Operational Definitions of Natural and Artificial Intelligence
Good Enough
Intelligence is a word we all use easily and often, but what exactly are we referring to with this word? It's a foggy notion, one of surprisingly many such words that we manage to use successfully every day, even without a crystal clear idea their meanings. We tend to not worry about that lack of clarity so much in everyday communication, because people seem to understand what we mean and that's good enough.
We often think of intelligence as a uniquely human capacity that may have some mystical, metaphysical aspect that sets us apart from the other creatures of nature. At other times we use intelligence to mean some objectively measurable quality that can be used to characterize or compare individual persons. Yet another sense of intelligence refers to some ill-defined capacity that lets us solve problems, act independently, process natural language or other such specific functions that can be approximated, emulated or replicated by machines for practical purposes. All of these views suffer from definitions that range from the uselessly vague to the uselessly narrow. While those definitions and the common-sense ideas they spring from are OK for our everyday business, they are not at all suitable for the purposes of science or engineering.
Not Good Enough
The ability to talk about intelligence in a way that everyone can be sure they are talking about precisely the same thing gets more important as intelligence becomes a practical concern in science and engineering. We hope for a standard definition that is simple, clear and intuitive. One that is both general enough to apply across the entire range of usage from casual conversation to rigorous academic analysis and specific enough to allow scientific study and the engineering of useful systems.
Useful and Meaningful Definitions
Trying to define intelligence in terms of other even more grossly introspective terms like 'thinking', 'reasoning', 'mind', 'consciousness', 'free will', 'sentience', 'soul' and so on will only continue to take us in the same circles as it has taken philosophers for millennia. An approach that shuns the speculative, top-down thinking of philosophers, many psychologists, spiritualists and even traditional AI theories may avert that artificial deadlock. Let's aim for a definition that allows valid and reliable measurement, and can thus found practical efforts to engineer functional and valid artificial intelligences, but let's also require our definition to be intuitive and grounded sufficiently in observable human and animal behavior to be accepted generally as well. Consider the following definitions.
A natural intelligence is a set of capabilities provided to a natural living organism by a nervous system.
An artificial intelligence is a set of capabilities that replicate or closely approximate a natural intelligence and is provided to an artificial machine or to a natural living organism directly or indirectly by engineered mechanisms.
How are these definitions better?
- They eliminate the baggage of mysticism and metaphysics by defining intelligence independently of consciousness, mind, soul and that entire lot of high-level ultra-abstractions.
- They ground the concept of intelligence in the intuitive and observable phenomena of ordinary chemistry, physics, physiology and measurable behavior, thus putting theorizing and experimentation on a firm scientific footing.
- They emphasize the evident integral relation of intelligence and a functioning physical body.
- They emphasize the hierarchical abstraction and integration of sensory inputs for pattern recognition, memory, induction and other demonstrable functions of nervous systems as the basic processes for concept formation, storage, manipulation, and interaction with homeostatic internal physical states.
- They allow us to see intelligence as manifest in a very broad range of animal behavior that parallels the tremendous variation in the complexity of nervous systems, yet they allow us to clearly separate the intelligent from the non-intelligent.
- They allow for analysis of the components of intelligence and provide a basis for rational study of the evolution of intelligence.
- They encourage cross-fertilization in the fields of mathematics, artificial intelligence, psychology, linguistics, physiology, developmental biology and brain science.
- They illuminate the comparative psychometrics of individuals.
- They lend intuitive and rational senses to the term intelligence as used in various non-technical contexts.
- They found a course of research that may eventually dispel the fog that ignorance, common sense, and the confusing, contradictory musings of philosophers have created to obscure the nature and functioning of the human brain.
Oh, really?
Grand claims were made above, and one must always be suspicious and skeptical in the presence of grand claims. Let's consider some possible faults of basing intelligence on the functioning of a nervous system.
"Set of abilities? C'mon, now. How vague can you get!
Good point. Substituting one vague term with another is no advance. No complete, "fully-defining set of capabilities that are provided to a natural living organism by a nervous system" has yet been identified. The important thing, though, is that these definitions allow such a list to be developed by scientific methods, as the relevant capabilities are measurable.
