|
Author
|
Topic: A Definition of Intelligence
|
warren_bergerson
Member
Member # 262
|
posted 19. December 2003 14:49
The following is a draft of an article proposing a formal definition of intelligence, intelligent behavior, intelligent processes, and intelligent systems. Any comments would be appreciated.
A WORKING DEFINITION OF INTELLIGENCE AND INTELLIGENT SYSTEMS
INTRODUCTION
The purpose of this article is to propose what will be called a working definition of intelligence. The proposed definition is intended to serve three key criteria. First, the definition is intended to provide a clear and unambiguous method of distinguishing intelligent behavior, intelligent processes, and intelligent systems from behavior, processes and systems which do not qualify as intelligent. Second, the definition is intended to be compatible with at least some intuitive concepts of intelligence. Third, and by far the most important criteria, the definition is intended to be useful and practical for the mathematical, scientific, and engineering analysis, modeling, simulation and design of intelligent behavior, processes and systems.
It is useful to recall that a definition of a complex phenomena like intelligence is based not on enumerating all the properties of the phenomena. A definition is instead based on a limited number of key properties which can distinguish occurrences of the complex phenomena from non-occurrences. The definition presented here is based on the concept that intelligence is a form of information processing. Intelligent information processing, it is proposed, is distinguished from non-intelligent information processing by being preceded by or being the product of a complex form of information processing that is labeled here as ‘learning efficient problem solving’ or LEPS. This ‘being preceded by LEPS’ property produces the following working definition:
DEFINITION OF INTELLIGENT BEHAVIOR, INTELLIGENT PROCESSES AND INTELLIGENT SYSTEMS: A behavior, process or system is characterized as intelligent if and only if the behavior, process, or system is dependent on, associated with, and a product of a complex and extended preliminary processing called ‘learning efficient problem solving’ (LEPS). Dependence on LEPS may involve any combination of 1)LEPS processes included in the system, 2)the results of LEPS processes incorporated into the beginning state of the system, and/or 3)the results of LEPS being included in the systems input.
As will be discussed, this is a broad definition of intelligence which includes a very wide range of information processing operations (provided the processing is preceded and dependent on LEPS processes and provided the processing does not occur in nature in the absence of preceding LEPS processes. As will also be discussed, this definition is useful in both the mathematical and scientific analysis of intelligence, intelligent behavior, intelligent processes, and intelligent systems because it makes it possible and practical to generate a variety of testable predictive hypotheses.
THE LOGICAL BASIS FOR THE WORKING DEFINITION
The proposed definition of intelligence is based on a number of key assumptions regarding the nature of intelligence and intelligent systems. These assumptions, it is suggested, appear sound and reasonable. based on currently available evidence. At least to some extent, the soundness of these assumptions can be reviewed and tested.
ASSUMPTION ONE: Intelligence, intelligent behavior, intelligent systems can be analyzed, modeled, simulated and designed using information processing concepts and mathematics.
ASSUMPTION TWO: Any known and well defined form of intelligent behavior or process which can be modeled and simulated with goal-directed mathematical algorithms or programs. There are processes which we can not currently precisely define and model due to the level of complexity involved, but there does not appear to be any specific process or operation which is logically impossible to model and simulate.
ASSUMPTION THREE: Any goal-directed algorithm or function which simulates intelligent behavior or an intelligent process can be generated mathematically by a member of a class of functions labeled here as LEPS processes.
ASSUMPTION FOUR: All behavior, processes, and systems which are recognized as intelligent appear to be associated with complex real world phenomena which may be modeled as LEPS processes. As far as is currently known, all LEPS processes are associated with life forms on earth.
The above assumptions are based in large part on concepts and techniques used in designing complex computer systems. It is not entirely clear to what extent if any the above assumptions have been subjected to formal analysis. The claim made here is that the above assumptions appear reasonable and sound based on currently available evidence.
EFFECTIVENESS IN DIFFERENTIATING INTELLIGENT FROM NON-INTELLIGENT
One requirement for a definition of intelligence is to provide a clear, objective and reasonably unambiguous basis for differentiating intelligent behavior, processes and systems from non-intelligent behavior, processes and systems. It will be noted that the clear distinction requirement is not the same as the compatible with intuitive meaning criteria.
