* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Processing and Interaction in Robotics
Survey
Document related concepts
Technological singularity wikipedia , lookup
Barbaric Machine Clan Gaiark wikipedia , lookup
Person of Interest (TV series) wikipedia , lookup
Mathematical model wikipedia , lookup
Human–computer interaction wikipedia , lookup
Intelligence explosion wikipedia , lookup
Existential risk from artificial general intelligence wikipedia , lookup
Embodied cognitive science wikipedia , lookup
Ethics of artificial intelligence wikipedia , lookup
Transcript
PROCESSING AND INTERACTION IN ROBOTICS* Francesco Amigoni1, Viola Schiaffonati2, Marco Somalvico3 Artificial Intelligence and Robotics Project; Dipartimento di Elettronica e Informazione; Politecnico di Milano; Piazza Leonardo da Vinci, 32; I-20133 Milano (MI); Italy [1] e-mail: [email protected], tel: (39) (2) 2399-3622; fax: (39) (2) 2399-3411 [2] tel: (39) (2) 2399-3622 [3] e-mail: [email protected], tel: (39) (2) 2399-3524; fax: (39) (2) 2399-3411 ABSTRACT Sensor science is a modern discipline which addresses issues, problems and solutions involving various facets of the phenomenon of perception of phenomena. Perception is oriented toward natural phenomena, both external and internal to man, moreover, perception is also oriented toward artifacts devised by man. In both cases any sensor device is itself an artifact that man has conceived and designed in order to substitute and to emulate his own sensory organs. The purpose of the paper is to present a perspective of sensor science oriented toward the emulation of man as a subject able to perform, firstly, interactive activities between himself, from one side, and the external world together with himself and other men, from the other side, and, secondly, intellectual functions driven by the sensorial aspects of interaction and directed to the actuatorial aspects of interaction. The paper, therefore, addresses the analysis of sensor science in relation with the most advanced artificial system, developed in the history of science by man in addressing the emulation of man, namely the robot. The global view, oriented to understand the relationship existing between the three hierarchically arranged disciplines of information science, robotics and informatics, from one side, and sensor science, from another side, is a very interesting arena for discussing in a novel approach the modern and dynamic outlook of sensor science. KEYWORDS * This paper has been presented as invited paper at Eurosensors XI, Warsaw, Poland, September 1997. Artificial perception of the world and of man, Machines and metamachines, Models of phenomena, Sensors and information sciences. 1 - INTRODUCTION Sensor science is a modern discipline which addresses issues, problems and solutions involving various facets of the phenomenon of perception of phenomena. The perception is oriented, accordingly with the needs of man, in various phenomenological directions. Sometimes, perception is oriented toward nature, both external and internal to man, sometimes perception is oriented toward artifacts which are the results of previous invention of man. Anyway, any sensor device is itself an artifact that man has conceived and designed in order to substitute and to emulate his own natural sensors. The purpose of the paper is to present a perspective of sensor science oriented toward the emulation of man as a subject able to perform, firstly, interactive activities between himself, from one side, and the external world together with himself and other men, from the other side, and, secondly, intellectual functions driven by the sensorial aspects of interaction and directed to the actuatorial aspects of interaction. The very general scenario [1] of emulating, with an artifact, the intelligent and interactive functions of man is the framework where we want to set up a strategic evaluation of sensor science within its various perspectives. The paper, therefore, addresses the analysis of sensor science in relation with the most advanced artificial system, developed in the history of science by man in addressing the emulation of man, namely the robot. The main activities of the robot [2] involved in emulating man, namely processing and interacting, are therefore illustrated in order to understand the important relationship existing among sensor science, robotics, informatics and information science. The global view, oriented to understand the relationship existing between the three hierarchically arranged disciplines of information science, robotics and informatics, from one side, and sensor science, from another side, is a very interesting arena for discussing in a novel approach the modern and dynamic outlook of sensor science. It is clear, however, that this quite broad approach requires an uncommon but necessary analysis of artificial emulation of man provided by computers (processing machines) and by robots (processing and interacting machines), with discussion spreading from the question “who a man is”, involving humanities, and “what a robot and a computer are” involving sciences. The power of the outlook provided in this novel approach toward understanding the future of sensor science is thus centered on an interesting synthesis between the culture involved by humanities and sciences. Moreover, this assertion enables the demonstration of the validity and the modernity of the proposed illustration of sensor science, since it is today clearly accepted, within culture discussion around the world, the agreement of the necessity of unifying culture and of exploring, with novel and interesting results, conceptual and experimental problems lying in the borderline between humanities and sciences. The paper is organized as follows. Section 2 presents a brief philosophical survey on the concept of machine and on the discipline that tries to emulate some of man intellectual performances by means of machines, namely Artificial Intelligence (AI). In Section 3 we propose a very general 2 taxonomy of reality into three classes denoted by man, machine and the world. Section 4 is about the concept of model which provides a description of the relationship between man and the world. Section 5 introduces the machine as a particular way of representing a model. A particular class of machines, namely information machines are presented in Section 6 together with some of their classifications. Section 7 deals with the proposed novel role of sensor science. Finally, Section 8, illustrates some issues of open research. 