Indeed, simple common sense and everyday experience offer us quite a few items for the list. All we need do is examine the behaviors of animals whose nervous systems vary in complexity and compare their capabilities. Many of those capabilities are already known and can be used for empirical modeling.
Leaky at the bottom?
The very simplest element of nervous system is a simple reflex arc, in which a particular stimulus always excites exactly the same response, just like throwing an electric light switch. We have to wonder how this is different from other simple chemical and mechanical systems that exist in living things that don't have nerves, or even in non-living or quasi-living things. Without some clear and significant distinction, our use of the nervous system as a minimal threshold for intelligence would be arbitrary.
The critical distinction that we need here is the ability to do three very basic things that form the foundation of intelligent behavior:
- Convert the continuous changes in the environment to discrete units, like the bits that computers operate on
- Integrate multiple specific inputs to filter out noise and increase reliability
- Integrate over multiple sensory inputs of the first two kinds to allow coordinated behavior.
Those three tasks are the basic capabilities possessed by nervous systems that contain more than two nerve cells. Having such a non-trivial nervous system, then, is what separates the intelligent from the non-intelligent, and it does so cleanly. Here, integration refers to a group of mathematical operations or transformations that are performed on sets of multiple input values, such as summation, averaging, thresholding, and pattern detection. This means that a set of inputs, which may be distributed over space and time, is considered as a whole in the production of an output. This relatively simple capability is the basis of rapid adaptability, memory, concept formation, different types of learning and other higher functions.
Although the reflex arc is the minimal element of a nervous system, no animal has such a simple nervous system. The simplest known nervous system in a living organism is found in the hydra, a tiny aquatic animal, and its cousin, the jellyfish. The nervous system of those creatures is a decentralized net of nerves much like a fishing net. Still, even this minimal nervous system can integrate sensory input from numerous sensors of different types and control many muscle-like cells to provide the hydra with surprisingly complex behavior, including movement, capturing prey and reproduction.
Barren at the top?
The most complex nervous system that we know of is our own. Can these definitions lead us to ultimately understand what we call the mental phenomena of the human being? We can't answer that question yet, but two trends suggest that this question will either be answered eventually, or simply mooted. One trend is the accelerating progress we see in neurophysiology and functional brain anatomy. The other is the success that has been achieved in applying the artificial neural network model in electro-mechanical systems that solve complex real-world problems.
One possible outcome of advancing science is the gradual erosion of the metaphysics-tainted traditional concepts by scientific advances until ideas like mind, will and soul either become irrelevant or are given practical definitions. Another possibility is that the advance of science will reach some true limit in explanatory and predictive power. It may come to be that, when all the (scientific) possibilities have been exhausted, like Sherlock Holmes, we will have to admit that what is left dualist metaphysics), no matter how improbable, must be accepted.
Validity
Some will argue that eliminating mysticism and foggy abstract notions from the definitions is to throw the baby out with the bathwater; they will claim that 'real' intelligence or that the 'essence' of intelligence will thus be lost. The standard blameful term used for this is reductionism. For our purposes, it doesn't matter. We simply want useful definitions that will help us move forward. We will judge them on their utility alone. We will hold the definitions as valid to the extent that they allow progress in the development of artificial intelligence and in other fields where intelligence in its commonly-used senses is an important concept. It may or may not turn out that these definitions generate such a rich understanding of what we all can agree is observable intelligence or its effects that the notion of 'real intelligence' as something different may just disappear as have phlogiston and the ether.
A Spectrum of Intelligence
These definitions allow us to say with clear certainty what is intelligent and what is not. Rocks, plants, solar systems, cities and the Internet are not naturally intelligent in any degree; they have no nervous system. (We'll leave open the possibility of artificial intelligence in social organizations.) Hydra, slugs, dogs, pongids and humans are all equipped with nervous systems, and so are intelligent.
