AI.md 16 KB
AI & Machine Learning
Programming
Fuzzy
AI and Machine Learning techniques might be better thought of as a form of "fuzzy" programming, whereby similar techniques are used, but with a qualitatively different expectation regarding the output.
Traditional Approaches
For comparison, let's examine the traditional approach to programming.
Lambda Calclulus
In traditional programming practice, the program, and the algorithm(s) encapsulated by the program, take certain inputs and output information which is expected to be correct or incorrect based on it expressing a high degree of accuracy, even a form of absolute accuracy. Case in point, a program might take numerical input parameters, perform a direct arithmetic operation on these parameters, and output the result of the operation. With such a program, its validity is based entirely on whether or not the output proves that the right operation was performed and that it was correct.
Verification
Verification can be performed relatively easily, as it is just as matter of confirming the operation by controlling the values of the input parameters, utilizing a secondary process for determing the expected result (such as doing math on pen and paper), and then running the program with those same input values and seeing if the program outputs the same value as that which was the observed result from the secondary process. Though arithmetic might seem like too simple of a comparison to be relevant to all programs, it is the right one, as all computed constructs, their behaviour and the expressions in which they participate can be abstracted to their numerical forms.
Modern Approaches
Now let's take a look at the approaches utilized by Machine Learning and Artificial Intelligence.
Verification limits
When using Machine Learning techniques, one cannot perform the same type of arithmetic verification. Though the Machine Learning techniques will make use of algorithms which may be similar, or the same, as those found in traditional computer programs, there is usually not any direct perspective on the workflow of the operational pattern wherein the inputs are taken through a neatly delineated lambda computation which outputs a single output value that can be scrutinized.
Inherent Fuzziness
Instead, what we have is a process which is often referred to as "fuzzy", in that no single output will necessarily reveal an answer which can serve an a-priori purpose. Rather, a massive repetition of the procedure is sought, possibly with the same inputs, or with a set of inputs.
Limitation, Modality, or Different Beast?
With enough repetition, increasingly accurate estimates can be made as to the outcome of a given data set without necessarily performing the entire computation. That is to say, the Machine Learning algorithms can take a snapshot of the computation without having completed it, and produce estimates about each stage of the computational procedure, including the outcome.
Is this Artifical Intelligence?
Perhaps. Perhaps not.
For us to know the answers to these questions, we need to define what it is to be intelligent.
Intelligence has been defined in many ways: the capacity for logic, understanding, self-awareness, learning, emotional knowledge, reasoning, planning, creativity, critical thinking, and problem solving.
Some arguments can be made to conflate, or at least infer an overlap, between intelligence and IQ. This is a reasonable vector of consideration and it is almost impossible to consider one from the context of a human being without also thinking of the other. Though no semantically complete representation of intelligence or cognition as expressed through human activity has been demonstrated, or even proposed in a capacity that can be scientifically scrutinized and found to be a viable mode of quantification, it is currently the most agnostic approach we've yet found and, as such, it will be considered in this essay.
Capacity for logic, understanding, self-awareness, learning, emotional knowledge (?), reasoning, application of logic, planning, perceiving, creativity, organizing, critical thinking, resolving, problem solving, ordering,
Acknowledging, verifying, naming, remarking, identifying, contextualizing, structuring, the ability to acquire and apply knowledge and skills. late Middle English: via Old French from Latin intelligentia, from intelligere ‘understand’ (see intelligent). problem-solving ability, spatial manipulation, and language skilled use of reason the ability to apply knowledge to manipulate one's environment or to think abstractly as measured by objective criteria (such as tests)
Definitions (Intelligence)
There are two primary contexts for the definition of intelligence which are employed by humans, with one consisting of a set of definitions pertaining to use of knowledge, and the alternate definition focusing on a domain of knowledge or information. We will be focusing on the former, as the latter would suggest that all computation is intelligence. Though a case can be made that all information is intelligence, that is beyond the scope of our essay and ventures too far into the metaphysical.