Based on the proposed definition, a behavior, process, or system is defined as intelligent if it satisfies two key criteria. A behavior, process or system does not qualify as intelligent if it fails either of the following two criteria:
CRITERIA ONE: The behavior, process or system is associated with and dependent on a LEPS process.
CRITERIA TWO: The behavior, process or system does not occur in the absence of LEPS processes.
The only known LEPS processes on earth are those directly or indirectly associated with life forms. In most instances, the first criteria can be evaluated by determining if the behavior, process or system is directly or indirectly related to life forms. The second criteria can similarly be evaluated by determining if the behavior, process or system can or does occur in the absence of the influence of life forms.
COMPATIBILITY WITH INTUITIVE CONCEPT
The second criteria to be satisfied by a definition of intelligence is compatibility with the intuitive or common sense concept of intelligence. Satisfying this criteria is something of a challenge because there are so many different and often inconsistent concepts of intelligence in use.
The intuitive concept of intelligence used here is expressed as "Intelligence is a complex form of information processing associated on earth exclusively with biological life forms." Three features of this intuitive concept are worth noting. First, it does not involve the concept of consciousness. Second, this concept does not limit intelligence to a property of humans or to species closely related to humans. Finally, the definition does not limit intelligence and the manifestations of intelligence to life forms. Under appropriate conditions, manifestations of intelligence can be exhibited by manmade systems. This includes systems which do not contain LESP processes.
USEFUL IN MATHEMATICAL, SCIENTIC, AND ENGINEERING APPLICATIONS
The third, and by far the most important, criteria to be satisfied by a definition of intelligence is usefulness in mathematical, scientific and engineering applications. Since computer simulation is one of the key tools used in the analysis of intelligence and intelligent behavior, a useful definition of intelligence means one compatible with mathematical modeling and computer simulation. It should be apparent that the definition proposed in compatible with mathematical modeling and computer simulation.
Even more important in scientific and engineering analysis is the requirement that the definition used by compatible with the formulation of testable, predictive hypotheses. The definition of intelligence offered here is compatible with a whole series of testable, predictive hypotheses including hypotheses with the following general format.
TYPE A HYPOTHESES: If behavior, process, or system is intelligent, then there exists some goal directed algorithm Y which can model and simulate X.
TYPE B HYPOTHESES: If behavior, process, or system X is intelligent, then there exists some goal directed algorithm Y compatible with known physical properties of X which can model and simulate X.
TYPE C HYPOTHESES: If Y is a goal directed model of an intelligent behavior, process or system X, then there is a LESP Z which can generate Y.
TYPE D HYPOTHESES 4: If Y is a goal directed model of an intelligent behavior, process or system X, then there is a LESP Z which is compatible with known physical properties and which can generate Y.
SUMMARY
At least based on initial review, the definition of intelligence proposed here appears to satisfy the listed criteria. The proposed definition is designed primarily for compatibility with artificial intelligence applications. It is, however, believed to be useful in analysis of a variety of complex behaviors and processes associated with biological life forms.
IP: Logged
|
|
warren_bergerson
Member
Member # 262
|
posted 21. December 2003 12:11
PHYSICAL PROCESSES VERSUS INTELLIGENT PROCESSES
One of the fundamental questions in science and philosophy is the question of the differences between physical processes and intelligent processes. Are the laws of nature governing physical processes the same as the laws of nature governing intelligent processes? From another perspective, are the scientific hypotheses or theories describing and explaining physical processes and laws the same or different than the scientific hypotheses or theories needed to describe and explain intelligent processes and laws? The definition of intelligence offered above provides answers to these questions.
Based on the definition proposed, intelligent processes can be characterized as dispersed and dynamic. Physical processes by contrast have traditionally been characterized as permanent and universal. It is sometimes useful, if not entirely accurate to think of physical processes as simple, elementary processes. Theses elementary physical processes operate in essentially the same manner at all times and all locations. Intelligent processes, from this perspective are made up of large sets of simple processes. In aggregate, these sets of processes produce complex relationships operate, or appear to operate differently at different times and locations. It appears that in terms of abstract mathematical and logical concepts, physical and intelligent processes are manifestations of the same basic causal concepts or processes. As a practical matter however, it appears to be useful to treat physical and intelligent causal processes as two distinct categories of causation.
