Chapter XIV

Alternative Systems

Advanced alternative systems will increasingly depend upon the power of information technology and processing to achieve sophisticated integration and complex articulation with the environment. Information processing systems have made tremendous advances in the last couple of decades, and remain at the forefront of the applied sciences. Artificial intelligence is the name we give for  this rapidly developing and multifaceted domain of information sciences. 

The conventional criteria for the evaluation of artificial intelligence has been the von Neuman standard of the Chinese Room--implicit to this criteria has been the model of human intelligent functioning as the goal of  artificial intelligence development. This kind of standard criteria is inherently difficult to apply in an objective manner, and, because it embraces the inherent issues of anthropological relativity, it does not transcend the basic dilemmas inherent to human knowledge and intelligence in the world.

Furthermore, it is quite apparent that machine intelligence has as well certain critical non-human constraints that is inherent to their design and functioning as human made machines. These constraints are the following:

            1. All machine intelligence exists, or functions, in a closed world. This world is one that is built, managed and operated by human beings. Intelligent pattern that is the result of machine intelligence is a product of meaningful design, and may be employed  in the production of meaningful design, but it does not by itself produce meaningful design.

            2. All machine intelligence exists, or functions, in a manner that processes information in a linear manner. It processes strings of information, in series that occur in sequential order. Even parallel processing architectures are essentially the cofunctioning of multiple strings.

            3. All machine intelligence exists, or functions, in a manner in which there is no duality of patterning--the signal string contains the information, and the information conveyed by the string is a part of the string itself. In other words, machine intelligence exhibits no duality of patterning in its signal pattern.

            4. All machine intelligence exists, or functions, in a dead, or non-living state. It cannot be attributed the essential synergistic features of living biological organisms, or of what is referred to as "life." A dead state is one that cannot change itself except entropically. Thus, intelligent machines perform a certain or general kind of work, involving energy transfers and heat as a by-product, that results in the manipulation and production of meaningful pattern. Again meaningful pattern is merely  a by-product of this work.

            5. All machine intelligence exists, or functions, in a manner that can be said to lack awareness, either of the self or of the sense of surroundings.

These constraints all occur  at the same time, and are interrelated to one another in the design of machine intelligence. These kinds of constraints are inherently non-anthropomorphic, as there is not  implicit comparison or contrast to human intelligence in their determination.

Technical reductionists would argue that human intelligence can be analytically reduced to the brain wave functioning of neurons that have an electro-chemical basis. This would not be an incorrect analysis to make. In other words, our own intelligence is machine-like just as much as any computer would  be by this reductionist model, and therefore ought to be subject to the same kinds of design constraints are are intelligent machines. Indeed, too, human intelligence is not unconstrained by basic design features and limitations. A brain too large for instance, or overactive, might face a fundamental problem of heat dissipation.

But, also in a technical way, each of these points can be used to contrast human intelligence with machine intelligence. Human intelligence does not exist in a closed world. It functions in an inherently non-linear manner. It has duality of pattern in its signal processing characteristics. It is a living machine, and it can be said to have an advanced form of awareness of both the self and the world in which the self is situated.

It follows that if these are the basic kinds of constraints that predetermine the possibilities of design for intelligent machines, then the design of more intelligent machines will proceed from understanding and as much as possible circumventing or nullifying these kinds of constraints. We measure the quotient of machine intelligence in terms of the degree of sophistication achieved in its functioning and existence along each of these five sets of points.

We can go further, if we wish to adopt a more anthropomorphic model of machine intelligence, then there are further criteria that we might wish to hold as human intelligence exhibits several other features of design that appear for the most part unique to our species:

            1. We are capable of the symbolization of experience, which is the symbolic definition of experience. Indeed, symbolization is such an inherent aspect of our intelligent design, that we cannot not symbolize experience except in the most rudimentary and impulsive of ways.

            2. We are capable of generalizing knowledge from one area or domain to another, and thus devising means of applying this knowledge to alternative domains to which it is not directly derived.

