THEORIES OF HUMAN COGNITION To Better Understand the Co …my.fit.edu › ~gboy › GAB › Book_chapters_files › UNESCO.Encycl... · 2016-02-26 · Boy, G.A., [2002], Theories

Boy, G.A., [2002], Theories of Human Cognition: To Better Understand the Co-Adaptation of People and Technology, in Knowledge Management, Organizational Intelligence and Learning, and Complexity, edited by L. Douglas Kiel, in Encyclopedia of Life Support Systems

(EOLSS), Developed under the Auspices of the UNESCO, Eolss Publishers, Oxford ,UK, [http://www.eolss.net]

Page 1

THEORIES OF HUMAN COGNITION To Better Understand the Co-Adaptation of People and Technology Guy Boy, European Institute of Cognitive Sciences and Engineering (EURISCO), Toulouse, France Keywords: artificial intelligence, co-adaptation, cognition, cognitive models, human-centered design, human-computer interaction, knowledge management, learning, organizational memory systems, socio-technical systems. Short content list 1. Introduction 2. Automation history and evolution of practices 3. Artifacts embeds human cognition 4. From individual intelligent assistance to multi-agent communication 5. Human memory models as analogs for external memory systems 6. An organizational memory application within the IMAT project 7. Co-development of human and artificial cognitive functions 8. Rehabilitating the Art of Memory 9. Conclusion and perspectives Glossary Activity. Denotes the result of the execution of a task. Activity theory. Russian approach to cognition that considers that people learn from their environment,

and human activity is mediated by surrounding artifacts. Affordances. Property of an object that suggests specific actions to users. Agent. An artifact or a person that/who acts and communicates with other agents of a society. AI. Artificial intelligence is a branch of computer science. Analytical knowledge. Knowledge that can be captured and represented. Art of memory. Mnemotechnique that enables someone to memorize loci and images imprinted on

human memory. AUTO. The AUTO pyramid helps designers and evaluators to analyze human-machine interaction

along four dimensions: Artifacts, Users, Tasks and Organizational environments. Autopoiesis. Model of self-creation that focuses on production mechanisms. CFA. Cognitive function analysis: method that supports human-centered design of safety-critical

systems. Cognitive function. A human or machine function that transforms a task into an activity. It has a role, a

context of validity, and a set of resources. Cognitivism. Main stream of cognitive science that supports cognition as information processing. Co-reliability. Within a multi-agent system deals with co-operation, co-adaptation and co-dependency

among agents. Distributed cognition. Scientific approach that considers human cognition as a distributed process

among a variety of agents. FMS. Flight Management System: on-board navigation system that enables pilots to plan a flight and

manage it.



Page 2

HCA. Human centered automation is an approach to automation that involves human factors at various stages of the design process.

Homeostasis. Approach that originates from cybernetics and focuses on regulation mechanisms. IAS. Intelligent assistant systems support people in the performance of complex tasks. LTM. Long term memory: seen as a mass memory of humans in information processing. OM. An organizational memory embeds knowledge and know-how of an organization. Phenomenology is interested in introspection and human experience. Positivism. Based on the fact that consciousness is not observable, and thus investigations of

consciousness are close to metaphysics. Interested in anything observable. Leads to empiricism. Situational knowledge. According to Dreyfus, situational knowledge cannot be captured and thus

cannot be modeled and simulated using artificial intelligence. STM. Short term memory: seen as a set of buffers of a cognitive processor in information processing. Task. Denotes the required instructions to reach a goal.

1. INTRODUCTION

"Celestial navigation capitalized on the European virtues of mathematical theory and on instruments of high technological sophistication. In contrast, navigation in Oceania emphasized the deliberate refinement of people’s intuitive sense of direction and the learning of direct perceptual cues from the natural environment. For a seaman of Oceania, making a voyage is conceptualized as being within a pattern of islands, the positions of which are represented in his cognitive map." K.G. Oatley.

1.1. Initial motivation

When I was approached to write this chapter, I wondered how I could relate theories of human cognition to life support systems. My background is in cognitive science, computer science and engineering. My everyday work for the last twenty years has been more oriented towards studying human factors in complex dynamic safety-critical systems, and aerospace systems in particular. I have been doing research on human cognition in the context of automation. Automation has been developed to relieve the operator of tedious work and make the system both more efficient and more reliable. Lots of automated tools have reached the hope of their creators. Unfortunately, many of them have introduced new issues. In both cases, i.e., positive and negative use of automation, theories of human cognition have evolved to the increasing need to better understand and predict human adaptation. This chapter was then a new challenge towards using this background to the wider service of life support systems in general. In this chapter, we target life support systems such as human/organizational learning and human-centered design.

1.2. Computers are everywhere in our life

I started from the fact that computers are everywhere, and will probably be even more present during the third millennium. People are faced with the tremendous challenge to adapt to these new artifacts that are becoming more and more invisible. Computers are more invisible because they are becoming integrated and usable. Obtaining money from a cash machine in the street is no longer a complex cognitive process. We have learned to rely on this kind of machine. Sometimes, we wonder if our credit card will come out after a transaction. New automatisms have arisen. New types of issues occur when you either make an error, or you perform an action that is not allowed by the computer program. We are becoming more and more aware of these types of problems. Credit cards certainly help support our lives when they work well, but they happen to be nightmares when they fail to provide what we expect.



Page 3

We need to manage our nerves. Sometimes, we expect things that are not in the contract. This may be due to our background or the non-explicit phrasing of the contract. In other cases, we make a manipulation error and we do not understand the consequences. Finally, the machine itself may be broken or deliver messages that we don’t understand. All these examples are related to human cognition. Since information technology, and automation in particular, evolves very quickly, emerging practices, and information management in particular, do too. A major issue is that many people do not have time to adapt and mature emerging practices rapidly enough with respect to the increasing speed of technology evolution. The management of knowledge and action supported by the use of new information technology leads to the creation of new cognitive functions that will be extensively described in this chapter. Humans are often the victims of new information technology because they do not assimilate and accommodate such cognitive functions in the right way, and/or at the right time. A major issue is the integration of computer technology with the current external memories as extensions of the human memory. Information technology enables knowledge management and storage. Will information technology give birth to better life support systems? What will be the role of human beings in these life support systems? What will be the repercussions of this artifactual evolution on the way humans live? Information-based life support systems are taken within a broad scope including their integration and use at school, home, work and public places for instance.

1.3. Modeling human cognition to rationalize co-adaptation of people and technology

In this chapter, theories of human cognition will be used, and sometimes extended, to rationalize the concurrent adaptation of people and technology. I will use alternatively the terms theory and model, because they are often the same in cognitive science, a field in constant evolution. Theories are often socially recognized models of data from field and laboratory experiments. The notion of model is important since it influences the way experiments are planned and conducted. Both theories of human cognition that support experiments, and experimenters themselves, are crucial in the genesis and interpretation of experimental data. Anytime we try to measure, assess or understand a life process, we attempt to use one or several models; whether these models are implicit or explicit. We refine our measures, assessments or understandings by refining these models. We might end up either with satisfactory models in the end, or with failures that might suggest drastically different directions of investigation. We are constantly in quest of models that support our lives. Some of these models are already available. Many people have tested them. They have become part of our life support conceptual tools. Models of human cognition have been debated for a long time, and still are.