Scientific models and hypotheses can be formulated for both physical processes and intelligent processes, but the types of algorithms involved are different. Physical processes and the laws governing physical processes are best modeled by simple deterministic or algebraic functions. Intelligent processes and the laws governing intelligent processes are best modeled by goal directed algorithms or programs. Although the mathematical algorithms involved are quite different, the scientific processes or paradigms used to develop and test hypotheses appear to be very similar.
The concept which is used to differentiate intelligent processes from physical processes is dispersed processing. If you isolate occurrences of intelligent processes and physical processes they appear to be very similar. However, while a physical process can exist or apply anywhere, an intelligent process is dependent on dispersed processing. Unlike a physical process, an intelligent process will fail or disappear if isolated from the results of dispersed processing. The connections or communications between an intelligent process and the associated dispersed processing can be used to analyze, understand and control it. The existence of an intelligent process can be identified and tested by disrupting the communication with dispersed processing.
IP: Logged
|
|
warren_bergerson
Member
Member # 262
|
posted 24. December 2003 08:48
The definition of intelligence offered above involves a paradigm shift. In order to understand or visualize the above definition it is necessary to abandon one of the standard premises or assumptions used in academic and theoretical science and replace it with one of the standard assumptions used in engineering, systems design, and other applied sciences.
Essentially all theoretical scientific analysis is based on some variation of the uniformity premise. This assumption asserts that the laws or causal relationships governing certain units are uniform with respect to time and location. Engineering and system design, by contrast use what can be labeled the ‘dynamic dispersed processing’ (DDP) premise. DDP asserts that the laws governing the behavior of an object are dynamic or changeable and dependent on processing occurring at different or dispersed times and locations.
The differences between the uniformity premise and the DDP premise are easily illustrated by computers or logic machines. The uniformity premise would assert that all members of a set of logic machines are governed by the same uniform laws and will produce the same output for the same input. DDP, by contrast, asserts that the logic machines are governed by dynamic laws or programs and that the governing laws or programs are changed as the result of processing or processing occurring at ‘dispersed’ times and locations.
It should be noted that ‘uniformity’ and ‘DDP’ reflect differences in viewpoint or perspective. The issue is not whether the logic machines being analyzed are ‘really’ governed by uniformity or DDP. The issue is which perspective leads to the more useful analysis.
Scientists have for centuries been trying unsuccessfully to unravel the laws governing intelligence using the uniformity premise. In fact, it can be argued, academic or theoretical science has failed to unravel or discover any of the laws governing life forms. These failures are all due to reliance on the uniformity premise.
Viewing and analyzing something from a different perspectives or using different premises is called a paradigm shift. Understanding intelligence, understanding human intelligence, and in fact understanding the processes governing living systems requires changing perspective or a paradigm shift. Understanding intelligence, I suggest, requires abandoning the uniformity premise. Fortunately, the required replacement premise or perspective or paradigm is available from engineering, systems design, and the applied sciences.
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 24. December 2003 18:54
I can't resist this. It's too bizarre to let go by. warren_bergerson wrote quote:
Essentially all theoretical scientific analysis is based on some variation of the uniformity premise. This assumption asserts that the laws or causal relationships governing certain units are uniform with respect to time and location. Engineering and system design, by contrast use what can be labeled the 'dynamic dispersed processing' (DDP) premise. DDP asserts that the laws governing the behavior of an object are dynamic or changeable and dependent on processing occurring at different or dispersed times and locations.
I spent 10 years, first as a technician and then a systems scientist, working in the aerospace and defense industry. I worked on the design, development, and testing of systems ranging from the Polaris intermediate range ballistic missile's autopilot computer to the Apollo Command Module stabilization and control system, and worked in organizations with names like 'Development and Evaluation Laboratory' and 'Systems and Research Center' in a major aerospace company. I worked with engineering designers and I did some engineering design myself. I never once saw an engineering designer assume that "the laws governing the behavior of an object ... are changeable ...". On the contrary, engineering design depends on the assumption that the laws governing the behavior of objects are "uniform" in warren_bergerson's terminology.
While warren_bergerson is perfectly free to offer an alternative approach to defining intelligence, it does not advance his case to misrepresent other disciplines.