            3. We are capable of creative concatenation of experience and knowledge, to derive new patterns that have no precedence.

            4. We are capable of the linguistic transmission of information that conveys such experience from one person to another. Hence, we are capable of learning new experience based upon the experiences of other people.

These secondary criteria of an anthropomorphized machine intelligence appear to be most useful to the extent that they involve a human interface in a manner that permits the adaptation and mediation of human communication and activities upon multiple levels. I therefore consider these  to be extrinsic criteria versus the intrinsic criteria of the design constraints listed above.

The dilemma of designing and developing more intelligent machines then is the challenge of trying to overcome fundamental, intrinsic and extrinsic constraints of design, that ultimately cannot be overcome in any known manner or by any known means. What is really accomplished in any simple mode is merely an Elizaesque-type parlour trick. Only by means of supercomplex programming and data-base structures might these constraints be approached in any meaningful manner. The challenge is that we do not have a firm idea in any detail of what kinds of designs these may entail, or that may lead us finally beyond the boundaries that conventional machine-like intelligence set for us. One of the best examples of a limited application is in chess and other game playing machines, which machines have increased in sophistication to approach the game-playing capacity of the masters, and even to exceed this capacity in exceptional circumstances. This is a set-piece type of problem, with finite search-solution spaces. The kinds and number of possible moves to be made at each turn are finite and fully determinable, though the number of alternative pathways that can thread through the entire system approaches an astronomical number. This kind of machine-intelligence solution to a limited and deterministic problem set was not achieved easily, but only by  along period of development and application that lead to refinement and sophisticated streamlining of the protocol. To apply a similar kind of complex solution to every deterministic kind of problem set that we can encounter, in whatever area or field of applied knowledge we wish to consider, exceeds by many degrees our greatest supercomputer capacities. This is much more the case if we take into consideration an even broader range of problem sets that do not have deterministic-type solutions, but remain relatively underdetermined in character.

It seems in this regard that intelligent machining in conventional problem solving is most  successful if focused upon narrowly definable goals, and if it proceeds gradually in time from the ground up. The only top-down approach that we can take at this stage is to define a machine-based system of information processing and problem solving that extends the capabilities beyond component machines to incorporate a vast network of machines that interdigitate and articulate with one another in a organic manner. In the construction of such a model, a great deal of unknown problem-solving needs to be subsumed within a critical-path flow-chart that allows an object-oriented and functional partitioning of the general system into a minimal number of component  subsystems. Each system and subsystem must be tackled both separately and interdependently. Each presents its own complex problem set that can be only solved partially and incompletely. Within a larger  system, there will occur deterministic components that define the operational efficiency and intelligent capacity of the system as a whole, though such key components may not be easily or readily identifiable as such.

This type of system puts a premium upon the communicative capacity between machines and operating systems. The  information bottleneck that is based upon the ability for processors to perform a certain speed of operations, is matched by a communication bottleneck that permits different machines to transmit, and receive, processed or raw information only at certain speeds or rates. Generally, in our current state of the art, machines have the be physically connected through transmission lines, and this has posed severe restrictions upon the ability to communicate. The alternative has been a kind of amplitude and frequency modulation of electromagnetic signals. Communicative capacity between machines is as much a challenge of devising a language of mutual intelligibility that would permit signals to transmit that were in a synonymous with the kinds of signals occurring within the operating systems of computers themselves. In other words, the encoding of communiques between devices should be in the same programming language as the computer normally operates in anyway. There should be little requirement for translation interfaces or mediation to be interposed between different systems.

The challenge of constructing a distributed information processing system is in solving the communication needs at various levels and in various areas simultaneously. Communication distribution can be seen as a kind of hypergrid, distributed multidimensionally, each dimensional unit having its own channel capacity for communication separate or at least separable from those streams other dimensional units.