1.4. Outline of the chapter

In section 2, Rasmussen's model will be used to explain historical facts on automation and the emergence of information-based practices. Section 3 will support the claim that artifacts embed human cognition. The study of affordances is very important for human-centered design because no matter what users are required to perform using an artifact, they will attempt to react to artifact affordances. Cognitive modeling is a primary topic of artificial intelligence that strongly supports cognitivism. Dreyfus' criticism of artificial intelligence had the merit to put forward situational knowledge developed and used by expert people, and to show the limits of analytical knowledge that was



Page 4

extensively used by artificial intelligence researchers to develop expert systems. The cognitive function paradigm was developed to take into account situatedness in cognitive modeling. It also fostered the development of function allocation among people and technology. In section 4, we will investigate the current shift from individual intelligent assistance to multi-agent communication in terms of human-machine communication models. Section 5 will present a variety of human memory models to support the analysis, design and evaluation of external memory systems. An example of an organizational memory system of training knowledge will be introduced in section 6. In section 7, we will discuss the co-development of human and artificial cognitive functions by using an evolution of Piaget's action schemes. The search for and satisfaction of the co-reliability of human and artificial agents contributes to the emergence of distributed cognitive activities. Section 8 will present Descartes's heritage and phenomenology (based on human experience) as a background for a new Art of Memory in knowledge management. In the conclusion, these rationalizations will be used to the present an epistemology of organizational cognition. The balance of the chapter provides a concurrent approach of human and organizational learning, and human-centered design.

2. AUTOMATION HISTORY AND EVOLUTION OF PRACTICES

Automation has been a major concern for a long time. The clock is certainly one of the best examples of an old automaton that provides time to people with great precision. People rely on clocks to manage their life. A watch provides precise time information to a user who will not be able to access it otherwise. In addition, a clock may be programmed to autonomously alert its user to wake up for example. People trust clocks, but they have also learned to know when a clock does not work properly. They have learned to interact with such an automaton. During the twentieth century, more sophisticated automata have been created.

2.1. From energy-based to information-based interaction

Several artifacts such as the telephone, the car, the airplane, television, and the computer have made the twentieth century a unique period of our era. All these technologies facilitate communication among people. The last born artifact, the World Wide Web (the Web for short), is certainly the most promising for that matter. Basically, computers are everywhere, and more than ever at the service of communication. The question is not if people use and will use newer information technologies to communicate, but rather how they use it, whether to improve their well-being and to become more aware of the needs of others. This fundamental shift from energy-based use of physical artifacts to information-based use of cognitive artifacts needs to be further investigated. Cognitive artifacts result from the computerization of our living environments. Because more cognitive artifacts were created to help manage our lives, conventional human-machine interaction models may become rapidly obsolete. We have started to observe this trend in studying the evolution of commercial aircraft cockpits and the emergence of new pilots' practices. In the past, pilots needed to calculate and think at the same they needed to act in the short term. Today, they tend to think more than they directly act. In addition, their thinking and (high level) actions are more directed toward long-term action plans that are finally implemented by the machine. Pilots have become flight managers of artificial agents that execute complex tasks that were executed by their predecessors. The emergence of new practices has forced cognitive scientists to take different human-machine interaction models into account.



Page 5

Information technology provides new challenging means to do business. Today, in some countries, someone can order food using the Web from his or her house. What he or she needs to do is to connect himself or herself to the appropriate Web address, browse the various product descriptions, check their prices (he or she may run a spreadsheet at the same time to have the total price of what his or her has selected so far), verify the complete order, press the "Order" button, and finally confirm. This is it. The delivery to his or her kitchen will follow in a few hours, or even minutes in some cases. This short example shows how we are moving from an energy-based world to an information-based world. People previously would have had to write a shopping list on a piece of paper, go to the supermarket, walk around the various food sections, choose what they need, load their trolley, go to the cashier, load the bags, pay, go back to their house unload their car and carry the food to their kitchen. The main difference lies in the energy and time that people used to spend in getting food to their home, and the more cognitive task that is now involved in programming what they need using both their past experience and the information that they have on the Web. To generalize what I already stated, life management is becoming more cognitive, i.e., it involves more (abstract) thinking than (concrete) doing. The result is that once someone has made a decision, a whole invisible energetic process is started, i.e., some other people will have to collect the food he or she ordered and bring it to his or her house. This person has delegated the physical tasks to people who he or she doesn’t know. He or she doesn’t know how many they are, and he or she doesn’t know what they do either. He or she has learned to trust the system. Since he or she is not close to the energetic process, actions are often irreversible, e.g., any erroneous choice of item that he or she made in his or her order will be taken as a definitive choice by the delivery person, there will not be any possibility of direct cancellation because the real product that he or she finally sees does not match the representation of the virtual object on the Web. Human beings have always tried to improve the control of their environment by extending their own capabilities. However, what do we loose by moving to the information-based interaction world? The Web removes the pleasure of going to the market place, and enjoying the delicate and multiple smells. Walking through the market stands, talking about the latest news with the merchant, weighing the food and smelling it; this is real sensory-motor experience! But, where is the pleasure in going to crowded supermarkets, waiting at busy counters, driving back home at rush hours? Automated shopping is much nicer in this case. It seems that information-based interaction makes a drastic distinction between things that are well formatted and things that require and deserve more human sensitivity and physical presence. Examples of the former things are packs of milk, sugar and mineral water. Examples of the latter are fresh fruit, fish and meat.

2.2. An interpretation of automation evolution

Bernard Ziegler, a former Vice-President of Airbus Industrie, made the following observations and requirements from his experience as a test pilot and distinguished engineer: • "the machine that we will be handling will become increasingly automated; • we must therefore learn to work as a team with automation; • a robot is not a leader, in the strategic sense of the term, but a remarkable operator; • humans will never be perfect operators, even if they indisputably have the capabilities to be leaders; • strategy is in the pilot’s domain, but not necessarily tactics; • the pilot must understand why the automaton does something, and the necessary details of how; • it must be possible for the pilot to immediately replace the automaton, but only if he has the

capability and can do better;



Page 6

• whenever humans take control, the robot must be eliminated; • the pilot must be able to trust automation; • acknowledge that it is not human nature to fly; it follows that a thinking process is required to

situate oneself, and in the end, as humiliating as it may be, the only way to insure safety is to use protective barriers."

Ziegler's very high level observations and requirements come from his very rich experience. Cognitive science could benefit from them by proposing appropriate theories of cognition that would rationalize this experience. Rasmussen’s model has been extensively used over the last decade to explain the behavior of a human operator controlling a complex dynamic system. This model is organized into three levels of behavior: skill, rule and knowledge (Figure 1).

Identification Decisionmaking

Planning

Situationrecognition

Situation(s)/ task(s)

Tasks(procedures)

Sensors Effectors

Goal(s)

Environment

Knowledge

Rules

Skills

Cognitive engineeringHuman sciencesSocial sciences

Operational researchOptimization

Expert systems

Electrical engineeringMechanical engineering

Control Theories

Figure 1. Rasmussenʼs model, the evolution of automation and the emergence of contributing disciplines.

