Newell and Simon:
The phenomena surrounding computers are deep and obscure, requiring much experimentation to assess their nature.
Each new machine that is built is an experiment. Actually constructing the machine poses a question to nature; and we listen for the answer by observing the machine in operation and analyzing it by all analytical and measurement means available. Each new program that is built is an experiment. It poses a question to nature, and its behavior offers clues to an answer.
One of the fundamental contributions to knowledge of computer science has been to explain, at a rather basic level, what symbols are. This explanation is a scientific proposition about Nature. It is empirically derived, with a long and gradual development.
The symbol system hypothesis implies that the symbolic behavior of man arises because he has the characteristics of a physical symbol system. Hence, the results of efforts to model human behavior with symbol systems become an important part of the evidence for the hypothesis… The empirical character of computer science is nowhere more evident than in this alliance with psychology. Not only are psychological experiments required to test the veridicality of the simulation models as explanations of human behavior, but out of the experiments come new ideas for the design and construction of physical symbol systems.
Definition, from Newell and Simon:
A physical symbol system consists of a set of entities, called symbols, which are physical patterns that can occur as components of another type of entity called an expression (or symbol structure). Thus, a symbol structure is composed of a number of instances (or tokens) of symbols related in some physical way (such as one token being next to another). At any instant of time the system will contain a collection of these symbol structures. Besides these structures, the system also contains a collection of processes that operate on expressions to produce other expressions: processes of creation, modification, reproduction and destruction. A physical symbol system is a machine that produces through time an evolving collection of symbol structures. Such a system exists in a world of objects wider than just these symbolic expressions themselves.
The hypothesis:
A physical symbol system has the necessary and sufficient means for general intelligent action.
Definition:
An expression designates an object if, given the expression, the system can either affect the object itself or behave in ways dependent on the object.
Designation is not part of formal models of logic or computing:
In none of these systems is there, on the surface, a concept of the symbol as something that designates. The data are regarded as just strings of zeroes and ones - indeed that data be inert is essential to the reduction of computation to physical process.
But designation emerged through list processing:
The next step, taken in 1956, was list processing… It took its cue from the emergence of random access memories, which provided a clear physical realization of a designating symbol in the address. The contents of the data structures were now symbols, in the sense of our physical symbol system: patterns that designated, that had referents… [Colleagues] found it strange that there were no bits [that held the content of the system], there were only symbols that designated yet other symbol structures.
Actually, what's being said here is very misleading, and this is the basis for a lot of misunderstandings, criticisms and emendations of Newell and Simon's view.
Definition:
The system can interpret an expression if the expression designates a process and if, given the expression, the system can carry out the process.
It's easy to get caught up in designation - and we'll be talking a lot about designation today - but interpretation is really the pivot that lets you leverage computer science in thinking about intelligence. In particular, interpretation allows you to think of symbolic reasoning as leading to choices of action: a system can output the name of an action to perform and then interpret that name to do what it has decided. We've seen that that is crucial to a knowledge-level view of system behavior. In addition, interpretation is an important element in explaining why physical symbol systems can exhibit general intelligence. Interpretation is the main building block of universal computers, which can execute any specified algorithmic process; so having interpretation in a physical symbol system means that the system can take its information about the world - in terms of expressions that designate objects and properties - and put it to use in an arbitrarily open-ended way. I think this is something that got Newell and Simon particularly excited but it doesn't feature prominently in the paper or (especially) in the reaction to it in the literature.
Talking about lists:
List processing is simultaneously three things in the development of computer science. (1) It is the creation of a genuine dynamic memory structure in a machine that had heretofore been perceived as having fixed structure. It added to our ensemble of operations those that built and modified structure in addition to those that replaced and changed content. (2) It was an early demonstration of the basic abstraction that a computer consists of a set of data types and a set of operations proper to these data types, so that a computational system should employ whatever data types are appropriate to the application, independent of the underlying machine. (3) List processing produced a model of designation, thus defining symbol manipulation in the sense in which we use this concept in computer science today.
Brooks:
The default assumption has been that the perception system delivers a description of the world in terms of typed, named individuals and their relationships. For instance in the classic monkeys and bananas problem, the world description is in terms of boxes, bananas, and aboveness… The effect of the symbol system hypothesis has been to encourage vision researchers to quest after the goal of a general purpose vision system which delivers complete descriptions of the world in symbolic form. Only recently has there been a movement toward active vision which is much more task dependent or task driven.
Intelligence in the world:
To build a system that is intelligent it is necessary to have its representations grounded in the physical world. Our experience with this approach is that once this commitment is made, the need for traditional symbolic representations soon fades entirely. The key observation is that the world is its own best model. It is always exactly up to date. It always contains ever detail there is to be known. The trick is to sense it appropriately and often enough.
[The evolutionary time-line of adaptive behavior] suggests that problem solving behavior, language, expert knowledge and application, and reason, are all rather simple once the essence of being and reacting are available. That essence is the ability to move around in a dynamic environment, sensing the surroundings to a degree sufficient to achieve the necessary maintenance of life and reproduction. This part of intelligence is where evolution has concentrated its time - it is much harder. This is the physically grounded part of animal systems.