What is immediately clear, however, is that the intelligence of a slug is both much less than and much different from the intelligence of a dog, a dog's intelligence compares the same way with that of a chimp, and a chimp's intelligence is less and in some ways different than that of a human. Similarly, the intelligences of a fetus, a newborn, a young child and an adult vary in parallel with the development of each individual's nervous system.
Intelligence therefore has degrees and probably types as well. It covers a very wide spectrum of behavior that matches the equally wide range in the complexity and degree of development of nervous systems. Very simple nervous systems provide very simple capabilities and very complex nervous systems provide very complex capabilities.
One implication is that any further development of natural intelligence beyond human intelligence is necessarily tied to an evolutionary change in the organization of the human nervous system, an event that may produce a new and superior (or inferior) species. Humans have the most complex nervous system of all animals and so will sit at the pinnacle of natural intelligence until something more complex comes along. Unless we kill it first, of course.
Variations in human intelligence
Differences among individuals are fundamentally important to us, both as individuals ourselves and as a society. When choosing mates or associates in any enterprise, we prefer those who have 'better' properties and capabilities, where 'better' means 'more able to help satisfy my needs'. Human intelligence is a term that wraps up a rather vague bundle of abilities and behavior that we all accept as 'mental' abilities that are somehow associated with brain function.
The operational definitions suggested here replace the indefinable concepts of 'mind' and 'mental' with the clear concepts of nervous system and sensory integration on the input side and planning and manipulation on the output side. The extent to which we understand what the nervous system does and how it does those things is the extent to which we understand intelligence. In contrast to philosophy and most of psychology, we are rapidly expanding our understanding of the nervous system, particularly the central nervous system and brain most of all.
From our definitions, we can conclude that identical nervous systems will necessarily produce identical intelligences, but why don't all people have the same intelligence? We have the same nervous system, don't we?
Well, actually we don't all have identical nervous systems, not even identical twins. We know that the gross structure, development and chemistry of the human nervous system is determined by our genotype, as it is for all animals that have nerves, and that our peripheral nervous systems are largely identical unless damaged or genetically different. The exact configuration of our brains, however, is dependent in important ways on our body's experience during life, and most especially during the first several years. That dependence or susceptibility to influence begins at or before the fetal stage; it diminishes with maturity, but never ends entirely. Old dogs can still learn new tricks. The nervous system is also easily affected by poor nutrition, illness and injury. Sometimes in subtle ways and sometimes in very dramatic ways. This dependence of physiological development on environment is basically what makes us unique individuals and why intelligence varies even among apparently healthy, 'well-formed' individuals.
Factors such as disease, poisoning, injury, nutrition, imbalances of body chemistry, malfunctioning of other body systems, malformation and deterioration with age also affect individual intelligence through direct effect on the nervous system or its development. There is also a category of 'psychological' factors that is very important and interesting. That is the realm of classical psychology and theories of self. These psychological factors also involve nervous system states, but are special in that they additionally involve the complex internal interaction of high-level abstracted concepts and models with other body systems, such a hormone production by the endocrine system and other body chemistry. Whether we should include those states within the scope our definition of intelligence or treat them separately as modulating factors is an interesting question.
Artificial vs. natural vs. just different
It is at least very difficult to fully emulate a nervous system of non-trivial complexity by means of electronic devices, and it may even be impossible. Much work has been done with artificial neural networks, and they are a fascinating and valuable advance that both derive inspiration from neurology and inform it. Still, there are fundamental differences in the way current electronic simulations of neural networks and natural nervous systems operate and behave. Those differences suggest that any practical artificial nervous system can never aspire to more than approximation of natural intelligence. How close they can come remains to be seen. To fully reproduce intelligence in full effect, including consciousness, may require us to reproduce it in kind.
The question of whether fully emulative artificial intelligence and natural intelligence are the same 'real' thing may forever remain a philosophical one, which is to say unanswerable. Still, if confronted with an artificial intelligence that fully, and I have to emphasize the term 'fully', reproduces the behavior of a natural intelligence, I predict that most people would accept it as a natural intelligence on an emotional level in daily life and react to it as such.