2 - STATE OF THE ART 2.1- Machines It can be difficult to define what a machine is, nevertheless it is fundamental to find a definition of machine in general, and of information machine in particular. The interest on information machines refers to its implications within artificial intelligence. The concept of machine is essential if we are discussing of the intelligence of man and of the possibility to build machines able to emulate performances of human intelligence. This discussion derives from philosophy, in particular from two opposite parts of it. On one side there are philosophers - Descartes (1596-1650) [3] [4] who is the most important - who consider the dualism between ‘res cogitans’ and ‘res extensa’, namely between a thinking soul and an extended matter. They deny that a man can be considered as a sort of machine because he always has his own soul and he is not only matter. On the other side, the French philosopher La Metrie (1709-1751) [5] in his book “L’homme machine” (1748) proposes that the distinction between soul and body does not exist and everything in the world is matter; in this way, the man can be considered as a mechanism and explained only through changes of the body. Without thinking whether a man can be considered as a machine or not, we adopt a general definition by Gregory [6]. A machine is a functional entity composed of determinate parts: if we know these parts and their interactions, we can understand its internal functions. In order to understand external functions of a machine, it is necessary to know the environment (surroundings) where the machine works. The dilemma about the Descartes and the La Metrie approaches will be encompassed and resolved in a novel presentation of the concept of machine which will move from the basic concepts of model, of architecture and of machine in order to define information machines as the class of machines necessary for philosophical discussion on AI. 2.2- A Short History of Artificial Intelligence Artificial intelligence was conceived in 1956 as a new scientific discipline during a summer seminar at the Dartmouth College, organized by Minsky, McCarthy, Rochester, Shannon. The aim of the seminar was to examine some future aspects, especially performances, of the machines (namely computers). Many methodological problems arose at once; there is still a strong debate on the best way to define AI ambits of program research. 3 Nowadays, AI is both an experimental discipline between engineering (namely its branch studying computers) and a science which studies basic principles of intelligence [7]. Therefore, there are many different lines of approach: from programs to solve problems that present human intellectual difficulties, like playing chess, to studies of intellectual mechanisms and techniques of problem solving that exercise these mechanisms. In any case, it is necessary to be clear about when we can say that an entity is intelligent [8]. We adopt McCarthy’s [9] definition: “an entity is intelligent if it has an adequate model of the world (including the intellectual world of mathematics, understanding of its own goals and other mental processes), if it is clever enough to answer a wide variety of questions on the basis of this model, if it can get additional information from the external world when required, and can perform such tasks in the external world as its goals demand and its physical abilities permit”. If we consider intelligence as a natural kind, it appears like a phenomenon that AI can model. According to this definition we can say that AI is composed of two parts: the epistemological part and the heuristic part [10]. The epistemological part concerns the representation of the world and its modelisation and shows that AI concerns information machines. The heuristic part concerns the methods to find a solution and to decide what to do. The representation of the world is fundamental for understanding what the heuristic part can do on the basis of information. This process identifies an abstract activity which can be helped by philosophy; hence it is important to consider what the philosophers had to say. Philosophy must be summoned when we are deciding how much of a phenomenon can be represented in its model and how much of modelisation can be done, which is the problem of the discussion between weak and strong AI. Our description of the world depends on our beliefs: common-sense sentences and scientific expressions must be accepted, without falling in the logical positivism that makes describing phenomena of the world too simple. 2.3 The Philosophy of Artificial Intelligence We adopt McCarthy’s proposal of the philosophy of artificial intelligence [11]: a human level computer program of AI requires philosophical attitudes, especially epistemological ones, which are the only chance for performing with the heuristic ability. With regard to an intelligent machine it is necessary to formalize many aspects like beliefs, hopes and goals; thus a philosophical point of view becomes fundamental. The problem of the perturbation levels and of the amplitude of new intelligent discoveries is called the frame problem [12] [9]: a new discovery implies a revision of the previous knowledge: how much of this must we change to conform it to the new discoveries? The frame problem is connected with monotony and non monotony of the incremental growth of the intelligent results. Incrementality versus time and versus space of an intelligent process can be derived from the definition of intelligence given by McCarthy. The frame problem is typically found when we discuss about the possibility of building a machine emulating human intelligence. Within the study of interaction between AI and philosophy we can drawn a preliminary explanation of the conditions that clarify the understanding of the frame problem. 4 3 - REALITY, NATURAL AND ARTIFICIAL 3.1 - Bipartition of the reality The set of all things that exist in the world is named reality. More precisely, the reality is the set of what is perceived by humans as present in the world. The reality can be considered, as well, as a source of various types of phenomena that man can perceive. It is useful to consider the reality divided into two component parts: the natural and the artificial (see Figure 1). The possibility to distinguish the two component parts is very ancient: it takes place in the Ancient Greece where rational thought was originated in V century BC. The transition from myth to philosophy and so to rationality creates the gap between man and the world (intended as the natural outside man). While in the mythic age man was one with the nature and the divinities, with the beginning of philosophical thought the man becomes ‘different’ from the world; he is no more one with the nature and he can act on it. By relating in this way with the world, man observes natural phenomena and tries to explain them. This process only allows to find rational answers to questions and therefore it pushes toward a scientific achievement. In this way it is possible to distinguish not only between man and the world, but also among man, the world and machine. Moreover each machine can be created by man to work on the nature. We intend the natural as the composition of man and of the world; in addition we consider the artificial referring to the notion of machine, so we can say that the artificial is the set of artifacts (i.e. machines) devised by the humans. The preliminary distinction between man and the world is fundamental to understand man’s acting on the world and so to divide the reality in the natural and the artificial and to speak about machines. They appear immediately as artificial objects: in fact man has planned them to achieve human means, they have different characteristics from those which are in nature, but have the character of being subjected to the physical laws [6]. 5 REA LITY NA TURA L MAN + WO R LD AR TIFICIA L MACHINE Figure 1 - The division of the reality in the natural and the artificial. 