Abstract or Die:

Life, Artificial Life and (v)organisms
 

Ron Cottam, Willy Ranson & Roger Vounckx

Abstract
 

What is “intelligence”?

David Fogel has suggested that it is “the ability of a system to adapt its behaviour to meet its goals in a range of environments”. We agree with him that this is the best currently available definition.

The difficulty, however, with this and other “top-down” natural language definitions, appears when we try to identify the essentials which will enable us to deconstruct it in the time-honored “reductive” manner and then build an intelligent system “bottom-up”. What is “ability”? What is “a system”? What is “behavior”? What are “goals”? What are “environments”? Unfortunately, the definition does not readily lend itself to reductive linguistic processing, as the elements of the complete expression are inter-dependent: we cannot establish definitions of the words ability, system, behaviour, goals and environments in isolation and then extract the complete expression’s meaning by simply combining them. Bruce Edmonds has presented a related argument as a tentative definition of “complexity”, namely that it is “that property of a language expression which makes it difficult to formulate its overall behavior even when given almost complete information about its atomic components and their inter-relations."

The situation is no better if we resort to formal language as our descriptive mode: in fact, it is a good deal worse! Even a humble Boolean AND expression suffers from this irreversibility: we can derive the single output from multiple inputs, but not the complete inputs from the output. Science itself is similarly flawed in its relationship to nature, in its use of rationalities which presuppose elemental interchangeability (e.g. in the presupposition if A = B + C that B + C = A).

Nature appears to use a different approach, where the possibility of contradiction between representations at different localities is provided by relativity, but even so global coherence is maintained. Local representations which are at odds with the global “picture” are destroyed in favor of their local-globally coherent companions: witness the quantum-mechanical collapse of multiply-superimposed hypothetical representations into one “real” conclusion.

We would do well to take account of the difference between our abstract formulations, which impose local-global separations, and the fundamentally different formulations of nature. The present paper examines this difference as part of a route towards building intelligent systems. We will refer to abstract logic and natural logic to differentiate between our usual (artificial) approaches and those taken by nature. The main target of the paper’s considerations is the meaning of the word “goals” which appears in David Fogel’s definition.

Our starting point is the observation that computers as we know them do not have goals ! Their formal style of logic precludes any integration of their data, and even the individual bits which make up the representation of a single number are formally separate and devoid of meaning in the global context. Computers as we know them are not integrated systems on their own, although they usually appear to be so because we inadvertently include ourselves in their operation! The first criterion for a computer is that it must be capable of doing nothing - otherwise how would we know it is doing what we want it to do? Notice that this removes all autonomy from a computer. Situations where a computer appears autonomous are simply those times when the (formal) complication of their operation is too great for us to comprehend in detail and our (formal) simplified description of their operation is incomplete.

Let us rather look at biologically-derived intelligence as a prototype. Biological information processing is integrated: note the singularity of our individual consciousness. However, now we apparently find a contradiction. Our bodies clearly operate under the constraints of natural logic, but our minds do not appear to do so. Survival demands that we relate to our surroundings through simplified representations, as a way of reducing information processing time, and we consciously do so using abstract logic. But how does this come about?

The situation becomes a little clearer if we look at the development of both biological organisms and computers from the bottom up.

Newly-born animals have instincts which enable them to survive and learn. These are built up from conception to birth as a pre-structuring of neural connections. A high degree of plasticity remains, however, enabling the animal not only to build on these instincts, but to replace them in many cases with environmentally-derived variants. Computers also have “instincts”. Their pre-programming is at an abstract level when they leave the factory, but initially it lies in the physics of device operation – in natural logic. The precursor of future “goal” implementation in both cases lies in natural logic.

If both organism and computer start off from natural logic and develop towards the use of abstract logic, where is the difference between them?

The developmental progression from natural to abstract logic in an organism is continuous. Although there are strong environmental and external directive effects, the development itself is wholly internal: an organism “does it itself”. Natural and abstract logics are integrated within the organism, and ultimately the abstract can only be divorced from the natural by internal decision (quasi-autonomously or not). Within a computer there is no integration of natural and abstract logics: their segregation is the first rule of computer design, to take decision-making out of the computer’s “hands” and keep it for ourselves.

So, if we are to build intelligent goal-driven systems, how should we do it? The current approach of taking a system which by its very nature includes us in its operation seems rather strange. Artificial life does not exist: it is nothing other than a functional simulation of life, projected into an abstractly-operating embodiment by us, the godlike designers. We must learn to create virtual organisms, whose goals are generated internally through the interplay of natural logic, abstract logic and environmental influence. But do we really want to create autonomous (v)organisms, or would we prefer that their goals are in any case related to our own desires? Nature has developed the quasi-integrated role of parent to its child. Maybe this is the position we should seek if we wish to expand our capabilities but still retain control of them.
______________________________________________
______________________________________________