Historically, automation of complex dynamic systems, aircraft in particular, led to the transfer of human operators’ skills (e.g., performing a tracking task) to the machine. Autopilots are in charge of simple tracking tasks since the 1930s. This kind of automation was made possible using concepts and tools from electrical engineering, mechanical engineering and control theories, such as mechanical regulators, proportional-integral-derivative controllers (PID), Laplace functions and stochastic filtering. Autopilots were deeply refined and rationalized during the 1960s and the 1970s. Human skill models were based on quasi-linear models’ functions and optimal control models. Human engineering specialists have developed quantitative models to describe and predict human control performance and workload at Rasmussen's skill level. They have been successfully applied to study a wide range of problems in the aviation domain such as handling qualities, display and control system design and analysis, and simulator design and use. The second automation revolution took place when the rule-based level was transferred to the machine. In aviation, a second layer was put on top of the autopilot to take care of navigation. The flight management system (FMS) was designed and implemented to provide set points for the autopilot. A



Page 7

database is now available onboard with a large variety of routes that cover most of the flights in a specific geographical sector. Pilots need to program the FMS by recalling routes from the database and eventually customize them for a specific flight. Once they have programmed the FMS, the aircraft is "capable of flying by itself" under certain conditions, i.e., the FMS is in charge of the navigation task in pre-programmed situations. Today, human factors research and practice have evolved towards cognitive engineering, and hermeneutics, because the control of highly automated systems does not require the same abilities and requirements as traditional tools. Human operators are mostly working at Rasmussen’s knowledge-based level. Basic operations are delegated to the machine, and humans progressively become managers of (networked) cognitive systems. Humans need to identify a situation when there is no pattern matching (situation recognition) at the rule-based level, to decide according to specified (or sometimes unspecified) goals, and to plan a series of tasks. These are typical strategic activities. Some people are good at strategic activities, others prefer to execute what they are told to do. In any case, the control of cognitive systems requires strategic training. Back to the food shopping example, using the Web has totally transferred the shopping task to Rasmussen's knowledge-based level. The selection of food items is made using virtual objects. The delivery is planned with respect to the customer’s schedule and the nature of the food. Technology has always shaped the way people interact with the world. Conversely, interacting with the world has direct impact on how technology evolves. Rationalization of experience feedback influences the development of theories that make new artifacts emerge. In a technology-driven society, this goes the other way around, i.e., the use of artifacts make new practices and new jobs emerge, as the film technology did for example. The twentieth century was rich in technology innovation and development. The speed of evolution of technology and resulting practices is very sensitive to economical impacts. In some cases, when economical benefits were not obvious a priori but the evolution of human kind was at stake, technological advances were decided at the political level such as designing and developing a technology that enables a man to walk on the moon. Today following these grandiose projects, we realize that human-centered automation, and more generally human-centered design, is not effectively taken into account at the political level yet. A new paradigm needs to be found to better understand the balance between human and machine cognition.

3. ARTIFACTS EMBED HUMAN COGNITION

People build and interact with artifacts by satisfying both physical and cognitive constraints. Different people may not see the same kind of physical and cognitive constraints in the same artifact.

3.1. Affordances

Creative people who make films use very different skills from the Lumière brothers who created the film technology. The Lumière brothers were engineers. Their films cannot be qualified of artistic productions. Several decades later, the film technology has become the support of what is now called the Seventh Art. When we go to a movie theater, we are not so impressed by the engineering side of a film, but by its content and the feelings that it created on us. The film technology that was born from analytical thinking has become an art that enables the expression of situations, feelings and thoughts. Current film technology affords creativity.



Page 8

Objects have affordances whether they are natural or artificial. Very simple artifacts such as door handles have different shapes. A horizontal flat door handles that is located in the middle of a door suggests pushing. A vertical cylindrical door handles that is located on one side a door suggests to pull. Human beings establish a relationship between door handles and the appropriate action to open doors, i.e., push or pull. Gibson defines this kind of relationship between a human and an artifact as affordances. This kind of relationship is not necessarily visible, known or desirable. It is thus important in some cases to identify affordances, e.g., in safety-critical systems. The important thing to remember is that the best designed operational documentation may not be useful in using artifacts that have counter-intuitive affordances. Affordances are properties of physical artifacts as well as properties of cognitive functions. Some cognitive functions are learned or artificially constructed and are called cognitive artifacts. Other cognitive functions are innate. Information technology also has affordances that need to be found in tools that enable people to: • generate information, i.e., tools that enable making information explicit to others; • maintain information awareness, i.e., tools that enable people to be aware that appropriate

information exists somewhere; • access information, i.e., tools that enable people to access appropriate information at the right time

in the right format; • understand information, i.e., tools that enable people to understand information chunks.

3.2. Expertise and artificial intelligence

Many models of human cognition are analogs to existing mechanical tools. The computer has been taken as one of the best analog since its inception. Alan Turing was one of the first theoretician to compare the computer to the brain. Current mainstream cognitive science is strongly based on the assumption that cognition is information processing. This approach of cognitive science is known as cognitivism, and was introduced during two meetings in Cambridge and Dortmouth in 1956 by famous scientists who launched the artificial intelligence field. Among these scientists were John McCarthy, Marvin Minsky, Noam Chomsky and Herbert Simon. The basic assumption of cognitivism and artificial intelligence is that intelligence can be reduced to computation of symbolic representations. The fact that most cognitive scientists share this assumption enables them to communicate and present consistent work. It is interesting to note that Dreyfus recognized that the early prediction of Simon and Newell (1958) "that within ten years most theories in psychology will take the form of computer programs", has been partially fulfilled. The mind as an information processor is the current model that presents cognition as a complex system that receives, stores, retrieves, transforms and transmits information. Hubert Dreyfus, in his famous critics of artificial intelligence, claimed that the expert system technology would never be able to reproduce human expertise. He argued that expertise is developed over time to produce skills that cannot be represented by IF-THEN production rules. These skills are not ordinary skills, they are skillful means as Varela, Thompson and Rosch put it. Dreyfus claimed that situational knowledge cannot be captured and thus cannot be modeled and simulated using artificial intelligence. Situational knowledge is quite different from analytical knowledge. This philosophical claim can be compared to the work of Schneider and Shiffrin (1977) who showed experimentally the distinction between automatic and controlled human information processing. Artificial intelligence systems and principles are based on analytical knowledge representations. Situational knowledge is incrementally learned from experience. Learning mechanisms and expertise are also too complex to be represented analytically according to Dreyfus. Consequently, some expert systems may crystallize



Page 9

human intelligence when they are used, but they are not unique in this regard. The airplane, for example, is not an expert system per se, but it crystallizes very clever thought on the flying process. People don’t fly naturally, but we understood what were the main parameters and relationships that characterize the flying process. New artifacts and skillful means were created that enable people to fly. Today, artificial intelligence has reached more maturity as a discipline in part due to the fact that computers are much faster and more powerful than we could not imagine twenty years ago. It is now possible to visualize abstract concepts due to the considerable improvement of user interface technology. On the one hand, computers are able to play chess better than people: Deep Blue is able to beat Kasparov! I believe that intelligence is not necessarily an intrinsic construct of software programs, but an emerging property that come from enhanced performance, quality of results, and appropriate interaction. When a system enables you to efficiently and purposefully achieve important tasks that you could only dream of performing before in extremely short notice, and are necessary to support your life, this system is likely to improve your life, and can be qualified for bringing artificially extended intelligence.

3.3. The cognitive function paradigm

The concept of cognitive function, that is presented in this chapter, was developed in the framework of human-centered automation in aeronautics systems, and more generally safety-critical systems, in the context of the traditional French school of ergonomics, that emphasizes activity analysis. This concept is very similar to Leont’ev’s functional organs. "Functional organs are functionally integrated, goal-oriented configurations of internal and external resources. External tools support and complement natural human abilities in building up a more efficient system that can lead to higher accomplishments. For example, scissors elevate the human hand to an effective cutting organ, eyeglasses improve human vision, and notebooks enhance memory. The external tools integrated into functional organs are experienced as a property of the individual, while the same things not integrated into the structure of a functional organ (for example, during the early phases of learning how to use the tool) are conceived of as belonging to the outer world." As Kaptelinin wrote. The concept of cognitive function helps analyze how people (human agents) and information-intensive systems may interact. A cognitive function can be interpreted in the mathematical sense or in the teleological sense. The former interpretation leads to the definition of an application transforming an input into an output. The input is usually a required task to be performed. The output is the result of the execution of the task. We usually say that a human uses a cognitive function that produces an activity or an effective task. The latter interpretation leads to the definition of three attributes of a cognitive function: • a role, e.g., the role of a postman is to deliver letters; • a context of validity, e.g., the context of validity of the above role is defined by a time period that is

the business hours and a specific working uniform, for example; • a set of resources, e.g., the resources necessary to execute the function are a bicycle, a big bag and a

delivery procedure, for example. A resource is a cognitive function itself. In a big town for example, the chief postman delegates the letter delivery task to other postmen.