3.2 - Interactions among man, machine and the world The three fundamental entities of the reality, namely man, machine and the world, interact each other as illustrated in Figure 2. The diagram, that can be interpreted as a new representation of the reality, points out the relations among the three actors and, more precisely, their interactions. Man interacts with the world by means of sense-organs (such as: eyes, ears, ...) and of activeorgans (such as: arm, leg, hand, ...), the first ones perceive the phenomena of the world, the second ones produce phenomena in the world. Machine interacts with the world by extracting information about the phenomena happening in the world, in this case it exploits the sensors, moreover the machine can produce phenomena in the world by means of the actuators. The third type of interaction is between man and machine, in this case we can say that, similarly to the interaction between machine and the world, machine perceives man through the sensors (since man is part of the natural) and produces effects on man by means of the actuators. This point of view is innovative, in fact the sensors and the actuators are usually regarded as referring to the world, and not to the entire natural. This is the reason why, in Figure 2, we have named the extended use of the sensors and of the actuators in parentheses ((SENSORS) and (ACTUATORS)). 6 MA N W ORLD (S E NS O R S ) A CTUA TO R S S E NS O R S (A CTUA TO R S ) M A CHINE Figure 2 - Interactions among man, machine and the world. 4 - PHENOMENON, MODEL AND LAW In the field of physics there is a great difference between a phenomenon and a model of the phenomenon, the first one is perceived, actually known, by man. The knowledge of the phenomenon is described in the model. Hence, a model defines, in a rigorous way, the knowledge, considered relevant for describing the phenomenon, acquired by the man. The empirical inductive deductive paradigm (see Figure 3), introduced by Galileo Galilei (15641642), is a good description of the relations and the differences between a phenomenon and a model of the phenomenon. In fact in “Dialogo dei massimi sistemi” [13] written in 1632 he states: “Bisogna veder col senso di che non dubitava l'intelletto” (it is necessary to perceive with senses what is clear to intellect), hence scientific knowledge is not a generalization of experience and observed phenomena, but it finds its validation when its discoveries are confirmed by experience. Starting from a phenomenon occurring in the reality, its model is devised by man with an intellectual process called abduction. Thus, the abduction describes the passage from a phenomenon to its model. Therefore, we are considering abduction as a bridge over a gap dividing two entities: the reality and the knowledge of reality. The abduction is a general process which can be called, as well, formalization in the case of physical sciences. 7 Within the knowledge of reality, man performs another intellectual process on the model in order to produce a law. The law is a new created element of the knowledge of reality. The intellectual process from the model to the law is called induction or deduction, depending whether it is a passage from lower to higher abstract knowledge or vice versa. We can call inference either induction or deduction. Inference identifies a passage within the knowledge of reality. Inference can be called, as well, in physical sciences, derivation. The new knowledge of reality, described by the law, can be utilized for predicting a new phenomenon of the reality. The intellectual process from the law to the phenomenon is called adduction, and it represents the passage from the knowledge of reality to the reality. The adduction can be called, as well, in physical sciences, expectation. The activities of abduction and adduction are called afference, whereas, as we mentioned above, the activities of induction and deduction are indicated as inference. ‘Afference’ is a word analogous to ‘inference’, in fact both derive from the Latin root ‘ferre’ which means ‘carry’ in English. The union of afference and inference intended as two general intellectual processes constitute the global process of intelligence. If we consider, on the basis of Bergson's approach [14], the intelligence as divided in two types, namely creative intelligence (or creating intelligence, from the verb ‘to create’) and fabricative intelligence (or fabricating intelligence, from the verb ‘to fabricate’), it is obvious that afference and inference, previously introduced, can be envisaged as having the same meaning of creative intelligence and fabricative intelligence, respectively. Thus, when man observes a phenomenon (where the phenomenon belongs to the reality) and when he creates a model of the phenomenon (where the model belongs to the knowledge of reality), the intelligence involved in the passage from the phenomenon to the model is the creative intelligence. The creative intelligence is exploited, as well, when man uses a law (where the law belongs to the knowledge of reality), and he expects that a phenomenon will take place. On the other hand, when man builds a law starting from a model (both the model and the law belong to the knowledge of reality), then the intelligence which is carried on is the fabricative intelligence. We stress that, from our point of view, the afference (i.e. the creative intelligence) is an activity performed exclusively by man, whereas the inference (i.e. the fabricative intelligence) is an activity that can be performed either by a man or by a machine. The whole empirical inductive deductive paradigm is subject to the metaactivity, belonging to the creative intelligence, of critique; it consists in revising and modifying the cycle of abduction, inference and adduction into a new similar cycle, actually an improved one with respect to the preceding cycle. The improvement is evolutionary, namely it permits a better knowledge of the reality that is exploited in an improved abduction, in a more effective inference and in a more satisfactory adduction. The empirical inductive deductive paradigm can be applied to any other experimental discipline, in general, and therefore to AI, in particular. In this case the paradigm explains the problem resolution activity performed by man together with machine. In AI the starting point is the intuitive problem which corresponds, within the paradigm, to the phenomenon in physics. The intuitive problem identifies an exigency arising in a man together with the desire of solving it. The formalization substitutes the intuitive problem with the represented problem, corresponding to the model in physics. The represented problem is successively solved in order to obtain the solved problem corresponding to the law in physics. In AI, as in the case of physics, the activities of afference and of critique are performed exclusively by man, whereas the activity of inference is performed either by man or by machine. 8 MODEL ABDUCTION INDUCTION DEDUCTION PHOENOMENON ADDUCTION GAP {REALITY} LAW {KNOWLEDGE OF REALITY} Figure 3 - Galileo Galilei: empirical inductive deductive paradigm. 