RBH
[ 24. December 2003, 19:03: Message edited by: RBH ]
IP: Logged
|
|
Salvador T. Cordova
Member
Member # 959
|
posted 24. December 2003 23:26
The more elementary a principle, the less likely one will be able to arrive at a non-circular definition. Intelligence I think fits this category.
For example a point in Euclidean Geometry is what is known in math as an 'undefined' term. One can take a dictionary, try to define a term with words. Look up all those words that define that word and they are described by other words! Reapeat the look up over and over and eventually circularity results.
That does not mean our understanding is not true, it is simply incomplete (Godel). We can enumerate characteristics of points in Euclidean Geometry, but aspects of it will remain undefined. Likewise, it is fruitless to define intelligence. We can only enumerate its characteristics relative to it's actions on the real world.
1. it makes choices
2. it can perform logic
3. it can solve problems
4. it can act on incomplete knowledge
etc.
Hope this helps,
Salvador
IP: Logged
|
|
warren_bergerson
Member
Member # 262
|
posted 25. December 2003 07:46
Salvador.
I would agree that definitions as an expression of absolute truth are not possible. However, in science, mathematics and everyday life serve useful functions in human communications. The definition or working definition of ‘horse’ may not provide an exhaustive, perfect, absolute knowledge of the concept of horse, but it provides a useful and relatively uniform basis for different humans to make similar distinctions between reindeer and horses.
In order to perform scientific analysis of intelligence, intelligent behavior, intelligent processes and intelligent systems, scientists need a working and workable definition of intelligence. The purpose of the definition is not to provide an absolute meaning but rather to provide a meaning so that different scientists analyzing different aspects of intelligence can have a reasonable expectation that they are using the term intelligence in a consistent manner.
Attempts to define intelligence in terms of vague imprecise traits or properties have not, in practice, proved to be scientifically useful. Take the question "Are evolutionary change processes intelligent or non-intelligent?". None of the vague informal definitions floating around even provides an objective verifiable method of determining whether evolution is or is not an intelligent process.
Based on the definition proposed, evolutionary processes would be non-intelligent processes if evolution resulted from the interaction of uniform, naturally occurring processes. Evolution is defined as an intelligent process if it is involves dynamic processes which are modified by deferred processing involving LEPS type processes. It is interesting to note that the early developers of the Modern Synthesis insisted on a similar type uniformity standard.
It is unfortunate that RBH is unable to present comments in a civil manner even at Christmas time. If engineering was based on the uniformity principle, then there would be no need to design new automobile since it would be assumed that all autos are uniform. In reality, engineering is based on the principle objects are not uniform and that the performance of objects can be modified by the various dispersed processing actions of engineers and those implementing engineering analysis. Engineers do use predictive scientific hypotheses developed using the uniformity or scientific determinism premise, but this does not mean they use the uniformity premise is designing objects.
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 25. December 2003 13:15
warren_bergerson wrote quote:
It is unfortunate that RBH is unable to present comments in a civil manner even at Christmas time. If engineering was based on the uniformity principle, then there would be no need to design new automobile since it would be assumed that all autos are uniform. In reality, engineering is based on the principle objects are not uniform and that the performance of objects can be modified by the various dispersed processing actions of engineers and those implementing engineering analysis. Engineers do use predictive scientific hypotheses developed using the uniformity or scientific determinism premise, but this does not mean they use the uniformity premise is designing objects. (Emphasis added)
But warren_bergerson's original claim was quote:
DDP asserts that the laws governing the behavior of an object are dynamic or changeable and dependent on processing occurring at different or dispersed times and locations. (Emphasis added)
As far as I'm aware, Newton's laws of motion apply to automobiles unchangeably regardless of the design of the automobiles. I've run on a volunteer heavy rescue team since 1973, and over the years the changing design of automobiles has not altered the relationship embodied in F=MA. When a 2,000 pound car hits a tree at 65 mph, F=MA still captures the total force of the impact. Altering designs, say by installing air bags, can alter the consequences of that impact, but constant physical laws still govern. Changing the slope of the deceleration of an impact over time by changing the design of the front end of automobiles to spread the force of the impact over more milliseconds depends on the constancy of Newton's laws of motion and on the constancy of the physical properties of matter.