Page 10

Note that when the postman returns home (different context) the above described cognitive function is no longer active, and he activates different cognitive functions such as father and husband (different roles) using different resources such as affection for his family and helping in homework activities. People develop appropriate cognitive functions to speed up, and increase both comfort and safety of their job, i.e., the tasks that they usually perform. These cognitive functions can be soft-coded or hard-coded. When they are soft-coded, they usually appear in the form of procedures or know-how stored in their long-term memory. When they are hard-coded, they usually appear in the form of interface devices or manuals that guide users in their job. In both cases, cognitive functions can be either implicit or explicit. When they are implicit, they belong to what is usually called individual expertise. When they are explicit, they belong to what is usually called sharable knowledge. Sometimes, cognitive functions remain implicit for a long time before becoming explicit and easily sharable. When a cognitive function is persistent, it can be elicited and transferred to a machine that will perform it for its user. This transfer process is commonly called automation. Various levels of automation can be implemented with respect to both human factors involved in the execution of the task and the technological limitations. Norman claims that the most important part of a successful design is the underlying conceptual model. The conceptual model of the use of an artifact and the artifact itself evolve incrementally. The effect of and the influence on the organizational environment are often not taken into account in conventional design. This was one of the reasons why I have proposed the concurrent investigation of Artifacts, Users, Tasks and Organizational environments (known as the AUTO pyramid), to develop more situated conceptual models. Conventional ways of doing design is based on goal-driven methods, i.e., designers start with an overall goal in mind and attempt to decompose this goal into sub-goals until basic actions can be derived and effectively performed. Goal-driven approaches to design are strongly anchored in industry since they lead to manageable and explainable products. Resulting products are usually technology-centered. In many cases, they are easy to maintain also. However, goal-driven design approaches ("I want to do this!") do not handle end-users requirements well. At the other end of the spectrum, event-driven approaches ("If this happens, what should I do?") to design tend to foster participatory design and use of experience feedback data. Since design is intrinsically iterative, event-driven approaches to design can be very time-consuming and very unstable. Boy introduced the Cognitive Function Analysis (CFA) approach that supports event-driven human-centered design. CFA can be seen as cognitive function allocation. CFA proposes to handle these issues in two necessary and complementary ways: categorization of experience feedback cases into cognitive functions that may be re-used in design; use of an integrated methodology based on the use of active design documents that enable design teams to implement participatory design.

3.4. Embedding cognition into tools: Looking for the right balance

No one questions the use of the clock today: the role of the clock cognitive function is to provide the time to its user. Its context of validity is determined by several parameters such as the working autonomy of the internal mechanism or the lifetime of the battery. Its resources include, for instance, a battery, the ability of its user to adjust time when necessary or to change the battery. Note that the user is also a resource for the clock cognitive function. Today, external cognitive functions have become more complex than the clock. People delegate to these external cognitive functions some actions that they used to before. They have to plan, monitor,



Page 11

negotiate, supervise, communicate, cooperate and coordinate with composite artifacts, e.g., a travel agent needs to work with a composite world-wide travel system (i.e., a composite artifact that embeds many external cognitive functions). This system is a network of a large number of computer systems. The travel agent learns about information traffic jams in the network, local crashes and tricks for booking a trip using a cheaper carrier, for example. These are cognitive functions that are relevant to his or her job. They are valid in time-specific contexts such as "during a holiday period", or "the Paris Orly airport is always very busy on Monday mornings". Travel systems have taken this into account for a long time with pricing. Cheap flights are usually available during the day, not in the morning or the evening. Today, such systems learn very fast, i.e., both human and external cognitive functions adapt very fast. The travel agent needs to assimilate and accommodate more cognitive functions to handle the increasing number of options. Tools have been designed by humans either to extend their capabilities or to create new capabilities such as flying. If computers are extensions of human brains, the World-Wide Web creates new capabilities that were not possible before such as consulting almost any kind of document located anywhere in the world after a few mouse clicks. It is now possible to involve a group of people distributed all over the world in a meeting. Such a virtual meeting induces new types of interaction that themselves influence the design of supporting technology. This mutual influence between the emergence of new types of interaction and the evolution of technology is a key factor that drives the evolution of our societies. Tools are either physical, e.g., hammers, forks, and bicycles, or conceptual, e.g., methods, theories and know-how. Tools are concrete models of intelligence, but they often require experienced people to be efficiently used. The user-friendliness concept introduced by human factors specialists needs to be used carefully. We should not believe that user-friendly tools would guarantee task efficiency and performance. No matter how sophisticated is a tool, the user often needs to be efficient to obtain satisfactory results. I was amazed to see how a 21 years old carpenter could design and produce a staircase that worked perfectly in my house in a few hours only. The physical tools he used were very basic. I deduced that his conceptual tools were extremely sophisticated, and he must have learned how to use the appropriate tool for the right job at the right time.

4. FROM INDIVIDUAL INTELLIGENT ASSISTANCE TO MULTI-AGENT COMMUNICATION

Cooperation among people is not a new concept. The novelty is that people now cooperate through technology. Technology was centered on single individuals for a long time. Today, the notion of socio-technical systems has emerged from the need to investigate and better understand the various relationships among people and technology at large.

4.1. Intelligent assistant systems

Some smart artifacts may not qualify for being artificial intelligence systems, but they implicitly include the use of appropriate human cognitive function resources that make intelligent the resulting user-artifact system. For example, alarm clocks are not artificial intelligence systems, but they are smart artifacts that provide intelligent assistance. Intelligent assistance can play different roles according to its level of automation (Figure 2):



Page 12

• a classical technical documentation that is either paper-based or computer-based; • a context-sensitive computer integrated documentation that is used as an advisor where the human

controls the machine directly; • an amplifier of user capacities that is located in between the human and the machine, and serves to

amplify human action or perception; Hollnagel defines this role as the embodiment relation where germane aspects of the task are highlighted while simultaneously others are reduced or excluded;

• an interpreter of machine capabilities that is also located in between the human and the machine, and serves to interpret machine input or output; Hollnagel defines this role as the hermeneutic relation where the user has moved from an experience through the machine to an experience of the machine.

Human

Human Human

Human

IAS

IAS

IAS

IAS

Machine

Machine Machine

Machine

Conventional technical documentation Context-sensistive help

Amplifier of human capabilities Interpreter of machine capabilities

Figure 2. Various types of intelligent assistant systems (IAS). There is a motto that I always remember when I am trying to solve a problem: "it is much better to ask someone who knows that hiring hundreds who search." It is often the case that someone in your environment knows what you are looking for. So why not ask this person for help. This is where cooperation takes place. Obviously, you need to know first that the person knows and can help you. How do you know this? Either you know the person very well, or you know that he or she belongs to a well-known stereotype. For instance, I would like to share this particular episode that I experienced when I was starting to use Microsoft Word. My goal was to display the current footnote. At that time, Word did not have all the current useful functions. I started to use the documentation provided by Microsoft. After a quick search in this documentation, I did not find any relevant information enabling me to display the footnote. During this information retrieval, a colleague of mine walked through the door of my office, and I suddenly understood that my problem would be solved in a minute. I knew that my colleague was mastering Word (this is an example of identification of a stereotype). I asked him: "How do you display the footnote, please?" He immediately replied: "Press simultaneously <Shift key>-<Option key>-<Command key>-<‘s’ key>, and the same to hide it!" It goes without question that finding this complex command was quite impossible without expert help. I recalled this episode many times and finally tried to analyze it. This episode influenced the design and development of the Computer Integrated Documentation system that will be briefly described later in the chapter. In this episode, it is obvious that the kind of information I was looking for was not in my head nor could I easily find it in the appropriate manual. Someone else knew it. That is to say that someone else



Page 13

provided me with this information. This is an example of distributed memory that is commonly used by people everyday.