5 - MODELS AND MACHINES As we have illustrated in the previous Section, a model is to be intended, in the Galileo Galilei's spirit, as a type of knowledge of a phenomenon of the reality. A model of a phenomenon is, then, quite different from, and could not be identified with, the phenomenon [15]. A model is: 1 finite: only a finite content of knowledge, extracted from a potentially infinite source of information, is drawn out from the phenomenon and it is inserted in the model; 2 objective: the model is such that anyone has the same perception of the truth embedded in the model itself; 3 experimentable: the model can be used in order to predict the happening of a new phenomenon. We think of a model as having the following three properties: 1 a model is perfect within itself, because it is built up as a form shaped by utilizing a formalism; we can adopt various formalisms such as mathematics, logic, etc.; 2 a model is imperfect in knowing a phenomenon of the reality, because the abduction is a creative process that expresses the knowledge embedded within the model, describing only some of the elements which contribute to the perception of the phenomenon; it is clear that the abduction cannot produce a model rich enough to describe the whole phenomenon; 9 3 a model is perfectible, since it can be indefinitely substituted with a better (in the sense of less imperfect) model. The new model is the result of a new abduction (namely a pulse of intelligence, creative in its nature) which takes into account additional elements of the real phenomenon. The process can be iterated in order to obtain a sequence of models each one better than the previous one but worse than the subsequent one. The illustrated paradigm is of great relevance, not only in discussing the types of intelligence that the man adopts in any experimental discipline, but it plays a fundamental role also in representing an interesting approach to the conception of any type of the artificial entities that we call machines. The key point consists in the notion of model, previously explained, and in the observation that each machine can be considered as a particular way of describing a model of a phenomenon. More precisely, a machine is an artificial entity which is a reification of a model of a phenomenon. It is important to point out the role and the meaning of the term ‘reification’, which derives from the Latin term ‘res’ that means ‘thing’. Thus, a machine is something made up as a thing that plays the role of describing a model. We further note that, since any model is made up by shaping a form within a formalism, in the case of a machine the form of the model is the architecture of the machine, whereas the formalism of the model is the set of components of the machine. In fact the way of composing together many components within an architecture provides the unique composite entity, namely the machine, which produces a global performance or function that is the result of the elementary functions that each component provides, composed together within the architectural design. It is therefore interesting to conclude that the performance of the unique composite machine is indeed the emulation (namely a partial reproduction) of the phenomenon whose model has been reified by the machine. Please note that we don't utilize the stronger wording of ‘simulation’ since, as previously stated, any model is just one expression of an imperfect knowledge about a phenomenon intended as the whole entity that we would like to known. Emulation is, thus, expressing how imprecise can be the replica of a phenomenon which is produced when a machine performs its functions. 6 - INFORMATION MACHINES 6.1 - Machines and Metamachines In the previous Section we have presented the important link existing between the concept of model, as it has been clearly illustrated within the Galileo Galilei paradigm, and the concept of machine, namely the particular type of artifact whose architecture (the form), composing together the individual functionalities of the various components (the formalism), describes the model of a phenomenon of the reality which is emulated by the functionality of the machine, which is, hence, the reification of the model of this phenomenon. In order to further carrying on our epistemological analysis of the notion of machine, it is now necessary to introduce a classification of the phenomena whose models can be embedded into the machines’ architectures. The most relevant classification distinguishes between the phenomena of the world distinct from man and the phenomena of man. 10 We will call world machines or Machines the machines whose architectures embed models of the phenomena of the world. We will call man machines or Metamachines the machines whose architectures embed models of the phenomena of man. While a Machine can be viewed as a rough (it just emulates) substitute of an element of the world distinct from the man, a Metamachine is a rough substitute of an element of man. In order to deeply understand the role of Metamachines, we have to address the identification of the phenomena that man observes in man himself and which he is able to model in a satisfactory way, where satisfactory means that the model, although imperfect, is a sufficiently clear rational description and satisfactory approximated emulation of the phenomenon observed by man within himself. It is evident, after the classification previously illustrated of the human intelligence into two categories, namely the creative intelligence and the fabricative intelligence, that we can conclude in identifying the phenomena of man, apt to be modeled and reified into Metamachine architectures, as the phenomena belonging to fabricative intelligence (or the intelligence of the ‘Homo Faber’ (the man who constructs)). Thus, we have illustrated that Metamachines are the rough emulators of the particular human phenomena which arise when man performs the intellectual activities of fabricative intelligence, namely the rational intellectual activity that man performs with his induction and deduction on a model of a phenomenon for constructing a law coherent within the model, both after having invented the model with abduction and before having utilized the law in interpreting new phenomena with adduction. We can describe the scenario by stating that, while man performs directly the intellectual activities of creative intelligence (or afference) immediately (without a medium) within his own body and within his own mind, the man can perform indirectly some of the intellectual activities of fabricative intelligence (or inference) mediately (with a medium) within a Metamachine, actually his Metamachine because man designs and constructs the Metamachine and because man makes the Metamachine to perform its own performance. It is now evident that the operations performed by a Metamachine, intended as an operator, are addressed toward operands which are models of phenomena of the reality since fabricative intelligence starts from the input models that man provides with his creative intelligence. It is now possible to assert the fact that any Metamachine is an operator emulating an intellectual activity of the fabricative type which acts upon operands which are models of phenomena of the reality. We can observe that the characteristic of each model, intended as operand, is to be described in a formalism which consists of a language enabling man to communicate to the Metamachine the model as an operand. The typical name that we adopt for such class of models is information, namely those models which are communicated by man to a Metamachine in order to become the operands of the Metamachine. In a parallel way we will adopt for the Metamachine the corresponding denotation of information machine. 