Given a known F, one can attempt to design an automobile's structure to distribute F over time and physical substructures so that the deceleration is less likely to harm the human occupants in the passenger compartment. But the same total F is produced (for given M and A) regardless of the design. The design alterations are attempts to deal with the unchangeable characteristics of the world that are described by constant physical laws. An automobile design engineer who worked on the assumption that F=MA does not describe an invariant relationship would soon be out of work.
Expressed characteristics of things, the performance of objects, their external appearance, the efficiency of their functioning, and so on, can be changed by the actions of humans acting (singly or in dispersed groups) to alter the design of those objects, but the alterations are constrained by uniform laws of nature that operate today exactly as they did millennia ago.
I'm sorry that warren_bergerson regards pointing out errors in his analysis as uncivil. But it would be less civil to let those errors go without comment: unremarked errors propagate and can infect a whole analysis. If warren_bergerson thinks my remarks here were uncivil, he should have participated in some of the design meetings I have. When one is designing objects on which people's lives will depend and where tradeoffs between inconsistent goals proliferate and where physical constraints are tight, as they are in designing spacecraft, disagreements can get really heated.
RBH
IP: Logged
|
|
Salvador T. Cordova
Member
Member # 959
|
posted 25. December 2003 22:12
Hi Warren,
First of all, I am a very strong ID advocate, I commend your attempt to define intelligence.
My personal bias is that we can successfully define CHARACTERISTICS of intelligence, aspects of it, but not intelligence itself.
My personal favorite characteristic of intelligences is the ability to form algorithmic and symbolic constructs. The proverbial 500 pennies all heads fits this idea very well. There may be other aspects of intelligence, but this is one that is mathematically detectable.
Computers and their ability to manipulate symbols, give birth to new algorithms (these algorithms exist through inheritance not novel creativity however). This quality gave rise to the term Artificial Intelligence.
I know it can be hard to be on this board and have our ideas questioned. But even if we are ultimately right, we can benefit from our critics comments. I've often found that sometimes I make an unfortunate distortion of my ideas and the critics have been the ones most helpful to improve my presentation. It is not a pleasant process, but I endure it when I believe the outcome is worthwhile.
Some of various critics on this board and others I don't even bother responding to. However, though RBH and I are often at opposite ends of the spectrum, I value his comments and criticisms very highly. If you can defend your thesis (politely) with someone as bright and knowledgeable as RBH, it will definitely raise one's confidence.
I encourage you that if you disagree, or if you know you are right, don't try to humiliate or fight your critics because they may be the very ones who will help make your ideas robust.
Merry Christmas,
Salvador
IP: Logged
|
|
warren_bergerson
Member
Member # 262
|
posted 26. December 2003 08:48
Salvador,
If you accept ID as a science or potential science then of necessity you must accept the possibility of defining both intelligence and design. The ability to effectively and reliably distinguish between intelligent behavior, processes and systems and non-intelligent behavior processes and systems is an essential starting point. To argue that intelligence is not definable is to argue that a science of ID is impossible.
The general definition of intelligence used in AI is that intelligence is or can be expressed as some complex form of information processing. I am simply proposing a refinement of a widely used and widely accepted approach to defining intelligence. The definition proposed starts by recognizing that intelligence involves DDP or dynamic dispersed processing ‘programs or laws’ as opposed to uniform or permanent and universal processing programs or laws.
All computer programs involve dynamic dispersed processing to a greater or lesser extent. Human engineering design processes also clearly involve dynamic dispersed processing. For both humans and computers, the existence of DDP can be demonstrated by disrupting communications between different distributed processes and observing the impact.
I must admit I have no idea why RBH would object to defining intelligence in terms of DDP. Clearly all known forms of complex information processing involve dynamic dispersed processing. As is also clear, engineers and systems designers have extensive experience and expertise in addressing DDP. Maybe he could attempt to clarify his logic.
There are some attempts or suggestions that intelligence might be definable in terms of a non-DDP or permanent and universal process(PUP). This would include some suggestions that intelligence might involve some as yet unknown type of physical force. The EAM concept supported by some ARN posters would appear to be a PUP approach to defining intelligence.
Making a scientific concept or definition ‘robust’ involves at least a two stage process. First, the proposed concept or definition is made available to a wide audience for questions and comments. Second, the proposed concept or definition must be subjected to formal review and analysis by appropriate experts operating under appropriate analytical standards and principles. ISCID can serve as a useful environment for the first stage of this process.