Just as computer processing streams are linear, so also do communication streams tend to be linear. Making multi-linear streams of communication are one way of attacking the problem, as is broadening the transmission breadth of the communication signal. A combined stream that mixes multiple signal carriers within the same grid unit, to be filtered separately by each receiving grid, is an alternative solution to this kind of problem. Within hardwired systems, this problem is readily solved by merely multiplying the number of separate lines interconnecting the various components of the system. Such a filter can be nothing but an embedded sequence of key identifiers that can recognize, for instance, every nth point of reiteration.

The challenge of intelligent communication is therefore the  challenge of constructing complex systems of non-wired transmission based upon some range or set of ranges of electro-magnetic radiation, either focused as in laser systems, or broadcast.

A distributed system can be said to be a remotely connected supercluster of multiple  processing systems interconnected by communication lines based upon broadcast transmission of signals of various forms. Clusters and subclusters of such a distributed system can be said to be hard-wire integrated multiple  processing systems within the larger supercluster grid, presumably that perform either generalized or specialized or both hybrid sets of functions in coordination with other operating clusters. Thus, an internet system such as the world wide web, that connects mostly through telephone lines, is largely as yet a kind of cluster network that is not a truly distributed system. On the other hand, infrared based transmissions connecting office equipment with computers may be considered to be a distributed system. The scale of the system is not so important, I believe, as is the structural design of the system we are dealing with at whatever level. One of the means for a distributed system to achieve a degree of partial openness is through the development of an effective form of broadcast transmission between units. It can be demonstrated anthropologically that human systems and human intelligence could not have arisen outside of the framework of open linguistic communication.

Wireless systems have developed in relation to satellite communication, and these have grown increasingly sophisticated and powerful, as well as with decreasing degrees of noise and static, though they are far from meeting the standards challenges that would be required of a genuinely distributed system.

It follows that strategies of heuristic design are of paramount importance in the consideration of top-down distributed systems in which  the theoretic components exist in complementary manner to the achieved technology. In other words, even if present state of the art technology is relatively primitive and crude to the challenges and goals of any given problem set, it is in the meeting of ground-up practical solutions with top-down design configurations that progress will be defined.

It is something of a paradox as well that devising distributed, wireless based systems on the criteria of relative openness, may be based as well upon solving several other sets of primary constraints in computing--duality of patterning of a limited form is achievable in distributed systems if these distributed systems can interconnect via a common input-output interface and if this interface includes as well feed forward/feedback loops that include effective environmental monitoring on one hand, and effective motor articulation with the environment on the other hand. I am not referring to the conventionally, anthropoidal robot that walks and talks independently of some human controller. Rather I am referring to robotized systems that function independently to achieve a limited range of functional tasks in relation to its environment--such machines can take any form and perform practically any task. The desire to put these machines to human form is as much a reflection of our own anthropocentrism regarding intelligence as anything. 

Achievement of a standard of duality of signal patterning can arise when a common communicative interface can be utilized in alternative contexts to achieve a range of different functional applications by independent and remotely connected machines. It entails the creation of a generalizing symbolic language in intermachine communication that can be adapted to fit a wide and open range of possible applications. This achieves a kind of limited duality that is based upon practical application of general terms to varying contexts. This is normally a trend that is opposite from what is expected with duality of patterning, especially if we adopt a strong psycho-linguistic model of language structure and patterning, though I believe it more accurately replicates what I believe are the actual parameters of communicative design in human language. It emphasizes the social aspects of language function as a communicative system around which cultural and psychological meanings can be built. In this alternative viewpoint, it is the intermediative function of language as a communicative system that is emphasized over the subjective meaning building aspects of any particular language system.

The challenge therefore of building a distributed network supercluster of machines that can perform a wide range of information-based functions in limited dimensions, is two-fold. It is a challenge of constructing a effective system of wireless communication that will permit the long-distance transmission of both large quantities of information at very fast rates, as well as a broad range of different kinds of information transmitted simultaneously or in tandem. It is also the challenge of constructed hard-wired systems as clusters and sub-cluster networks that fit within this multi-dimensional grid structure and that are capable of performing a wide range of alternative information-processing functions simultaneously.

