processing is inefficient and lacks the type of insight that is at work in human thinking. In order to emulate the human's ability to remember and learn, the un-creative chess program will refer to large sets of games and chess knowledge in its memory, but the effectiveness of the chess program still depends on set patterns that it has seen work before based on games that have already been played. Despite a combination of thousands of chess styles and situation-based information, its decisions are still limited to its past and its improvement depends upon new memory input from the outside. As the game progresses the computer becomes more likely to miss a move because it is less probable that it will perfectly 'recognize' a situation from its memory bank.
A chess program driven by the model of a creative machine acts differently because it does not require input in order to learn. Similarly to the un-creative program, the creative program has a wealth of knowledge that guides and influences its decisions, but it also incorporates low-level random perturbances that will cause the program to break away from its historical patterns and 'test out' decisions that go against its history. The results of this experiment will then be re-introduced to the program's history so that it knows why something does not work that it usually ignores (2). By controlling the perturbance level, the program will be more likely to deviate when it has little background evidence supporting a decision. It turns out that by trying out moves that have no basis in memory a program will create its own discovery drive where more unfamiliar situations are given more flexibility which abets learning rather than deterministic ignorance (6).
Chess is a unique conceptual space because chess' objectives exist independently from human revaluation, and therefore a creativity machine can verify which creative ideas are useful according to a set of rules. According to sources Thaler's creativity machine has figured out 11, 000 new potential hooks for pop songs and 15, 000 mixed drink combinations, but whether or not these ideas are good is a function of human validation (2). Creativity is the combining together of old ideas together to form new ones, but these ideas must be useful or interesting in some way. Combination-theory is the idea that the most improbable combination of old ideas are also the most creative idea, but this theory fails to recognize the complexity behind human evaluation (1).
Thaler's creativity machine will continue to prove most valuable when its purposes can be objectified and quantified, but when it seeks to make drawings, music, and literature a program's ability to learn and adapt at very fast rates is put into check by human variability. Even as a creative machine begins to learn how to create songs or food that people like, these opinions may change as tastes and fads tend to do. The machine's creativity can never get too far ahead of itself because humans will adapt to the machine's creations and factor this into their likes and dislikes. They will begin to reject certain creative ideas that they would have liked five years ago to pursue a niche that is inherently more mystifying and unsuspected. Where the objectives are clear, the creative machine can continue a process of hyper-evolution. Already the human imagination can see how scary the machine's imagination could be because it will improve upon its own methods the more it works on a task. The implications of a creativity machine whose objectives are dangerous do not need to be unpacked. When the words "All good men go to good earth in one eternal silent night" echo, the poignant fear is not about threatened identity or a de-mystification of the human spirit, but what resonates profoundly is the fear that machines with creative technology could be set to many horrible, destructive tasks, and they could learn to complete them with more mastery than was ever imaginable.
1)Creativity and Unpredictability
2)The Creativity Machine Paradigm
3)The Biology of a Creative Mind
4)Artificial Neural Networks
7)Thaler's NDE Experiment
| Course Home | Serendip Home |