ASIDE: Domain of knowledge / Information
In the previous paragraph it is alleged that in order to posit that intelligence and "domains of knowledge" or "information" are synonymous with one another. What follows is a deconstruction of this argument.
What exactly is a "domain of knowledge"? A domain of knowledge is the conception of encapsulating an expanse of information through arbitrary demarcation. As all known information, as communicated amongst humans, is necessarily structured by a human capacity, it follows that the very concept of a domain of knowledge is a purely human one. If it is possible to encapsulate an expanse of information by non-human means, we are limited in that we cannot make such an argument without also conceiving of it within the human frame. As such, it must be assumed that a domain of knowledge is one which is necessarily delineated by human thinking.
The other alluded proposition put forward as a comparable conception is that of information as a whole. In its most complete form, information as a whole must consist of any and all matter which exists physically in the Universe. It might be possible to declare a form of information which is separate from the physical aspects of the Universe, but its communication is bound to physical properties within said Universe. Another aspect of this is that there is no explicative mechanism which illustrates the manner by which human thought is represented physically, except by incompletely understood indication patterns, such as firing of neurons and other similar artifacts of analysis which consider the nervous system, the brain and the mind in any quantifiable form.
How does it follow that either of these beliefs are bound to the statement that "all computation is intelligence"? Quite simply, to produce a valence describing any information, in any form, requires, in tandem, as an essential component, or even as the component in its absolution, that a quantification is being produced. That is to say, the description is in and of itself a form of computation, and without a description, a production of valence, a communication about said information, or even by having a conception of thought which is never communicated from one being to another, there is no evidence that the information in question even exists.
Therefore, it is necessary to infer a supposition that a belief that "information" is intelligence is interchangable with the supposition that all computation is intelligence.
Intelligence
Though there are semantic variations for the definition of intelligence, their underlying similarities allow us to genericize our approach to evaluating whether or not the toolsets utilized within the realm of AI are a cause for the genuine manifestation of intelligence.
Let us explore some of these definitions and then proceed through the methodology of factor analysis in order to remove unnecessary components.
Capacity for logic
One's capacity for logic is the ability and extent by which an entity can observe the environment, identify its components, contextualize, possess awareness of the relationships of these components within the environment/system, evaluate.
One could make the case that logic itself can exist outside of a capacity, but again we are limited in that all conceptions of the universe, including the interactions of its components, as described by physics, are human conceptions. Thus, the concept of an interaction demonstrative of logic, or reason in any form, is a necessarily human one.
Relationships of Components within a System
In identifying a given system, we must, through our examination, acknowledge a few properties before concluding that the system does indeed exist.
Overview to properties of a system: A system should be expected to exhibit the following properties: encapsulation, state, transport/transformation, output,
TODO: verify the following - output, transformation -> see Dr. Shiva's summary on systems and system science
Qualifying Systems
1. Encapsulation
If a system exists, it would be redundant if we were not able to demarcate what separates this given system from all other things. Though there can be overlap to the extent that the system can be part of another system, or that a given component within this system may also be relevant, enumerated or bearing a contextual relationship to another system, there must be a reasonable predicate to compel the declaration that the proposed system exists.
There's not necessarily a singular factor by which the pronouncement of delineating this system must occur. There are, however, several factors which naturally yield more obvious paths by which to weight the predicate:
a) Spatial / Location
The location predicate is qualified under a presumption that a physical space exists, and that there are degrees of movement possible along one or multiple dimensions. This is most naturally observable to a biological organism which moves through time and space, and which must do so to continue its existence. Even in the case of an organism which does not bear the possibility of ambulation, movement still does occur in order to sustain production and transport of energy, as is the case with breathing and metabolic function. Metabolically viable particles mobilize through mechanical movement, compelled through characteristics which, in our Universe, can be understood as being electromagnetic, interact with other materials and are processed to yield substrative components possessing their own properties, or a varying weight of a given property, such as to differentiate them from their originating particle. Through continuation of this behavior, it can be observed that, even with an organism that is stationary, its continued existence demonstrates some form of expression of movement through space; a change of location, in some respect.
