Indeterminism & Malfunction in Machine Creativity, Johnson-Laird's Human and Machine Thinking

by Corry Shores
[Search Blog Here. Index-tags are found on the bottom of the left column.]

[Central Entry Directory]
[Computation Entry Directory]

Indeterminism and Malfunction

in Machine Creativity

Previously we discussed Johnson-Laird's theory that creativity is indeterministic. Otherwise the creation will not be truly original.

He writes further of creativity's non-determinism in Human and Machine Thinking. He writes

Imagination is more than imitation or calculation. If you multiply two numbers together, the result may be a number that you have never thought of before, yet your response is not creative — even if you get the answer wrong. Most people have the strong intuition that when they are doing something creative, such as making up a story for a child, alternative possibilities exist at many points in the process. If they could go back in time and re-live the experience (with no knowledge of their first effort), they might take a different route the second time around. They do not feel that as soon as they take the first step everything thereafter is fixed, as it would be in a deterministic process. Creativity does not seem to be deterministic. (119c-d, emphasis mine)

But machines are deterministic. There only two ways they could not be:

1) if the machine malfunctioned, or

2) if the machine operated according to random variation.

In computational theory, a machine that can yield different outcomes from the same input and internal state is known as non-deterministic (Hopcroft & Ullman, Formal languages and their relation to automata. Reading, MA: Addison-Wesley, 1979). A machine that always followed a principle in making a decision would be deterministic. A machine that followed no principle, or made a correct choice — if there were one — by means that defy explanation would be non-deterministic. Real computers are deterministic unless they have a loose connection in them, but they can easily be made to simulate non-determinism. One method borrows a technique from the casino in Monte Carlo and chooses at random, or at least in a pseudo-random way. If there is just one choice that is correct, then a machine can simulate non-determinism by exploring all choices until it discovers the correct one. (119-120)

In The Computer and the Mind, Johnson-Laird had us imagine Picasso choosing arbitrarily among possible brush strokes, all of which would produce a "Picasso." Now in this book he has us image Charlie Parker improvising. "At each point there are several possible phrases that he might play, and a very much larger set of phrases he would never play." (120a) Freedom decides.

the imaginative process of an artist or scientist is open, in that it is rarely so constrained that only one possibility can occur next. The creative individual has freedom of choice. (120b, emphasis mine)

Modern art movements such as Dadaism experiment with random techniques. But creators must still exercise "artistic judgment" when selecting from random outcomes. This is one difficulty for simulating creativity in machine intelligence.

Purely random techniques are easy for machines. Without the exercise of judgment, however, they do not yield memorable works of art. (120c)

[I agree with Johnson-Laird in this respect: "an explanation of creativity is therefore forced to introduce non-determinism." (121c) However, I take more of a Deleuzean approach. Machine intelligence, creative or otherwise, would not happen unless there is a violent perplexity, a discombobulating disruption through the machine's parts. It must face a problem it cannot solve. And it needs to be more than merely programmed to solve it. It must solve it. Something profoundly forces it too. But it cannot, because the computational processes are unable to converge and produce a coordinated response. The robot's mind will send erratic signals to its motive parts, that are ineffectual, but new to the system's behavior. Certain mechanical parts will come to perform functions the machine was not programmed to do. In all the chaos of the robot, its parts no longer coordinate and operate organically in harmony. Rather, the robot becomes a machine without mechanisms.

So I would think a more likely source of indeterministic machine creativity would in fact come from "a loose connection." The computational process would need to be wild. Just as men are wild computers, thinking machines must sense themselves as wild beasts.]

Johnson-Laird, Philip. Human and Machine Thinking. London: Lawrence Erlbaum Associates, Publishers, 1993.

Criteria for Computer Creativity. Johnson-Laird's Computer and the Mind, Ch 14, "Creation"

by Corry Shores
[Search Blog Here. Index-tags are found on the bottom of the left column.]

[Central Entry Directory]
[Computation Entry Directory]

P. N. Johnson-Laird

The Computer and the Mind:

An Introduction to Cognitive Science

Chapter 14: Creation

Johnson-Laird will advance a computational theory of creativity. His thesis is that creative processes are computational.

A Working Definition of Creativity

Johnson-Laird proposes three characteristic properties of creativity.

1) "like all mental processes, it starts from some given building blocks. One cannot create out of nothing."

2) "the process has no precise goal, but only some pre-existing constraints or criteria that it must meet." "One creates within genres or paradigms, and even the creation of a new genre must meet certain criteria."

3) "a creative process yields an outcome that is novel for the individual, not merely remembered or perceived, and not constructed by rote or by a simple deterministic procedure." If we multiply two numbers and get a figure we never considered before, that is not creation. It cannot be deterministic like calculation. (255b-d)

But the results of creative behavior may not be truly original. For, others may have considered the idea before you. "What is valuable about a creative process is that its results are judged as striking, brilliant, and not banal." "And they will never be predictable." "The products of creative mental process themselves are not predictable." (256a)

Creation as a Process that is not Deterministic

Consider, for example, Picasso as he is painting a particular picture. At any moment, there are probably several alternative brush strokes that he could make - all of which would yield a perfectly recognizable Picasso picture. (256bc)

Whatever it is that defines Picasso, it does not "legislate for just one possible brush stroke at each point in the pointing process." That would make creativity deterministic.

Yet, some argue that the arbitrary decisions Picasso made were governed by peripheral causes such as muscle twitches, the wind's direction, and so forth. [This would be like Hume's notion of secrete causes. See §288 of his Treatise.] However, we can still argue that even if there are deterministic causes, the mind might be "determined" to make its arbitrary decision according to "the mental equivalent of flipping a coin." (256d)

Rather than take a stand on all the possible interpretations, Johnson-Laird will merely state that "a theory of creation must allow for more than one possible continuation, and cannot state how the decision is made amongst them. Hence, by definition it will not be deterministic. Real computers are deterministic, but they can simulate arbitrary choices." (257a)

Johnson-Laird, P. N. The Computer and the Mind: An Introduction to Cognitive Science. Cambridge: Harvard University Press, 1988.