6.2 - Information The word ‘information’ can have four different meanings: it is important to analyze them to understand the original meaning of the word. This operation allows us to find out the connection between science and Greek-Hellenistic culture and to examine the cognitive paradigm which 11 considers the thought like a manipulation of representations [16]. In this way it is possible to conceive the thought as an operation on symbols and to think about the mechanization of it that brings us toward the information machines. The four meanings of ‘information’ are the following. 1 In a conventional meaning, information is some knowledge, about the reality, that is made available. 2 In a scientific-technical meaning, information is a signal, a marker, something that conveys knowledge, following the work of Shannon [17] and his information theory. 3 In an operational meaning, information describes the operand on which a particular machine, namely an information machine, performs its operational activities. 4 In a more essential meaning, which summarize the previous meanings within a more general assumption, information is the act or effect of informing or being informed. The word derives from the Latin verb ‘informare’ which means to give a form. The process that brings the word ‘form’ (in Greek ‘eìdon’, from which derives the English ‘idea’) from the meaning of ‘ontological foundation’ (classical conception) to ‘content of thought’ (empirical conception) is the condition to create autonomous information machines. We conclude the Subsection remarking that, as previously indicated, among the various ways in which we may conceive models, there is a particular one which individuates information as the kind of model description which enables the model to be utilized as operand of a Metamachine, thus called information machine. In other words, information is a peculiar class of models of phenomena of the reality which can be made to become operands subject under operations performed by a Metamachine. After this explanation we can better understand and further expand the role of information in the context of our paper. 6.3 - The computer and the robot As pointed out in the previous Subsection, an information machine (or a Metamachine) is a machine which performs, as an operator, some operations on operands which are information. A general classification of information machines can be introduced by appropriately classifying the types of phenomena which, modeled as information, are the operands subject to the operations performed by information machines. On the basis of our previous discussion of the two types of intelligence, we propose the epistemological thesis that the creative intelligence can not be satisfactorily modeled, and, by consequence, we consider inappropriate, for the operation of an information machine, the activity of invention of a new model. However, it is possible to perform on the models invented by man, with abduction, the following four types of operations. 1 Transformation, namely the class of operations on models that consist in getting a model, namely a given form shaped within a given formalism, and in substituting it with another model made by shaping the same form of the previous model within a new formalism, thus providing a model that is the transformation of the previous one. The particular discipline involved in designing information machines which perform transformation is called electronics. 2 Processing, namely the class of operations on models that, under the common denotation of information, contemplate three types of information and, thus, three types of processing: 12 2.1 algorithm as a type of information and execution of algorithm as a type of processing; 2.2 data as a type of information and management of data as a type of processing; 2.3 problem (or knowledge) as a type of information and solution (or inference) as a type of processing. It is interesting to add to this classification the case in which processing is integrated with interaction between information machine and the external world. Thus, we call the resulting class of operation interprocessing. The interaction, specific of interprocessing, between machine and the external world has not to be confused with another type of interaction, existing between machine and man, which is typical both of machines performing processing and interprocessing. The particular discipline involved in designing information machines which perform processing is called informatics or computer science. The particular discipline involved in designing information machines which perform interprocessing is called robotics (or interinformatics). 3. Control, namely the class of operations on models of potentially influenced phenomena that external influencing phenomena can disturb: the activity of control is dedicated to maintain the same values of some typical parameters of the models of the influenced phenomena. The particular discipline involved in designing information machines which perform control is called control science. 4. Communication, namely the class of operations on models that transfer the knowledge of a model from one subject (a man or a Metamachine) to another subject. The particular discipline involved in designing information machines which perform communication is telecommunications. The information machines which perform processing or interprocessing can, therefore, be divided into two classes, the first one composed of computers, and the second one composed of robots. A computer (also called processor) is an information machine which performs, as an operator, the operation of processing of models; the processors characterize the field of informatics or computer science. A robot (also called interprocessor) is an information machine which performs, as an operator, the operations of processing of models and of interaction by means of models; the robots characterize the area of robotics. The interaction is composed of perception of phenomena, whose models are established within the architectures of sensors, and of production of phenomena, whose models are established within the architectures of actuators. Now it is clear the nature of the relations among man, the world and machine illustrated in Subsection 3.1. In Figure 2 we have three types of interactions. The first one, existing between man and the world and represented by dotted arrows, indicates not only that man and the world are source of phenomena not yet modeled but that the interaction between man and the world is itself a phenomenon. The second one, existing between man and machine and represented by continue arrows, is already a model because one of the interacting entities is a machine, which is a model (actually a reification of a model) and it implies that the way in which machine interacts with any other entity is also a model. The third one, existing between machine and the world and represented by continue arrows, is already a model because, as in the previous case, one of the interacting entities is a machine. 