A third challenge arises with the issue of control and coordination structures, in both hard and soft information architectures, that will be heuristically effective in incorporating the entire grid structure in a systematic and synergistic manner. I see such control and coordination as being decentralized and itself distributed at various levels in such a system. Control and coordination remains ultimately a human endeavor, except to the extent that a sense of relative autonomy of function and design can be designed into the architectures of such systems themselves. Self-replication of structure, learning and modification of architectures to fit alternative frameworks would be standards to achieve  in such control structures. Machine systems that are capable of running and managing themselves, with the fewest possible human inputs, and are even capable of  building and repairing themselves, seem to be distant science fiction goals of intelligent design.

There is a sense in this issue, when viewed from the top-down, of a central strategic problem, a general or even universal problem set, that once articulated and fully defined, will lead by deduction and logical inference to the solution of a great many different kinds of problem sets. I do not believe there exists as yet any universal programming language to date that is capable of encompassing all possible logical chaining structures that are typical of intelligent systems. Machines capable of handling such languages would also have to be designed and built, and I do not believe this has yet been accomplished either.

 The  problem and challenge of constructing an intelligent distributed supercluster involves  an entire range of problem sets at multiple levels, each of which must be addressed separately, as well as in relation to the entire structure. We do not know yet what the best or most streamlined design or set of designs would be for the construction of such a system. It is apparent that no single kind of programming system, whether neural networks, or object oriented programming, or  Lisp or Prolog programming, will completely  address every dimension and aspect of the entire problem. It entails putting together the common and conventional approaches in Artificial Intelligence research, in the various applied and theoretical areas, into a common problem set that defines a single advanced distributed system. Thus the challenge of visual pattern recognition and vision is as much a part of the general problem of building such as system as would be the problem of voice  recognition, or of symbolic dependency, or learning or decision making or robotic manipulation or circumlocation.

*****

There occurs a higher level criteria for these kinds of systems. This has to do with the achievement of a degree of generalization of worldview and of self awareness, and what can be called the emergent pattern of mental functioning from mechanical signal transmissions. Grossly, and in an unqualified way, we can refer to this as "consciousness" and we can say that a computer system, however sophisticated in design, lacks intrinsic consciousness. We can attribute  a sense of consciousness to mice and rats, as well as to humans and dolphins. We might even attribute some kind of limited  consciousness to insects and other non-mammalian animal forms. But  we  do  not attribute a state of consciousness to Deep Blue  or to any other supercomputer we have built. The critical question to be answered is "why."

Integration proceeds at different levels and in different ways in the construction and design of distributed architectures. Functions are not completely separable from one another, and there occurs a great deal of overlap that, from the standpoint of informational efficiency, represents a load and a form of noise intrinsic to an underdeveloped and partially unintegrated system.  Components must replicate similar kinds of procedures in the course of normal operation. In the best of possible worlds, each procedure would only need to be performed one time by one machine: the results of this procedure  would then be stored and made  available for use by any other machine further down the road. Often, there are diminishing returns if retrieval of  stored  information, or the storage of information itself, requires a more informationally expensive procedure than reiteration of the original procedure in the first place.

There is  a fundamental trade-off it seems, between the problem of integration on one hand, that combines subsystems into a single hard-wired "cluster" and the problem of distributed processing, which serves to link different systems or clusters into a coordinate network. It seems that we can improve systems integration through hardwiring, but only at the  expense of maintaining truly and remotely  distributed networks. On the other hand, if we wish to extend distributed networks to encompass broader ranges, then the price  we pay is in our ability to integrate  systems as a single operational unit. In a sense, with the problem of distribution, the  challenge of effective communication between different systems becomes paramount over the challenge of processural integration into a single system.

The concept of unit operations is an important approach to  take in applied metasystems and in the design and coordination of different systems. Operational units define unit operations as basic common functional denominators, and provide thus a shorthand for design of more complex systems. A limited number of basic operations, for instance, can be recombined in a countless number of  ways to achieve alternative complex systems.

Meaning and learning

Awareness

Artificial life.

Light research and Light computing.

Advanced Systems

by Hugh M. Lewis