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 26. December 2003 12:33
warren_bergerson remarked quote:
I must admit I have no idea why RBH would object to defining intelligence in terms of DDP. Clearly all known forms of complex information processing involve dynamic dispersed processing. As is also clear, engineers and systems designers have extensive experience and expertise in addressing DDP. Maybe he could attempt to clarify his logic.
I have not objected to the notion of distributed (dispersed) processing. Indeed, I use distributed processing in the course of my work. warren_bergerson's definition of "dispersed processing" is unobjectionable: quote:
The concept which is used to differentiate intelligent processes from physical processes is dispersed processing. If you isolate occurrences of intelligent processes and physical processes they appear to be very similar. However, while a physical process can exist or apply anywhere, an intelligent process is dependent on dispersed processing. Unlike a physical process, an intelligent process will fail or disappear if isolated from the results of dispersed processing. The connections or communications between an intelligent process and the associated dispersed processing can be used to analyze, understand and control it. The existence of an intelligent process can be identified and tested by disrupting the communication with dispersed processing.
warren_bergerson's notion that intelligence involves processes distributed over some array of separate 'modules' or structures is reminiscent of Minsky's Society of Mind conception that was published 15 years ago.
The "dynamic" part of "dynamic dispersed processing" is defined by warren_bergerson to mean that the underlying laws of the universe are changeable, mutable, and that intelligence somehow depends on that mutability: quote:
Engineering and system design, by contrast use what can be labeled the 'dynamic dispersed processing' (DDP) premise. DDP asserts that the laws governing the behavior of an object are dynamic or changeable and dependent on processing occurring at different or dispersed times and locations. (Emphasis added)
It is that "dynamic" component, the assertion that engineering design uses the assumption that the laws underpinning the physical universe are mutable, changeable, that I object to because it is false: Engineering design depends on there being a stable set of laws and principles that describe the physical universe. Whether it's building bridges, spacecraft, or automobiles, engineering design requires that the laws be "uniform" in precisely the sense that warren_bergerson uses that word.
If by the term "dynamic" warren_bergerson intends to say that the various components of a dispersed processing system interact with one another and thereby affect one another's behavior, as his operational definition (knock out communication links) suggests, that's a different kettle of fish. That's not "laws governing the behavior of an object," that's merely a restatement of the "dispersed" condition.
RBH
IP: Logged
|
|
warren_bergerson
Member
Member # 262
|
posted 26. December 2003 15:07
RBH,
To begin with, you seem to be confusing ‘the nature of causal relationships’ with the mathematical conventions used to model causal relationships in scientific hypotheses. In physics, scientific hypotheses are generally based on modeling causal relationships using an algebraic permanent and universal format. There are other mathematical forms which can be used to express causal relationships and in some instances the alternative mathematical formats are more practical.
It is generally useful to analyze and model behavior and intelligence in terms of information processing. It is generally useful to express or model this information in terms of computer programs. In such modeling it is generally useful to recognize both the role of dispersed processing and to recognize that the programs or laws governing information processing are changeable or dynamic. It would be possible, but not practical, to express information processing in terms of permanent and universal causal relationships.
Second, your comments appear to be confusing the engineering design process with the permanent and universal scientific hypotheses used in the engineering design processes. The goal in engineering a new object, I suggest, might be described as creating a new program which creates a new set of rules or laws by which the object interacts with the environment. The engineer uses and initiates a variety of dispersed processing to create the new object and new program controlling the relationship between the object and its environment. There is much more to the engineering design process than a small number of permanent and universal hypotheses.
None of the points you raise precludes defining intelligence in terms of DDP.
IP: Logged
|
|
warren_bergerson
Member
Member # 262
|
posted 27. December 2003 10:49
A least in modern times, there has existed a very large gap between knowledge in the applied sciences and knowledge in the academic sciences. No where is this gap more noticeable or more problematic than in the study intelligence. It is generally recognized that intelligence involves complex causation. By far and away the best tool for modeling and analyzing complex causal relationships is the computer and the computer program. But academic scientific and mathematical analysis of complex causal relationships have not kept up with advances in computer technology and systems design. Academic science in large part still operates on the outmoded premise that causation is limited to simple algebraic relationships and stochastic variations of these simple algebraic relationships.