13 We can now understand why we have utilized two distinct types of arrows: while the dotted arrows represent phenomena of interaction between man and the world, the continue arrows represent models of interaction between machine and the world or man. It is also important to clarify that, as said before, the modeled interaction between a man and an information machine always exists. This is true first at all because “the man makes the machine” since the man creatively invents the model which is reified in the machine. This is true as well because “the man makes the machine to perform” since the following statements hold. 1 The man decides which types of operand (information) are sent to the machine (thus defining the syntactic aspects, namely the form of the sent model). 2 The man activates some of the potentially available functions of the machine (thus defining the semantic aspects, in a operational way, namely the particular functions identified by the man within a sent model). 3 The man, by sending the model and by activating the functions obtains from the machine the delivery of a given performance (thus defining the pragmatic aspects, namely the fulfillment of the goal that has driven the man to utilize the machine in a given way). Thus, the second part of the man intent, namely the fact that “the man makes the machine to perform”, requires the necessity that interaction between man and information machine is always existing. On the other hand, the interaction between information machine and the world is not necessary. By examining the second class of information machines we will therefore distinguish between the following two cases. 1 The case in which there is no interaction between machine and the world, that is represented by computers belonging to informatics. 2 The case in which there is interaction between machine and the world, that is represented by robots belonging to robotics. We are now in the position to introduce a classification of the information machines of informatics and of robotics, namely the computers and the robots. The classification is based on the presence or absence of the sensors or the actuators. Therefore the criteria for classifying the information machines of informatics and of robotics are in the type of interaction existing between machine and the world. Thus we have the following cases (see Figure 4). 1 Computer (or processor), namely the information machine of informatics which has no interaction between itself and the world and, thus, which has neither sensors nor actuators. 2 Robot (or interprocessor), namely the information machine of robotics which has interaction between itself and the world, with sensors, or with actuators, or with both sensors and actuators. It is therefore appropriate to introduce a further subclassification of robots as follows: 2.1 black robot, namely a robot with no sensors and with actuators only. Please note that ‘black’ is here utilized in order to stress the fact that a black robot, having no perception, is ‘similar’ to a ‘blind’ individual; 2.2 blue robot, namely a robot with both sensors and actuators. Please note that ‘blue’ is here utilized in order to stress the fact that a blue robot, having both the activities of perceiving (with sensors) and of producing (with actuators) models of phenomena, is ‘similar’ to a ‘blue collar worker’ who, in the assembly of a factory, performs simple operations, guided by his intellect, his natural sensors and his natural actuators, in carrying on a task of industrial production; 14 2.3 white robot, namely a robot with no actuators and with sensors only. Please note that ‘white’ is here utilized in order to stress the ‘similarity’ of a white robot to a ‘white collar worker’ who is not involved in performing a task of industrial production (similarly to the blue collar worker). This fact is a consequence of the absence of actuators, which are needed to perform any task of industrial production. The white robot has only the sensors which are used to observe, to monitor and to control the activities performed by other machines (and by other information machines, such as black robots, blue robots and computers). Thus, the white robot brings up the results of such observations, monitoring and controls to the attention of man. In fact a white robots has no actuators, and, thus, it has no output toward world, while its only output is toward man. White robot is an artificial assistant which supports man (its assisted individual) with the best decisions that man has to take in order to carry on his best way of utilizing and evaluating the artificial activities of machines and of information machines. We can conclude that the first type of robot has a monodirectional interaction as output, the second type of robot has a bidirectional interaction of input and of output, whereas the third type of robot has a monodirectional interaction of input. COMPUTER (PROCESSOR) PROCESSING ROBOT (INTERPROCESSOR) PROCESSING AND INTERACTION (types) BLACK BLUE WHITE ACTUATORS SENSORS NO NO YES YES NO NO YES YES Figure 4 - The computer and the three types of robots. 7 - INFORMATICS, ROBOTICS AND SENSORS SCIENCE 7.1 - Informatics and robotics From the considerations presented in the preceding Sections, we may underline the fact that a robot contains, as a part of itself, a computer with all its functions. However a robot is able to add to the functions of a computer, namely the processing, that it fully can perform, a new set of functions, namely those devoted to perception and production of models of phenomena. Thus we can state that robotics (or interinformatics) is a discipline more general than informatics or computer science. As illustrated in Figure 5, robotics, namely the discipline that deals with robots, expands in two directions informatics, namely the discipline that deals with computers; the two new directions are 15 related with the perception of phenomena of the reality by means of sensors, and with the production of phenomena of the reality by means of actuators. PERCEPTIO N (SENSO RS) PR O CESSING (CO MPUTER) PR O D UCTIO N (ACTUATO RS) IN FORMATICS ROBOTIC S Figure 5 - Robotics extends the functions of informatics. 