"A mind that is stretched to a new idea,
never returns to its original dimension"Oliver Wendell Holmes (1809-1894)
"In 1968, George Land gave 1,600 5-year-olds a creativity test used by NASA to select innovative engineers and scientists. He then re-tested the same children at ages 10 and 15. The test results were staggering! 98% at age 5 registered genius level creativity, 30% at 10 year and 12% at 15 years of age. The same test given to 280,000 adults placed their genius level creativity at only 2% ! In his book 'Breakpoint and Beyond', co-authored by Beth Jarman, Land concluded that non-creative behavior is learned".
Breakthroughs in science or art are rare. Most of the time we see variations on a theme, work that, whilst new, is easily classified as part of a school or discipline. Here we will look into the meaning of innovation and the steps that may be involved in creating an artificial innovative system, steps that should also be applicable to creative novelty in humans.
It is often thought that the essence of creativity requires that the entity be conscious, thus able to deliberate on a particular problem. If this is essential then we need to consider how we can make our artificial system conscious, and that problem itself presents a major task. Here we will try to outline creativity in a way that does not need consciousness, but will also look into what being conscious brings to the question, and hopefully show that this also has a valuable role in creative thinking.
Before we look into the new, let us try to outline the old, the position we start from. The way that our current ideas are stored in the mind is unknown in detail, but we can usefully model this using variants of the complexity science idea of attractors. Here, a concept represents an attractor, and those states of mind that cause us to bring that concept to consciousness form the basin of attraction for that concept. Other competing attractor basins meet in idea space, and the one that wins our attention relates to the relative strength of the components making up the attractor (the neural weighting) in relation to its neighbours.
It seems clear that, given this sort of arrangement, any particular input to our senses will need to be matched to an available concept, if we are to recognize it. Thus the variety and extent of our world view must relate to the number of attractors present. This limit sets the current boundary of our thought, we cannot directly understand (recognize) stimuli that are unclassifiable within this autopoietic matrix of available mental ideas.
Before we can deal with an unknown datum we need to establish a suitable new concept for it. This relates to creating a distinct pattern distinguishing the new input from those seen before. In most cases this pattern will have several elements, common features that already relate to existing concepts, yet with some novel aspects also, giving a mismatch or ambiguity when compared to current concepts, and generating in us a sense of unease.
To create new attractors we must either use new areas of brain space or partition those already in use. If the latter then this would weaken the old attractors, lowering the walls. Yet it seems that we can retain our old ideas and just add on extra levels of subdivision (e.g. we learn the word 'tree', add lower levels of 'deciduous' and 'evergreen', divide these into say 'oak' and 'acer', add specific varieties and so on). This suggests additions rather than rebuilds, and the possibility that the additional space used is co-existent with the current attractor - in other words that a specific network can support multiple simultaneous, overlapping or re-entrant attractors.
It is often said that genius is 10% inspiration and 90% perspiration. In our context here we can say that the inspiration aspect relates to creating a new framework of attractors, and the perspiration is the effort required to define the attractors within this framework. This latter task is the repetitive strengthening of the links, the conversion from a weak boundary to a strong one, wall building.
We can perhaps view the inspiration part as a form of random mutation (temporary connections) followed by selection. Given an unresolved input, we activate various areas in various ways until a stable state is found - we design an attractor by trial and error. As this persists over time the neural weights will start to etch themselves a niche by the Hebb learning process, and this will be made permanent later in sleep (thought to be the housekeeping state - dreams are the testing of associations to be finalised perhaps).
Given a set of variables (which we can call parameters), we can arrange to change these in turn to generate a new pattern, similar in most respects to the old one. The totality of these possible variations defines the extent of the concept, the range of possible inputs that can be forced to fit. In this way, a concept such as 'table' can have variable parameters that define number of legs, height, shape etc., plus relatively few fixed parameters that are crucial to the concept (e.g. flat top).