The prospect of determinable locality, insofar that a delinated locality can be produced, manifests a factor by which the encapsulation of this system can be qualified.
b) State
If a system is to exist, it must have some set of components contained within it which can be identified and described. These compose the system's state. This will be elaborated on later in this essay.
c) Behaviour
An identifiable system must contain and/or exhibit behaviour. System behaviour will be elaborated on later in this essay.
2. State
The state of a system is the identification of all components which exist or bear affects within the bounds of that system. These components can be thought of as being members of the system, and may exhibit behaviours ranging from eliciting affects to other members, to simply bearing a quantifiable and determinable valence. There could be a potential for these members, behaviours and valences to exhibit change across a dimension of time, but this is not requisite, as the said system could be a frozen snapshot that would never change, regardless of the existence of time. That time is observed to exist in our Universe, however, should reasonably cause one to assume that the attribute of time exists either within the system, or as a relevant factor to the observer of the system.
The whole of the members and their determinable valences composes the state of the system.
3. Behaviour
A system's behaviour can be evaluated through 2 possible mediums. These mediums, however, should intrinsically bear a composite-type relationship. That is to say, one medium bears consequence to the evaluation of the other medium. That isn't to say that the semantics of the behaviours are identical to one another, and only differentiated by scale, but that there should be a manner of reasoning to describe the analoguous relationship such that it becomes feasible to hypothesize about the mechanism through which one influences the other, and that this mechanism is quantifiable.
i. Overall
TODO: I'm unsure of the declarations in the following paragraph. In particular, I am not convinced of the veracity of making the statement that we, as observers making declarations of a system, must also operate within systems which are separate from the system in question. That is to say, perhaps it is more accurate to say that our ability to make declarations about a system to which we are a member, is contingent on our membership to a subsystem of that system. This may suffice to assert that we are able to make declarations of a system to which we are members.
That a system exists and bears a behaviour is relevant to the observer of the said system. As we are entities making pronouncements of systems, it follows that we, as observers, do not exist solely and exclusively within the confines of that system. That is to say, that we are able to remark and identify the existence of this system compels the belief that our own existence and behaviour is not limited solely to the constructs of that system. The evaluation of systems, for example, can be seen as a being either a separate system, or a subsystem to the system in question, and one which we are partaking in, regardless of whether or not the identified system.
Nevertheless, it can be reasoned that a system has a deducible behaviour.
TODO: Is it possible for a system to exhibit behaviour which is separate from the behaviours of its members. Would a system's behaviour not necessarily be the culmination of its members behaviours?
ii. Member behaviour
It is far more common to evaluate a system's behaviour based on the behaviours of its given members. Though it follows that the behaviours of these members should produce, in part or whole, the behaviour of the overall behaviour of the system, it is not necessarily a requirement for one to consolidate the implications of member behaviours to overall system behaviour.
Behaviour of members can be conceived as the range of operations, effects, values, and consequences which can be evaluated or inferred from the existence of these members. In short, the behaviour of the members can be proclaimed as being the behaviour of the system.
System Behaviour without members
Is it possible to envision a system behaviour which does not include members? If there are no members to the system, does the system exhibit a behaviour? The system's locality within another system could be a dynamic characteristic which can be assessed and deduced, but this depends on there being another overarching system and, as such, the behaviour described by the system's locality is actually a member behaviour of the surrounding system.
Perhaps the temporal measure of a system can be a deduced behaviour, but this is again only relevant because there are other systems from which to make a temporal determination about the system in question.
PROPERTIES:
MEMBERS: Affectors, actors, operators, receivers, transformers, transceivers, reducers, instigators, detractors, descriptors, producers, occupiers, collectors, identifiers, motivators, catalyzers, destroyers, creators,
consists, behaves, imposes, elicits, performs, expresses, acts, works, manifests, cultivates, qualifies