To bridge some of this gap, it is useful to start by noting that causal relationships of the form "A always causes B" are modeled by mathematical functions. Relatively simple causal relationships are modeled by relatively simple mathematical functions. Complex causal relationships are modeled by complex mathematical functions. Computer programs, it will be recognized, are a complex type of mathematical function which model complex causal relationships. Computer programs are the types of mathematical functions used to model the causal relationships associated with intelligence.
Dynamic dispersed processing or DDP is one of the ‘interesting’ properties of computer programs and of certain types of complex causation. A DDP causal relationship is described as being dynamic or as having the appearance of being dynamic. The relationship between cause and effect or input and output changes over time. This change is attributable to dispersed processing which may occur within the program or which may be the result of external dispersed processing.
While DDP is apparently not a recognized phenomena in academic science, it is a widely used concept in engineering (even if the term DDP is not used). Engineering, it can be argued, involved complex dispersed processing used to change the complex causal properties (or the causal programs) controlling designed objects.
One we understand that DDP is an important and useful form of complex causation, then we can begin to consider goal directed or purposeful DDP’s which can be used to express predictive scientific hypotheses. Finally, once we master goal-directed DDP’s we can begin to consider the LEPS processes or programs proposed here to define intelligence. Although the terminology being used may be new, DDP, LEPS and goal directed DDP’s are all concepts currently in use in engineering and system design.
IP: Logged
|
|
Rex Kerr
Member
Member # 632
|
posted 29. December 2003 15:13
Warren claims that "DDP is apparently not a recognized phenomena in academic science".
I am curious to know what his definitions of "academic science" and "apparent" are. Neuroscience textbooks for undergraduates assume a "DDP"-like model for neural information processing. Defining intelligence is fine, but it's advisable to address drawbacks of current approaches from a position of knowledge and support the claims in detail. Otherwise the reader may be distracted from the valuable parts of the presentation by claims that they know are false (as I have just been).
Warren's definition seems to hinge on something called "Learned Efficient Problem Solving", but there is little detail on what LEPS actually is.
For some tasks, a multilayer perceptron can efficiently solve a problem after training (learning). Is that an instance of LEPS?
Reflexes are hardwired responses to given stimuli. It's a very efficient way to solve a problem (e.g. to avoid injury), but the "learning" takes place at the genetic level not the individual level. Is this an instance of LEPS?
I can learn how to do arithmetic, but learn how to do it inefficiently. Even the best methods are shockingly inefficient when computed using the human brain instead of, say, a Pentium 4--and most people learn rather sub-optimal methods for doing arithmetic to begin with. Does this fail to be an instance of LEPS?
[ 29. December 2003, 20:29: Message edited by: Rex Kerr ]
IP: Logged
|
|
warren_bergerson
Member
Member # 262
|
posted 30. December 2003 09:08
Rex,
DDP is a form or class of complex causation. DDP type causation is used routinely in systems design and, I suggest, in engineering applications. I am not aware of any academic or theoretical science which formally recognizes any form of causation other than permanent and universal causation and stochastic causation. Any number of fields move back and forth between applied science concepts and abstract science concepts, but again to my knowledge, no one formally recognizes DDP as a theoretical science concept.
If I read your comments correctly, you are suggesting that a formal definition of intelligence should start from existing concepts. The proposed definition does build from existing concepts and principles used specifically in systems design and more generally in engineering and applied science. In order to formulate a scientific definition of intelligence, I am proposing, you need to start with the concept of DDP as complex causation. No one, to my knowledge, in any field of academic or theoretical science has started from this concept and thus no one has developed a working definition of intelligence.
While academic or theoretical science continues to suggest and even insist that ‘intelligence’ can not yet be formally defined, the systems design experts working to simulate various types of intelligent processes and intelligent behavior already have an effective if informal definition of intelligence. In AI and systems design, intelligence is a form of information processing which can be expressed by a computer program (DDP). In order to create a successful simulation of intelligent behavior, and there exist a variety of successful simulations, it is necessary to use a goal-directed form of DDP. Finally, if you wish to construct a goal-directed DDP from a primitive system(or from an existing goal-directed DDP), you need what I have labeled a LEPS program.