4.2. From individual user assistance systems to distributed cognition

It is amazing how people tend to become more autonomous using new technology and, at the same time, more dependent. Even if this not new, information technology tends to reinforce this trend. When we use email for example, we feel empowered, able to communicate with the whole world. However, I have observed people in my institute totally lost after a single electrical-power failure because they were unable to interact with their virtual world. They had "important" electronic messages (emails) waiting and seemed unable to use other communication means or wait and do other things! This autonomy/dependence paradox is technology-induced. One result is that new theories of human cognition tend to emerge accordingly. In particular, human-centered automation models have been investigated during the last decade due to the information-intensive nature of recent accidents in safety-critical systems, e.g., in aviation, nuclear plants and the food industry. Key concepts induced by the use of information technology have emerged such as situation awareness, trust, cooperation, coordination, human resource management, experience feedback, and traceability. Again, these concepts were there in the past, but they have become essential for safety and efficiency reasons. Distributed cognition is an emerging theory supported by a growing number of authors. For example, Edwin Hutchins' well-known article on distributed cognition in an aircraft cockpit describes how speed limits are set by people and machines. He shows how the same speed information may take various representations when it is remembered by a human being or by an artifact, and how these representations are complementary and provide useful redundancy. Situational constraints provide a framework for the information flow in the cockpit among people and artifacts. Speed information can be transferred either through the visual channel or the auditory channel. In some situations, e.g., under time pressure, the same information may be sent through several channels. Cockpit artifacts are then memory extensions of the crew. The evolution of information technology and the influence of anthropology and ethnomethodology induced the development of socio-technical models describing distributed cognition among people and artifacts. The concept of an external memory emerged. Analog memory models will be proposed in the following of this chapter. This leads to the point that a well-designed human-machine system includes the necessary balance of cognitive functions that are correctly allocated to humans or machines. The main difficulty in cognitive function allocation is the non-monotonic nature of situations. Distributed cognition is situated and dynamic. In addition, individual human cognition is acquired and refined from interaction with the external world according to the activity theory. "Activity theory is a commonly accepted name for a line of theorizing and research initiated by the founders of the cultural-historical school of Russian psychology, L.S. Vygotsky, A.N. Leont’ev, and A.R. Luria, in the 1920s and 1930s." For Kaptelinin, human activity is mediated by surrounding artifacts. People learn from their environment. In other words, human and organizational cognitive functions are constructed or discovered from interaction with the environment. In turn, designers produce new artifacts that are intended to help people. This co-evolution of human and artifactual cognitive functions characterizes our socio-technical societies. The Web is introducing new forms of distributed cognition. It mediates interaction among people who could never have exchanged messages before, and provides information access with the invisible help of appropriate software engines. However, this new type of interaction requires references to and



Page 14

continuity of the world that people know. This is why metaphors are extensively used. People need to be in control of their lives, not lost in this emerging hyperspace. My observation is that they adapt very quickly to the Web. What do they learn? This is still an open question.

4.3. Evolution of human-machine communication models

In this chapter, an agent is an artifact or a person that/who acts and communicates with the other agents of a society (see Figure 5). An agent produces actions that produce effects. Agents are taken in the sense of Minsky's terminology. An agent is always associated to several cognitive functions. In this perspective, an organizational memory system is a cooperating "society" of artificial and human agents. With the evolution of information technology, artificial agents have become software agents. They are computer programs facilitating human-machine interaction, as well as human-human communication. Software agents and metaphors can be seen as remembering facilitators. An agent can be natural or artificial (artifactual). The former type includes people, therapeutic or atmospheric agents, for example. We try to better understand how they work, and model them in order to better deal with them. The latter type includes automated power plants, sophisticated vehicles, advanced computer networks or software agents, for example. We have built them, but it is time to better understand how they are used and influence our lives, model them in order to better control them. A major issue is that artificial agents cannot be studied in isolation from people who are in charge of them. The type of interaction among agents depends, in part, on the knowledge that each agent has of the others. An agent interacting with another agent, called a partner, can belong to two classes: (class 1) the agent does not know its partner; (class 2) the agent knows its partner. The second class can be decomposed into two sub-classes: (subclass 2a) the agent knows its partner indirectly (using shared data or a mediating space for instance), (subclass 2b) the agent knows its partner explicitly (using communication primitives clearly understood by the partner). Any of these classes may lead to conflict. Conflict arises from unshared high-level goals or competition for available resources. Thus, it is necessary to define a set of synchronization rules for avoiding problems of shared high-level goals or resource allocation between agents. Synchronization rules may be handled following three different models: • (A) supervision (class 1); • (B) mediation (subclass 2a); • (C) cooperation by mutual understanding (subclass 2b). In the supervision model, the agent is totally ignorant or vaguely aware of the cognitive functions of the other agents. Typically, synchronization rules have to be handled by a supervisor (Figure 3). The supervisor can be one of the partners or an external agent using an appropriate knowledge base, that may be an operation manual. Typical interaction between a user and a video recorder (VCR) may lead to conflicts for example. When a user attempts to program a VCR, he or she usually requires assistance. In this case, the programming cognitive function includes the delegation to a supervisor who is either someone who knows how to do it, or an association of the user and a user-guide manual.



Page 15

Human Agents Supervisor Machine

Figure 3. Supervision: agents need to have

a supervisor to assist or manage their activities. In the model of mediation, the agent knows that cognitive functions of its partner exist through the results of (at least) some of the partner’s actions. Both of them use an active shared database, called a mediator (Figure 4). Such a shared database can be an agent itself if it actively informs the various agents involved in the environment, or requests new information (self updating) from these agents. Agents use and update the state of this database. An example would be both agents noting all their actions on a blackboard to which the other agents refer before acting. Agents have to cooperate to manage the shared database. This is no longer a problem of resource allocation, but a problem of sharing data which each agent can use as it is entitled to. This paradigm is called a data-oriented system. Such a system has to control the consistency of the shared data. Cooperative relations between agents do not exclude competitive relations, i.e., shared data are generally supported by resources for which the corresponding agents may be competing. In this case, synchronization rules have to deal with resource allocation conflicts and corresponding data consistency checking. Direct manipulation interfaces of current personal computers are based on this model. Shared interaction spaces

Figure 4. Mediation: agents manage to communicate through an active database mediating interaction among agents. Each agent has an allocated interaction space that can be shared by other agents.



Page 16

The previous models do not allow for co-adaptation. In the model of cooperating by mutual understanding, agents interact directly with the others using a mental model of their organizational environment (Figure 5).

Figure 5. Cooperation by mutual understanding: agents incrementally construct a mental model

of their organizational environment. Today, only human agents are able to cooperate by mutual understanding. They share a common goal and a common language expressed by messages, e.g., experts in the same domain cooperating to solve a problem. Agents communicate by incrementally constructing and sharing a common context. Each agent always attempts to construct a meaningful representation of the other agents in order to anticipate their behavior and reactions. Cooperation by mutual understanding involves learning about the other agents. Difficulty in or absence of learning may result in switching to either requiring supervision or mediation. In our everyday life, when people try to communicate, they attempt to understand the cognitive model of the other. They incrementally learn the cognitive functions of the others in order to improve direct communication. This natural communication process fails when mutual understanding is no longer possible. A few software agents learn from users, and support cooperation by mutual understanding, in very specific contexts. However, outside of these contexts, the process of cooperation by mutual understanding is not guaranteed. These three models are applicable to represent a single agent as a society of agents, i.e., cognitive functions, or a multi-agent (human and machine) system. A society of agents works towards reaching a conscious state of mind. As Daniel Dennett put it: "Enjoying a certain renown is not limited to appearing on television, at such and such a time. It means, more broadly speaking, enjoying the power of being able to influence the course of things. Conscious states of mind are, in an analogous way,



Page 17

those that are able to dominate our brain. And I would gladly define the spirit not as a "theatre", but as an "arena" (an arena without spectators, or course) in which different sequences of events, competing with each other, fight for domination. That which we refer to as "memory", for example, would be nothing more than a series of events that, once taken place, have enjoyed a "renown" in our mind superior to that of all the others, or that have held a position of influence for a long time — from which our consciousness is made up."