Herbert the robot cleans up your office:
The laser-based soda-can object finder drove the robot so that its arm was lined up in front of the soda can. But it did not tell the arm controller that there was now a soda can ready to be picked up. Rather, the arm behaviors monitored the shaft encoders on the wheels, and when they noticed that there was no body motion, initiated motions of the arm, which in turn triggered other behaviors, so that eventually the robot would pick up the soda can
The advantage of this approach is that there is no need to set up internal expectations for what is going to happen next; this means that the control system can both (1) be naturally opportunistic if fortuitous circumstances present themselves, and (2) it can easily respond to changed circumstances, such as some other object approaching it on a collision course.
As one example of how the arm behaviors cascaded upon one another, consider actually grasping a soda can. The hand had a grasp reflex that operated whenever something broke an infrared beam between the fingers. When the arm located a soda can with its local sensors, it simply drove the hand so that the two fingers lined up on either side of the can. The hand then independently grasped the can. Given this arrangement, it was possible for a human to hand a soda can to the robot. As soon as it was grasped, the arm retracted - it did not matter whether it was a soda can that was intentionally grasped, or one that magically appeared. The same opportunism among behaviors let the arm adapt automatically to a wide variety of cluttered desktops, and still successfully find the soda can.
Reaction (Nilsson):
The designation aspect of the PSSH explicitly assumes that, whenever necessary, symbols will be grounded in objects in the environment through the perceptual and effector capabilities of a physical symbol system. Attacks on the PSSH based on its alleged disregard for symbol grounding miss this important point.
Nilsson:
It is often claimed that much (if not most) of human intelligence is based on our ability to make rapid perceptual judgments using pattern recognition… It is difficult to devise symbol-based rules for programming these tasks. Instead, we often use a variety of dynamical, statistical, and neural-network methods that are best explained as processing analog rather than discrete symbolic data…What parts are regarded as symbolic and which parts non-symbolic will depend on choices of the most parsimonious vocabulary and the most useful programming constructs - which, after all, are intimately linked.
But this seems like another place where the typical practice of using symbols has replaced the more abstract definition that Newell and Simon explicitly give. According to the actual definition, a physical symbol system is just a collection of structured patterns that evolve according to a collection of processes that operate on them. This seems straightforwardly to embrace patterns that represent numerical quantities and processes that update numerical quantities according to mathematical equations. There's nothing discrete in the definition.
A further way Newell and Simon's work has served as a lightning-rod for criticisms of practical models in AI concerns the kinds of meanings that representations should have. This kind of criticism is important for understanding what a symbol could be and what the physical symbol system hypothesis actually commits you to.
From Agre:
Representations in an agent's mind have been understood as models that correspond to the outside world through a systematic mapping [as in logic]. As a result, the meanings of an agent's representations can be determined independently of its current location, attitudes, or goals. Reference has been a marginal concern within this picture, either assimilated to sense or simply posited through the operation of simulated worlds in which symbols automatically connect to their referents.
A designer moves back and forth between two activities: synthesis of machinery and analysis of dynamics. The question of intentionality arises when it comes time to understand how one's artifact will interact with the people, places, and things that will populate its environment. The artifact will most likely interact with a given object in virtue of the object's role in the artifact's activities, not through the object's objective identity.
Conventional design practice in AI understands this phenomenon of time-extended relationships between agents and objects in terms of the problem of "keeping track" of an objective individual. An agent might have in its head a symbol such as SHIRT32 that serves three purposes: (1) to name a certain shirt, (2) to participate in the logical sentences that express knowledge of that shirt, and (3) to mediate concrete interactions with that shirt on particular occasions of buying it, wearing it, washing it, ironing it, and finally packing it off to a charitable organization.
If the purpose of representation is to express states of affairs in the world, the principal criterion on representational languages is "expressive power". For example, one might invent a new representation language to express temporal and causal relationships, certainty and uncertainty of beliefs, default assumptions when definite knowledge is lacking, or beliefs about the mental states of other people. One demonstrates the power of these languages by exhibiting their capacity to express certain scenarios that previous languages cannot. Having formulated a mathematical semantics for one's new language, one may proceed to realize it inside a computer, probably using conventional data structures and pointers, and presumably building computational facilities that perform deduction within the new language.
…but an agent must recognize not only the functional properties of objects but also their indexical properties. In other words, the agent must detect not simply the abstract "functionality" of objects but the role they play in the particular activities in which the agent is currently engaged. A car is a car in a wide variety of circumstances, but it is the-car-I-am-passing only when one is actually in a certain complex relationship to it. Complex as this relation is, though, it is readily perceptible because it places the car-being-passed in a standardized location in one's visual field, where its relevant properties (relative speed, police insignia, etc.) can be registered in standardized ways.
This conception of the role of perception in activity makes strong claims on accounts of visual architecture. The principal purpose of vision, on this view, is not to deliver a world model; nor it is it to identify the objective identities of things; nor is it even to recognize and enumerate the categories into which visible objects might be classified. Instead, the principal purpose of vision is, starting from the agent's intentions and the changing visual field, to register the indexically and functionally significant aspects of the entities in the agent's environment. The work of registering these aspects frequently requires the agent to engage in complex interactions with the environment, and this work is indissociable from the rest of the agent's interactions. Vision is in this sense an active process, not the passive reconstruction of the world as it is projected onto the retina.