From the existence of successful simulations of certain types of intelligent behavior and processes, we can conclude that at least a handful of systems designers have discovered that you can successfully simulate intelligent behavior with goal-directed DDP’s. I am not aware that anyone has proposed that ‘you can model or simulate any form of intelligent behavior or process with a goal-directed DDP’, but the hypotheses appears fairly obvious.
Once you recognize that intelligent behavior can be modeled and simulated by goal-directed DDP’s it is reasonable to ask the question where do goal directed DDP’s come from. Given a generalized form of mathematics for expressing goal directed DDP’s, it is not particularly difficult to define or develop abstract mathematical programs for transforming and creating goal directed DDP’s. I have labeled such programs LEPS. While it is not difficult to develop an abstract LEPS, it is very challenging to try to develop a practical or efficient LEPS (although I know of one person who is attempting to address the issue).
Again, if you are suggesting that a definition of intelligence needs to build on existing knowledge, then I am in agreement. However, the knowledge base being used is different from the traditional academic and theoretical science approaches which have consistently failed to produce successful or effective working definitions.
LEPS
LEPS can be operationally defined as ‘a type of information processing uniquely associated with life forms (at least as far as is currently known on earth). LEPS can also be defined as a type or class of mathematical algorithm or program. For a more detailed discussion of this see Message 4 of the Definition of Intelligence topic at http://groups.msn.com/LifeEngineering/messages.msnw .
It should be noted that learning efficient problem solving is not the same as ‘learning optimal problem solving’. Biological problem solving and human problem solving are in some respects extremely efficient and in other respects extremely inefficient.
If you wish to better understand LEPS and the proposed definition of intelligence, it may be useful or at least interesting to ask the question "Is biological evolution an intelligent or non-intelligent process?". Are evolutionary and adaptive changes in biological systems the result of non-intelligent chance interactions of simple naturally occurring processes? Or are evolutionary and adaptive changes in biological systems intelligent processes dependent on the existence of complex LEPS pre-processing?
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 30. December 2003 11:16
Once again we are in a morass of unsupported assertions. warren_bergerson wrote quote:
While academic or theoretical science continues to suggest and even insist that 'intelligence' can not yet be formally defined, the systems design experts working to simulate various types of intelligent processes and intelligent behavior already have an effective if informal definition of intelligence. In AI and systems design, intelligence is a form of information processing which can be expressed by a computer program (DDP).
Here is a review of half a dozen "working definitions" of intelligence from the AI literature. Not one of them defines intelligence in terms of a computer program. One mentions "computer program," but only to say that it does not matter whether intelligence is instantiated in a computer program, a human, or a Martian. To be sure, they all sooner or later want to model intelligence in a program, but they do not define intelligence in terms of a program.
And there actually are "working definitions" of intelligence in the academic literature. For example, six are mentioned here.
And here is an overview of Sternberg's Triarchic model of intelligence. It forms the basis for a good deal of academic work.
Computational models entered cognitive psychology (the academic study of perception, learning, memory, decision-making, and so on) in a pretty big way starting in the 1960s. For example, Miller, Galanter, and Pribram's Plans and the Structure of Behavior, originally published in 1960, proposed a computational model of cognition. When I spent a day as his host in 1969, Pribram (a neuroscientist) was very high on computational models of human cognition. At that time he was pursuing the conjecture that information processing was embodied in interference patterns in 2-D holographic representations in the DC microcurrents of Layer VI of cortex.
The PDP Research Group's Parallel Distributed Processing: Explorations in the Microstructure of Cognition, published in 1986, proposed detailed parallel distributed models for a variety of cognitive processes, and has been very influential in academic cognitive science. My copy of the two-volume work is getting pretty battered but I still consult it on occasion.
A Google search on "computational model" AND "cognitive psychology" yields 11,000 hits. Hardly a dearth of material there.
I'd sure like to see some references documenting warren_bergerson's claim that academic science insists that intelligence cannot be usefully defined.
Once again, warren_bergerson is completely free to propose an alternative definition, but also once again, it does not advance his argument to misrepresent the professional literature of the last half-century or so.
RBH
[ 30. December 2003, 11:31: Message edited by: RBH ]
IP: Logged
|
|
|
Printer-friendly view of this topic