5. HUMAN MEMORY MODELS AS ANALOGS FOR EXTERNAL MEMORY SYSTEMS

A variety of human memory models has been proposed in cognitive science that includes the fields of experimental cognitive psychology, neurosciences and artificial intelligence. These models may serve as analogs in the analysis, design and evaluation of external memory systems, i.e., systems that extend human memory capabilities, and a universe of distributed cognitive functions in general.

5.1. Explicit versus implicit memory

In cognitive psychology, a major distinction is made between explicit and implicit memory. Explicit memory is consciously stored in the hippocampus, and includes declarative knowledge composed of facts related to people, locations, objects, and numbers for instance. Implicit memory is unconsciously stored in the cerebellum, and includes procedural knowledge composed of perceptual and motor skills such as ability to play piano, write, or ride a bicycle. Access to information in the memory is a function of frequency and recency of access. The more you use implicit information the less you forget it. Information vanishes with time, i.e., when it is not used. When a piece of information is similar to a number of others, an interference process takes place and people do not remember. Every morning, I cannot remember my parking place since I am confused by previous places I used on previous days. Context of storage is often very different from the context of recall, and causes memory troubles. The way information is stored and retrieved very much influences memorization. This distinction between explicit and implicit memory can be extended to organizational memory systems. Since organizations incrementally build their own rules, they document these rules based on facts that they try to regulate. These rules are basic declarative knowledge characterizing the organization, i.e., they explicitly say what should be done. In addition, people within the organization have their own ways of doing things. Implicit practice takes place in everyday activities, and persists until people stick to the same organization pattern. Human resources personnel try to make some of this practice explicit, and transform it into explicit rules by adapting it to the requirements of the overall organization. It is often the case that organizational rules are developed within a context that is likely to change later when these rules are being used. Thus, organizational rules should be designed with their attached context of use.

5.2. Short-term versus long-term memory

In classical models of human information processing, the short-term memory (STM), that is seen as a set of buffers of a cognitive processor, is distinguished from the long-term memory (LTM) that constitutes (as a computer analog) the mass memory (hard disk metaphor) of humans. The short-term/long-term memory distinction adds several characteristics to the implicit/explicit memory distinction:



Page 18

• The LTM is constituted of a network of chunks accessible through the STM by indices. The probability of retrieving a chunk of knowledge in the LTM is strongly related to the number of associations to this chunk.

• The LTM contains both facts (declarative knowledge that is related to explicit memory) and formal procedures (procedural knowledge and know-how that is related to implicit memory).

• The STM can be considered as a subset of the LTM. The cognitive processor activates this subset. Several chunks can be organized into a larger-size chunk. For instance, the letters N, L, A, I, P, A, E, R constitute 8 chunks. These same letters, in a different order A, I, R, P, L, A, N, E constitute in general, for someone speaking English, a single chunk. Chunks can be related to other chunks. Thus, a chunk activated in the LTM may itself activate other chunks and so on (notion of associative memory). Activated chunks are added in the STM.

• As new chunks are incrementally activated, old chunks disappear from the STM (limited capacity). The capacity of the STM is about 3 (2.5~4) chunks. It can reach 7±2 chunks due to the combined use of the LTM.

• Chunks are not directly added to the LTM. The insertion of a new chunk from the STM into the LTM essentially depends on the possibility to construct associations between the new chunk and existing chunks. This activity necessitates the activation of chunks in the LTM that use a part of the STM capacity. Thus, storing new chunks involves a longer time than a regular access time.

This distinction between short-term and long-term memory can be extended to organizational memory systems. The organizational short-term memory takes place in the current practice of the organization, it is usually constituted of active relations that evolve according to current events happening in and requirements decided by the organization. The organizational STM is sensitive to both external and internal events, as well as corporate knowledge stored in the organizational LTM. The organizational LTM includes all facts and rules that were previously designed and refined by the organization. The Web, for example, is used via an interface software, such as Netscape or Microsoft Explorer, that enables its user to connect, access and visualize appropriate information. This is the STM of the Web's external memory. The external LTM includes all accessible sites worldwide.

5.3. Space-time windows memory systems

Historical facts and events are usually remembered using space-time windows. For example, we remember the period of the Second World War within several space-time windows according to the focus of interest. If we are interested in French issues, a possible space-time window could be the period of the resistance on the French territory. More specifically, we could be interested in the 1942-1945 period in the Vercors maquis. Within this space-time window, a network of agents enters into play with their actions and fact productions. Space-time windows specify contexts of validity of the described facts and events. The main difficulty here is to describe this highly dynamic context. It helps to figure out what the invariants of the context are. The notions of persistence and obsolescence are crucial to categorize and describe contextual conditions. If we consider a space window, the town is more persistent than the building; the building is more persistent than the room. If we consider a time window in aviation, a flight phase is more persistent than a flight sub-phase. Context can be organized into context islands with respect to its attributes and the persistence of these attributes. Context islands may be: • hierarchically dependent, they are then defined as mutually inclusive context patterns with respect to

their relevant attributes and the degree of persistence of these attributes;



Page 19

• independent, they are then defined as mutually exclusive context patterns with respect to their relevant attributes and the degree of persistence of these attributes;

• interdependent, they then share some context patterns or attributes. Any cognitive function, whether it is embedded in a physical tool, a concept or a piece of information, may become obsolete because the current context is totally different from the initial context in which it was constructed. Some cognitive functions are only available from human experience. Since they are highly situated and dynamic, they are usually too complex to describe in ordinary terms, and formal representations become rapidly limited. These cognitive functions may disappear when the people who own them disappear. Even if they are "formalizable", if they are not used for a long period of time, they are progressively forgotten, and are eventually replaced by other cognitive functions. But there are universal cognitive functions that remain valid across space-time windows. An organizational memory system should be seen as a complex dynamic process where context plays a crucial role, instead of a context-free database management system. It is composed of human and software agents exchanging information that can be concretely stored either on paper or electronically. It can also be stored in people’s minds. What remains in people’s minds and is not stored on concrete media is forgotten in the long term. The major issue is to better understand space-time windows that characterize vivid remembering. When people handle this vivid remembering, usually time windows are around forty to fifty years if there is no information transfer to other people. What will be the time window for the information stored on CD-ROM’s?

6. AN ORGANIZATIONAL MEMORY APPLICATION WITHIN THE IMAT PROJECT

IMAT (Integrating Manuals and Training) is a European ESPRIT project that includes a research effort on organizational memory of training knowledge. IMAT organizational memory issues are based on two basic questions: What happens when an expert leaves the organization? What happens when someone is hired into the organization? The rationale of IMAT is to better understand the way training materials (TMs) are managed and determine interaction or lack of interaction among various agents who deal with TMs. Each IMAT agent has his or her own cognitive function or role. An IMAT agent may be a provider (database administrator, information supplier, transformer, organizational memory manager), an instructional material developer, an instructor, and or a learner. The multi-agent world of IMAT is represented in Figure 6 that provides a model of the information flow of the organizational memory.



