What is Artificial Intelligence? Consider this excerpt from Tom Holt’s novel “Almost Human”:
“The robot hesitated, while the Appeal Court of its mind pondered the nuances of the Laws of Robotics. Eventually they handed down a decision stating that the overriding law which supervened all others was that no robot shall say anything, no matter how true, that will inevitably earn it a smack in the mouth with a 5/8” Whitworth spanner. “Sure thing, boss.” it said”
Is “artificial intelligence” then the point at which a machine’s ability to think can override programming, or is it the lesser test of applying mere rules/programming to provide answers to a variety of problems?
At present our best efforts to create artificial intelligence have produced little more than the amazing, human-like ability of a computer program to understand that the letter Y means “yes” and the letter N means “no”. This may seen a little pragmatic however this is ironically not far from the truth of the situation.
If we forgo any preconceptions as to the semantics applied to the word “intelligence” with respect to a technological form as apposed to a human, it becomes apparent that this is nothing akin to using the word “flying” to describe both birds (biological) and aircraft (technological) forms of heaver than air flight.
The field of study into the possibility of artificial intelligence necessarily assumes that it is possible to synthesise something that satisfies the conditions for “intelligence”, not everybody accepts the current presumptions made about human cogitation and deductive system which from time to time are ridiculed by critics whom argue on a variety of grounds that artificial intelligence is doomed to failure. A good example of such a philosophy is known as Tesler’s law, which defines artificial intelligence as “that which machines cannot do” which implies that any possibility of an artificial intelligence is impossible and that concepts and attributes such as intuition are abilities that are unique to human.
At this point I would like to draw the distinction between artificial intelligence as inferred in the hypothetical procedures based on interrogation in the Turing test, which in effect is merely a test of the systems ability to imitate human-scale performance, through programming, and as such is a simulation of the desired effect on the one hand, and a system’s intellectual capacity to learn, manage, and manipulate natural language or exhibit free will; etcetera on the other.
For example using the Turing test as a model, if a computer exhibited the ability to take decision that if made by a human would indicate the use of intuition, the system would pass due to the fact that it is not a test of human-scale performance, but is simply testing its ability to react to a process of pure stimulus-response replies to input (not action of its own accord).
The study of artificial intelligence, is a sub-field of computer science primarily concerned with the goal of introducing human-scale performance that is totally indistinguishable from a human’s concepts of symbolic inference (the derivation of new facts from known facts) and symbolic knowledge representation for use in introducing the ability to make inferences into programmable systems.
An example of inference is, given that all men are mortal and that Socrates is a man, it is a trivial step to infer that Socrates is mortal. Humans can express these concepts symbolically as this is a basic part of human reasoning; in this manner artificial intelligence can be seen as an attempt to model aspects of human thought and this is the underlying approach to artificial intelligence research.
If for the sake of argument we were to assume that ‘intelligent’ processes are reducible to a computational system of binary representation, then the general consensus amongst artificial intelligence authorities that there is nothing fundamental about computers that could potentially prevent them from eventually behaving in such a way as to simulate human reasoning is logical. However this necessarily assumes that practical everyday reasoning is not the optimum form of human cogitation and deductive, mathematical, and logical reasoning is all that is required to be ‘intelligent’.
If however we assume for the sake of argument that intelligence is not a mutually exclusive entity, and is rather the convergence of characteristics other than logical deduction or mathematical reasoning, such as emotional characteristics that together play a collective role in thought, decision making and creativity, then the greatest part of human intelligence is not computational, and consequently it is not precise and the development of artificial intelligence based the current model of pure binary logic would potentially result in only precise forms of human thought being simulated.
A great deal of research has been done on inference mechanisms and neural or nerve networks which has ironically been of more use in learning about human intelligence through the process of simulating intelligence in the machine, rather that the other way around. Such research has however produced an uncertainty about our own thought processes.
Such concepts require that we clarify a number of interesting anomalies, the most fundamental of which is that we have no adequate theories to explain the nature or origins of phenomena such as the mind, of consciousness, nor of intelligence This would require understanding of the relationship between the essence being and the brain where at present we simply have no true theories.
For the time being, although computers can solve with ease the most difficult mathematical problems, there are currently many problems that humans solve instinctively which are unresolvable artificially, where advanced heuristic rules and conceptual networks have collapsed due to the amount of contextual information and common sense knowledge they seem to require, such as natural language processing, or even “What clothes shall I wear?”.
It is the level of shared understandings required in our most inconsequential forms of social interaction which necessarily require that individuals assume complicated shared knowledge that is too complex for even the must sophisticated forms of artificial intelligence as conceived to date, in which propositions are either true or false and premises must follow deductively.
We need to give computers the ability to process imprecise concepts such as high, low, hot, warm, or very near, by substituting precise rule-like logically deductive structures of knowledge and mathematical measures for an approximation.
At the very least in order to program machines to simulate human mental processes, one needs to understand and clarify, how these processes function, therefore our attempts to replicate those processes that will spawn machines capable of doing any work that a man can do, can only really start when we understand the processes themselves.
The questions remain, “how can you create intelligence when there is no definition for what it is?” and “How would you know you had done it?” Faced with such questions that effectively invalidates artificial intelligence as a science due to it’s as yet unprovable assumptions, the fie Turing Test was devised. However this seems to indicate that machines can only become more intelligent as they become better able to simulate a single human’s reasoning ability.
It may be we should be setting our sights lower – and trying to determine the simplest form of animal or insect life which demonstrates intelligence, and working up from there. The mere process of identifying what is intelligent, however primitive, will help set the parameters for what we are trying to achieve.
Fore example. Is the ability to hold a conversation a true test of intelligence, or merely of human intelligence – a possibly irrelevant side issue? This has been the reality of the Turing Test since 1950, but has it lead us down a blind alley? Consider a hypothetical race of aliens who communicate by extra sensory perception, the fact they have no need for speech will not make them less intelligent, probably more so because less of their brain will be being used up in wasteful processes.
We can possibly take this further, and state that humankind needs speech to give its otherwise chaotic thought processes some order, and therefore intelligence, whilst a computer’s more logical structure obviates that need, as a machine intelligence is by nature computational, and precise and we should be concentrating on what we want that AI to achieve on its own merits, not restrict it to mimicking our own inadequate characteristics, but rather an approach that is not a result of clever programming, but where the AI can initiate its own actions, not just reactions, and can override, not just adjust, its programming.
Perversely, an expert system called the CYC project may almost by chance deliver the closest approximation to human reason, that has yet been devised, by its realisation of the parallels between the internet and the distributed connections within the human brain.
Because the knowledge stored on the internet is so diverse, and the product of so many different levels of human intelligence and experience, we may have in fact already achieved the most difficult part. All we need now is the machine’s ability to organise, access, and process that ‘consciousness’, so that the answer it gives to any problem is always contextually relevant, and we have come very close to our Artificial Intelligence. At the present time it seems that the development will remain stalemated until single machines have at present undreamed of computational and memory attributes.
Notwithstanding that this is a cheat, because firstly in general, humans themselves have to learn to think more like the expert machine, rather than the opposite; And even so, it is the continuing, apparently irrelevant, input by humans across the world which will keep this relevant, but that is very little different from the stream of consciousness we have all experienced since birth which informs our own daily decision making.
What is then left is the question of creativity – the ability to act, not just react, the ability to initiate, not just follow orders, the ability to self improve and, taking us back to where we started, the ability to lie where circumstances dictate that the truth is not enough.