7.2 - The role of sensor science In the previous Section 6, we have presented the taxonomy of the various information disciplines (electronics, informatics and robotics, control science, telecommunications). We have presented, as well, the refinement of the taxonomy of informatics and robotics and we have examined the four different configurations of the interfaces between the world and machine represented by the presence or the absence of the sensors or the actuators. In this Subsection we want to address, in a novel way, the role of sensor science, with the whole set of its various implications and subdivisions, at the light of informatics or computer science and of robotics (or interinformatics). Since sensor science is actually the “science of perception”, it is clear that sensor science is important within both informatics and robotics, for its role in designing techniques which enable the machine (both the computer and the robot) to perform the perception of phenomena produced 16 by man as one of the two aspects of man-machine interaction (the other aspect being the production of phenomena perceived by man). Moreover, sensor science is important, as well, within robotics only, for its role in designing techniques which enable the machine (the robot) to perform the perception of phenomena produced by the world as one of two aspects of the world-machine interaction (the other aspect being the production of phenomena in the world). In this way, we stress the well known consideration that identifies, within sensor science, an important subdivision which addresses the problems of designing sensors of robots, intended as the machines (submachines of the robot machines) which perform the activity of perception of the world phenomena. Moreover, we introduce, with a new light, a further role of sensors, which brings sensor science to play a new role both in informatics and in robotics. This role, as previously indicated, lies in the area of perception of phenomena produced by man. In fact, machines of informatics, namely computers, and machines of robotics, namely robots, interact with man and, therefore, computers and robots utilize input and output interfaces, which we can rename as (sensors) and (actuators). The parentheses are here introduced in order to distinguish these interfaces of man-machine interaction from the corresponding sensors and actuators (here indicated without parentheses) individuating the interfaces of world-machine interaction (see Figure 2). Thus, we can summarize the three roles of sensors and their corresponding impact within sensor science as follows: - sensors are required in robotics to perceive phenomena from the world; - sensors are required in robotics to perceive phenomena from man; - sensors are required in informatics to perceive phenomena from man. In conclusion, Figure 6 addresses the fact that sensor science individuates two areas (with non void intersection) of impact within informatics and robotics, together with a third area which has no relevance neither within informatics nor within robotics but which is relevant within other disciplines (e.g. nuclear physics). 17 INFORMATIC S S ENS OR S CIEN C E ROBOTIC S Figure 6 - Role of sensor science 8 - ISSUES OF OPEN RESEARCH From the standpoint of the new conception of the sensor’s role in informatics and robotics, we list, in the next Subsections, some of the issues that we deem as relevant in open research. 8.1 - Passive versus active perception We distinguish between passive and active perception. The first one refers to sensors that perceive phenomena that are autonomously happening in the world. The second one refers to sensors that perceive phenomena not autonomously happening in the world, but reactively produced in the world when it is being stimulated by phenomena originated from a source considered as external to the world. More precisely, while the world indicates an environment, unknown to the robot, the robot has the task to explore the world in order to perceive some of the phenomena happening in the world. The outer source, in this case, is represented by the robot itself that, before enabling its sensorial functions, activates the production of some phenomenological stimuli, which produce in the world a corresponding phenomenological reaction which is subject to the perceiving activity that the robot executes in exploring the unknown in the world. The morphology of the stimuli is 18 selected in order to trigger the most appropriate reaction which provides with the greatest impact the information needed to detect the unknown in the world. As an example, we consider the active sensor, developed at Politecnico di Milano [18], that, with a single machine vision, allows to obtain a three-dimensional vision. The machine is able to control the direction in which a laser beam actuator produces an indefinite cone of light. The vision sensor recognizes the reaction of the unknown world, a wall, produced by the stimulus, the cone of light. Thus, the vision apparatus perceives the ellipsis produced on the wall by intersection with the cone of light. Since it is known the exact position of the laser beam actuator and, therefore, the direction of the cone of light and the value of its angle, and since it is possible to measure the length of the two axes of the ellipsis, then the machine can determine the three-dimensional position of the wall. Thus, by means of active perception, we can obtain a monocular three-dimensional vision system. 8.2 - Multisensory data fusion In order to detect more precisely a complex and unknown entity, for instance a complex phenomenon happening in the world (or in man), it can be more appropriate to consider more subphenomena of different nature, which are related to the complex original phenomenon. Each subphenomenon is, therefore, independently perceived within an individual model embedded in an individual sensor. Thus, we are facing the problem of merging together the individuals models of the various component phenomena (i.e. subphenomena) in order to integrate them for obtaining a detailed model for the complex original phenomenon. As an example, we illustrate a machine, developed at IRST [19] [20], that is a rough emulation of a librarian. The machine, also called electronic librarian, utilizes a scanner in order to identify a rented book, and it is provided, as well, with a vision recognition sensor and with a speech recognition sensor in order to identify the face and the voice of the person who is renting the book. By means of a bimodal sensor data fusion, namely the fusion of the data from the vision recognition sensor together with the data from the speech recognition sensor, the machine is able to identify correctly a person with a 90% percentage of success. In the illustrated example, the complex phenomenon to be detected is the identity of a person, while the subphenomena, whose models, by means of their integration, enable us to determine the model of the complex phenomenon, are the face and the voice of the person. 8.3 - Multimedia sensors for hypertext The interaction between man and machine has been mainly based on written and spoken text considered as the only channel of communication that man can utilize for man-machine interaction both with computers and robots. Some extensions of the established interface paradigm are at the center of attention. The first extension is centered around multimedia, where other channels of communication are considered, such as sounds, music, graphics, drawings, photos, videos, audiovideos, and so on. The second extension, that is well investigated, is based on organizing texts, not in a sequential way, but in a hypertextual one, where the texts can be explored along several paths of exploration. 