Page 20

IMAT documents

Experiencefeedback DB

Raw data

Instructor

OM manager

Transformer

Learner

DB administrator

Informationsupplier

Instructionalmaterial

developer

Figure 6. Model of IMAT organizational memory information flow. Better understanding of interactions among roles and training materials requires a cognitive function analysis of the TM management process. The main objectives of this analysis are: • elicit the functions that are necessary to handle the organizational memory; • find out whether they are human-based or computer-based; • elicit the functions that can be analyzed according to the organizational memory information flow

model being used. Cognitive function elicitation is implemented as a requirement engineering process where users' requirements (UR) are gathered, technological possibilities (TP) are defined, an ontology of the domain is then constructed, UR/TP matrices are developed to document design rationale of the organizational memory, the organizational memory is experienced, and evaluations are carried out using the appropriate organizational memory model. This process is performed several times until a satisfactory organizational memory model is found. The evaluation of an organizational memory model is based on: • usefulness criteria, e.g., information obsolescence control, version control, agent participation, costs,

organizational memory useful functionality; • usability criteria, e.g., learnability, knowledge retention, errors and recovery from errors, efficiency

in knowledge management, subjective satisfaction of the agents. An additional goal of IMAT is to support organizational learning by improving: • experience feedback, e.g., linking documentation topology to semantics; • categorization of training materials, e.g., re-using fragments and instructional patterns; • traceability of organizational decisions, e.g., tracing rationale, generalization and forgetting. The nature of the organizational memory model may be centralized, i.e., typically supervised by a librarian, or distributed, i.e., based on individual annotations and a set of coordination rules. In the centralized model, the organizational memory manager is the librarian. In the distributed model, the experience feedback database is the repository of individual annotation. It may be thought of as a



Page 21

blackboard where the various agents write and read information. Each agent learns and uses coordination rules. The IMAT overall human-centered design model is built according to a loop process for each domain of investigation (Figure 7). There are three domains in IMAT: aircraft maintenance, car repair, and road traffic control systems. The various organizational memories are developed in parallel in order to foster cross-evaluations, comparisons, and mutual benefits. They are different because: • the agents are in either the same location, two different locations (training and maintenance), or in a

distributed environment; • the levels of maturity of the models are different; • usefulness and usability of the various models are examined differently.

Elicit requirements

Improve the OM model

Implement the model

Test the model

Figure 7. Loop process for each IMAT domain. Our first investigations have shown that people are aware of emerging information technology, but they do not know how to integrate it. Routine practice is perceived as safer than unpredictable new practice using new technology. Computers are not perceived as dangerous per se, but they are not used if they do not provide the right result at the right time. The co-adaptation process may break sooner than expected when technology is not ready. Co-reliability, as described later in this chapter, is a key issue in the co-adaptation process. Experimentation is ongoing. An organizational memory model for each mentioned application domain is expected by the end of the year 2000.

7. CO-DEVELOPMENT OF HUMAN AND ARTIFICIAL COGNITIVE FUNCTIONS

As we have already seen, human roles have become more cognitive with the emerging use of complex automated artifacts. In order to avoid unnecessary confusions, I consider airplanes as complex automated artifacts themselves. The addition of computers onboard does not change the cognitive character of flying. Computers have removed some of the physical tasks from humans. In this sense, flying has become more cognitive, even if today some of past cognitive tasks are allocated to the machine. In our everyday life, the distance between the actual physical work and us tends to increase. We should not forget that there are still people acting physically in the background doing invisible work. The push button generation of people might become physically lazy, and lose the sense of (old) reality. New types of reality emerge. When you open the appropriate Web page and select: "I would like a pizza in



Page 22

15 minutes!" A mouse click and that is it, you eventually will have a pizza in fifteen minutes. The resulting interface (usually perceived as black box) between people and everyday life facts has become deeper. This black box includes more agents mediating people’s actions. These agents are either human or software. This type of deeper interfaces was created and developed to amplify people’s actions. This works well most of the time. However, there is an increasing need for interpreting what the mediating agents do when there is a failure, e.g., when the pizza is not delivered after 15 minutes. This is because people either have trouble understanding the behavior of the overall process, or simply don’t trust the process at all when the perceived level of risk is too high. Many people become concerned or frustrated by not trusting processes taking place within these deeper interfaces. People are today in a situation very similar to children discovering the world. People are discovering a fast moving information-intensive world.

7.1. From Piaget's action schemes to cognitive functions

Jean Piaget developed a constructivist theory of mind that claims a series of stages in early child development. According to Piaget’s theory, children represent the world only in sensory-motoric terms. As they interact with the world, they learn relationships between actions and sensations, i.e., effects of their actions on their body. Young children gradually understand the concept of physical object. Their subjective impressions of the world’s reactions to their actions transform into objective, more abstract, representations that Piaget called action schemes. Individuals construct their own action schemes with respect to their own experiences with the world. The schemes are not innately supplied but constructed. Scheme construction is supported by two mechanisms: assimilation and accommodation. First, a scheme is assimilated when an action is successfully linked to an effect sensed by the originator of the action. The context in which this assimilation is performed is not necessarily recorded. Many other tries will eventually reinforce this scheme, i.e., the relationship between the action and its effect. If this reinforcement is often performed in the same context, then a specific action scheme will be accommodated; otherwise this will lead to a more complex learning mechanism. Context associated to an action scheme is progressively extended based on the reliability of the scheme in other situations. I propose to use the cognitive engineering framework of a memory-based incremental learning that extends Piaget’s theory to knowledge management in organizations. In this framework, the world is modeled by a set of agents that construct and use their own cognitive functions. Since Rasmussen’s model falls short at describing cognitive activities in a distributed environment, the cognitive function analysis methodology was developed by Boy to bring together a systemic approach and a holistic approach to design human-machine systems. It takes into account a distributed-cognition view rather than a single-agent view. The distributed cognition paradigm states that knowledge processing is distributed among several agents that can be humans or machines. It provides a framework to explore how agents respond to their environment. In this sense, the cognitive function model is useful to investigate ways the same information is represented, and how representations are modified and exchanged among agents. These multiple views of cognitive functions are analyzed and elicited using the already-mentioned framework of the AUTO pyramid.

7.2. Co-reliability of humans and artificial agents

Co-reliability within a multi-agent system deals with co-operation among agents. Human error is a major concern in safety-critical systems today. According to Tom Sheridan, human errors are mainly



Page 23

due to a lack of feedback. When people do not know what the results of their actions on a system are they have problems to control this system. More generally, humans who do not obtain feedback from their acts have problems to control their lives. In the multi-agent modeling perspective, agents in charge of a job that do not timely obtain appropriate feedback from other appropriate agents have problems to accomplish this job. To the question: "who is in control?" The answer is anyone that has the appropriate role to be in charge in the right context. Safety barriers are very efficient in many cases, as human beings are unique to recover from unanticipated situations in other cases. Cognitive function analysis of multi-agent systems is very important to determine who should be in charge, when, why and how. Co-reliability within a multi-agent system deals with co-adaptation of agents. People tend to construct cognitive skills from training. In the multi-agent modeling perspective, cognitive functions of humans and artificial agents are constructed from interaction among agents. Errors occur and are sources of subsequent modifications of the various cognitive functions distributed among agents. Human adaptation is not the only possible mechanism, artificial agents can adapt very quickly also. In the past, artifact adaptation to human needs was a very slow process. Since today technology evolves very rapidly, human-centered technology-adaptation mechanisms should be put in place in order to better control the evolution. Experience feedback mechanisms and usability engineering are partial answers to machine adaptation. Co-reliability within a multi-agent system deals with co-dependency among agents. Safety-critical systems are commonly certified with extremely low probability of failure per hour of use. They are extremely safe from an engineering viewpoint. However, when humans use them, they make errors. More than 70% of aircraft incidents are due to so-called pilot errors. Some of these errors are induced by artifacts themselves, i.e., errors are observable at use time, but are really made at design time because designers did not take into account end-users characteristics and the many situations in which artifacts will be used. In other words, artifact usability needs to be better tested and understood. The key issue, that is emerging from current practice with modern technology, is then awareness of what other agents are doing and have done so far, why they are doing what they do, and what they will do next. The difficulty is that there might be a long chain of agents in space and time in this situation awareness problem. People cannot decide and act without a reasonable awareness of what is going on in their environment. They can work with failing machines when they know what failures are and why they occur. In some cases, they successfully and safely act without knowing much about a failure if they know that it is not serious. They need to have the right information, in the right format and at the right time. I claim that the search for and satisfaction of the co-reliability of human and artificial agents guide the emergence of distributed-cognition activities, i.e., cognitive function allocation among agents.