This ability relates to the basin of attraction of the concept, defined in a multidimensional space (rather than the small dimensions usually used to illustrate attractors). By allowing many dimensions we can use a probability view to categorize the attractor, within a fuzzy logic mathematical framework. We can define the strength of an attractor as the fuzzy normalized sum of the various dimensions and relate this to the other attractors possible, to decide either the winner (strongest) or all matching categories (exceeding a threshold value).
Some features of our search for creativity seem forced from our experience of it:
- Interconnection Flexibility - creativity is change, constraints prevent it.
- Tolerance of Mistakes - trial without error isn't possible
- Parallel Operation - time precludes any serial search of state space
- Selection - a way of evaluating a solution and stopping the search
- Analogies - using existing patterns to apply to other areas.
- Randomness - ability to add a little chaos and see what happens
These features imply a system based not on certainty but on probability. Our concepts are often ambiguous and initially are weak, only gaining strength and definition over time. This would be expected if they began analogue in nature (continuous variables with no boundary between them) and gradually became more digital (step boundaries appear as the sigmoid transitions of the neural net harden and sharpen with learning and exercise of the concept).
Little is yet known about multidimensional attractors, so we may be premature in trying to use them to implement creativity. Nethertheless we know from attractor studies that networks can support multiple attractors, and that small changes can vastly alter the attractor structure. To generate new concepts we need a method that can classify various inputs automatically, and can then adjust the classifications automatically as greater discrimination becomes possible, adding new levels.
Prototypes of such networks are available in the Kohonen Feature Maps, a form of neural network, yet these seem too simplistic so far to perform this sub-classification, generating the lower level details that come so easily to children as they learn by experience. Yet this area holds promise for the task of generating creativity, since what is classified can depend critically on initial seeding, the order in which samples are presented. A way of allowing multiple internal options to be all tried in turn may well be possible in more complex networks, and this needs further investigation.
Consciousness is a serial mode of thought, it is like a searchlight that focuses on some part of pre-existing brain reality. But whilst it is possible to be conscious in the sense of aware without any subject (in meditation for example), as soon as we try to think we use the concepts (e.g. words, categories) already available. We must therefore regard consciousness as secondary to creativity. Using logic to derive new consequences amounts to little more than permutations of existing concepts, we can generate new links between concepts (associations) perhaps, but no conceptual novelty.
Given this, we must ask why we have consciousness ? As a teaching aid it is invaluable. We are unable to communicate very well in a parallel mode, body language being notoriously unreliable as a method of conveying complex ideas and useless for abstract ones. This mode thus conveys inadequate and incomplete information, but a finding from cybernetics shows that given inadequate information we can sometimes restore predictability by adding past data, aspects of the system history. Parallel states must exist in the eternal present, there is no past or future available. However a serial mode, given past memories, enables us then to consider future plans. To plan means trying to order our goal states, to impose a timeline on our actions, a prioritisation, a prediction from past behaviour. This has been so valuable in maintaining our civilisations (society and technology both) that it seems quite adequate as a justification for the evolution of a serial state - and this ability is, it seems, consciousness itself.
It would seem that creation does not need conscious attention, and this seems to match the data from creative people. Genius, when it acts, comes out of the blue. Even when a problem has been consciously studied for many years, the actual solution seems to be the crystallization of mostly random attempts at a solution, and regularly occurs in sleep - at the time we suggest new connections are laid down and new associations are activated in dreams. In other words it is associated with attractor formation.
The role of consciousness seems more to do with directions, with the choice as to which of life's many paths we wish to follow. It adds goal directed behaviour to the randomness of lesser animals. By doing this we have clear fitness advantages, we can pre-test options and choose not to follow harmful ones. We avoid the 'failure of the unfit' stage of natural selection, a stage that culls in one dimension and loses as a result all the acquired skills in many others. Consciousness allows efficiency in the search through state space, and this efficiency itself helps us to converge on the creative areas of possibility and leave aside what are often (literally) dead ends...