19 The role of sensors in the newly conceived methods of interaction between man and machine is that one of perceiving, for example, the cognitive intentions of the user when he produces a text. The text can be narrative, explanatory, informative, and so on. Since the text does not contain the information about the type of information content that the text itself contains, this information can be added to the text by means of graphic system as the CPP-TRS [21]. By means of the described technique and by exploiting the additional information about the type of the text, that is added to the text itself, the machine is able to perceive the cognitive intentions of the user. 8.4 - Sensors of cognitive indicators Recent advantages in sensors developed either for detecting phenomena in the world or for detecting phenomena in the human body, have suggested to give attention to a variety of new communication channels oriented toward the human body, also called techniques of biofeedback. A first class of advantages, more simple ones, is based on the newly conceived development and use of vision and sound sensors, for detecting the non-textual aspects of the so called ‘body language’, namely prosody, facial expression, arms and hands settings and movements. A second class of advantages, more complex ones, is centered around the adoption of sensors devoted to the biofeedback from brain and other body parts. The human body, and more precisely some of its component parts such as brain, heart and lungs, provides several signals that can be interpreted, after appropriate processing, for detecting stress, attention, boredom in man-machine interaction. The biofeedback is used in an intelligent tutoring system, which has been developed at Politecnico di Milano [22], that is oriented to teach mathematics to disabled people (in particular affected by the Down syndrome). The system is able to identify the cognitive state of the student, by means of a biofeedback sensor which detects the EEG signal from the brain of the student. Thus, the system is able to utilize the detection of stress, attention and boredom in order to select the next question that has to be proposed to the student for maintaining alive his attention. 9 - CONCLUSION In this paper we have discussed various issues, concerning both humanities and sciences, useful to understand strategic perspective around sensor science. We have shown how the kernel issue of artificial emulation of man, with his abilities of both intelligent and interactive entity, provides an enlightening and deep insight of various facets of methodologies and applications in sensor science. The paper has presented, in a balanced way, firstly, a taxonomy of issues for understanding present state of the art and, secondly, an array of future research directions, in order to statically and dynamically clarify the modern scope of sensor science. The authors trust that future investigations will be carried out, along the lines developed in the preceding discussion, by scientists interested, not only in doing research work along the lines indicated, but also in further enhancing attention and results on the issue of unifying the role of humanities and sciences, when the aim of understanding future and strategic broad directions of sciences such as information sciences and sensor science is addressed. 20 ACKNOWLEDGEMENTS This paper is a peculiar blending of scientific and humanistic culture. We are particularly glad to acknowledge the pivotal contribution given by Federico Cabitza who has provided a classical example of very talented and sophisticated integration between sciences and humanities. REFERENCES [1] M. Somalvico, Artificial Intelligence (in Italian), Enciclopedia Treccani Aggiornamento, Treccani, Roma, 1991. [2] M. Somalvico, Robotics: development directions and industrial applications (in Italian), Enciclopedia della Scienza e della Tecnica, Mondadori, Milano, 1984. [3] R. Descartes, Discourse on Method and Meditations on First Philosophy, Donald A. Cress (Translator), Hackett Pub. Co., 3rd Edition, 1993. [4] R. Descartes, Principles of Philosophy, Valentine Miller (Translator), Kluwer Academic Pub, 1984. [5] J. De La Metrie, Man a Machine, Open Court Pub. Co., 1974. [6] R. L. Gregory, Mind in Science, Cambridge University Press, Cambridge, 1981. [7] M. Genesereth and N. Nilsson, Logical Foundations of Artificial Intelligence, Morgan Kaufmann, San Francisco, 1987. [8] A. M. Turing, Computing machinery and intelligence, Mind, 59 (1950), p. 433-460. [9] J. McCarthy and P. J. Hayes, Some philosophical problems from the standpoint of artificial intelligence, in B. Meltzer and D. Michie (eds.), Machine Intelligence, Edimburgh University Press, Edimburgh, 1969, vol. 4, p. 463-502. [10] J. McCarthy, Epistemological problems of artificial intelligence, Proceedings of the Fifth International Joint Conference on Artificial Intelligence, Cambridge, Massachusetts, 1977, p. 1038-1044. 21 [11] J. McCarthy, What has AI in Common with Philosophy?, Proceedings of the Fourteenth International Joint Conference on Artificial Intelligence, Morgan Kaufmann, August 20-25 1995, p. 2041-2044. [12] A. Barr and E. A. Feigenbaum, The handbook of artificial intelligence volume I, AddisonWesley, 1989. [13] Galileo Galilei, Dialogue Concerning the Two Chief World Systems: Ptolemaic and Copernican, Stillman Drake, 1967. [14] H. Bergson, Creative Evolution, University Press of America, 1984. [15] M. Minsky, Matter, Mind and Models, Proceedings of the International Federation of Information Processing Congress, 1965, vol. 1, p. 45-49. [16] V. Pratt, Thinking machines: the evolution of artificial intelligence, Blackwell, Oxford UK, 1987. [17] C. Shannon, A mathematical theory of communication, Bell System Technical Journal, 1948, vol. 27, p. 379-423, p. 623-656. [18] M. Somalvico and C. Viviani, Vision in Computer aided manufacturing (CAM) (in Italian), Automazione e Strumentazione, IXXX, 4, April 1981, p. 272-279. [19] R. Brunelli, D. Falavigna, T. Poggio, L. Stringa, Automatic person recognition by using acoustic and geometric features, Machine Vision and Applications, 8 (1995), p. 317-325. [20] R. Brunelli, D. Falavigna, Person recognition using multiple cues, IEEE Transactions on PAMI, vol. 17, no. 10, October 1995, p. 955-966. [21] G. Tonfoni, Writing as a visual art, Intellect, Oxford, 1994. [22] C. Carbonelli and M. Somalvico, A multidisciplinary platform for developing biofeedback Intelligent Tutoring Systems (in Italian), Proceedings of the Convegno SMAU/UGIS "Lo sviluppo tecnologico al servizio dei disabili” V edizione SMAU95, Milan, Italy, September 23, 1995. BIOGRAPHY Francesco Amigoni got the “Dottore Ingegnere” degree in Informatics Engineering at the Politecnico di Milano (Italy) in 1996, he is currently Ph.D. student in Control Science and Informatics Engineering at the Politecnico di Milano (Italy). His interest is about multiagent systems of distributed artificial intelligence. 22 Viola Schiaffonati is getting a “Dottore” degree in Philosophy at the Università Statale di Milano (Italy). Her interest is about philosophy of science and its connections to artificial intelligence. Marco Somalvico got the “Dottore Ingegnere” degree in Electronics Engineering at the Politecnico di Milano (Italy) in 1965, he is currently full professor of Artificial Intelligence and director of the Artificial Intelligence and Robotics Project at the Politecnico di Milano (Italy). His interest is about multiagent systems, virtual museums and systems for disabled people. 23