7.3. Emerging behavior of distributed cognitive activities

Virtual shopping has become a distributed cognitive activity, i.e., several actors are involved in it. Some actors are involved in the design of interface objects that are provided to customers on the Web. Their contribution is crucial. The more people understand what they order and are satisfied when delivered, the more the cognitive function "understand-what-you-buy" is satisfactory. Other actors collect the orders and execute the physical tasks of storing the frozen items in appropriate places according to the delivery time. Others actually deliver the food to your house. They need to be fast



Page 24

enough according to the nature of the food to be delivered. Finally you are a crucial actor, since you are the decision-maker and the evaluator of the overall process. You are the manager of this team of invisible actors. As I said earlier, you may actually ignore who some of the actors are. You only have indirect control over them by assessing the result of the delivery. In a sense, that is good since you are not involved in micro-management! Similarly to Piaget’s schemes, cognitive functions are constructed over time from experience. At the start of the learning process, a cognitive function is assimilated within a limited context of use. This context evolves when the owner of the corresponding cognitive function accommodates it to various situations. In knowledge management systems, cognitive functions are related to information generation, retrieval, understanding or interpretation, for example. As action schemes, they may evolve towards specialization or generalization based on experience in various contexts of use. One of the main applications of the cognitive function analysis of human-machine systems is human-centered automation, i.e., human-centered cognitive function allocation among humans and machines. Computerized knowledge management involves human-centered automation. Some information is generated automatically. User interfaces and new jobs are co-designed. Design in this context has become a real challenge because the resulting co-evolution is very fast. In safety-critical systems, it becomes crucial to incrementally document this co-adaptation or co-construction of human and machine cognitive functions to insure further traceability.

8. REHABILITATING THE ART OF MEMORY

Knowledge management has become a very important topic for many reasons such as rapid turnover in organizations, rapid evolution of technology, and change in nature and use of the produced artifacts. A very interesting paradox is that people tend to adapt very quickly to this evolution, but they have problems understanding it. Today, rationalization of technology evolution and of new practices emergence is much slower than these evolution and emergence themselves. This is why such rationalization is a crucial research topic today. In this section, I propose three viewpoints that might be useful in rationalizing our socio-technical evolution: the Art of Memory, the influence of Descartes’ heritage, and human experience.

8.1. The Art of Memory

The Art of Memory, invented by the Greeks, is not "officially" used today in our occidental societies. People have almost forgotten it after the invention and practice of printing. This art enables someone to memorize loci (locations) and images imprinted on his or her memory. It is usually considered as a mnemotechnique. A locus is easily remembered, e.g., a house, a balcony, an angle, etc. Images are forms, distinctive signs or symbols of things that we need to remember. The Art of Memory is like internal writing. Even if it is not necessary, people who know the letters of the alphabet are able to write and read. Similarly, people who know the mnemotechnique are able to put what they have heard into specific loci and repeat it by heart. If we want to remember many things, we need to have a number of loci. A major condition is that the loci must be organized into a series that needs to be remembered in order. This way, one can go forwards or backwards from any locus. Yates considers that the loci have attributes such as: put distinctive signs every five loci; create these loci in isolated places; create memory loci that are different from each other.



Page 25

The Art of Memory is a particular indexing mechanism that enables people to invent loci and images (indices) that help remember things. Emotional events tend to facilitate the formation of such loci and images. Images can be shocking and unusual, beautiful or ugly, funny or rude. Good stories create emotions that are likely to create useful indices that will facilitate remembering. "The reason that we remember the stories teachers tell is that human memory is set up to retrieve and tell stories, as well as to capture the stories that others tell. The story is a unit of memory. Furthermore, good stories contain good images, novel ideas, or particularly poignant passages that enable our memories to create indices that make retrieval of these stories easier. Storytelling depends on being reminded of a good story to tell. And, reminding depends on having labeled the stories we have heard or have created well enough so that when those labels appear naturally in the course of a day, we can use them to find relevant stories." As Schank and Jona put it.

8.2. Influence of Descartes’ heritage

Despite the very rich background of the Art of Memory, our current world is dominated by rationality that was first introduced by René Descartes in 1619. The Discourse on the Method (Discours de la Méthode pour bien conduire sa raison et chercher la vérité dans les sciences) developed by Descartes consists of: • accepting only what is so clear in one’s own mind as to exclude any doubt; • splitting large difficulties into smaller ones; • arguing from simple to the complex; and • checking, when one is done. Among other great contributions, Descartes’ heritage has led to: • the development of current information technology that itself turns out to rehabilitate a new Art of

Memory; • the development of phenomenology. Computers are the latest tools that have emerged from this mathematics-driven world introduced by Descartes. Today, computers are everywhere: at work, in administration, in amusement places, at home, etc. Computer games have become so popular that children have developed skills that many adults could never pretend to have such as rapid reactions, moving target tracking, and computer commands discovery by exploration. These are positive arguments in favor of computers. On the negative side, computers define a mathematics-based world where reality is made of severe approximations. The simulated world is discretized and often does not provide a complete account of the real world. People may accommodate to the simulated world and not to the real world. In the Cartesian approach, problem solving consists of mechanical problem decomposition. Knowledge is divided into chunks that can be learned individually. Quantitative assessments can be made. They are used to incrementally select students. This approach deals with more academic problems dealing with closed world situations (otherwise mathematics could not be applied). Consequently, the "cognition as information processing" paradigm has made cognitive science an axiomatic science looking for a set of phenomena. However, world phenomena that emerge today from the use of information technology are difficult to capture using this approach. As Varela et al. put it:



Page 26

"A growing number of researchers in areas of cognitive science have expressed dissatisfaction with varieties of cognitive realism. This dissatisfaction derives from a deeper source than the search for alternatives to symbol processing or even mixed society of mind theories: it is dissatisfaction with the very notion of a representational system. This notion obscures many essential dimensions of cognition not just in human experience but when we try to explain cognition scientifically. These dimensions include the understanding of perception and language, as well as the study of evolution and life itself." In the Art of Memory approach, it is difficult to assess students’ progress using mathematical criteria. This approach deals with real world situations and problems. Investigations are performed in an open world. New information technology provides a chance to combine these two different approaches. Software agents are likely to facilitate cognitive linking between humans and machines. On the one hand, they enable people to enhance their memory capabilities with the crucial condition that people master these external memory extensions. Otherwise, people take the risk of relying too much on software agents and forgetting more than remembering. On the other hand, they provide analytical means to enhance reasoning and remove heavy low-level interaction problems to the benefit of more interesting tasks.

8.3. Human experience

Varela, Thompson & Rosch link the dissatisfaction with a variety of cognitive realism to what they call "the Cartesian anxiety": "... We have the two extremes, the either-or of the Cartesian anxiety: There is the enchanting land of truth where everything is clear and ultimately grounded. But beyond that small island there is the wide and stormy ocean of darkness and confusion, the native home of illusion... This feeling of anxiety arises from the craving for an absolute ground. When this craving cannot be satisfied, the only possibility seems to be nihilism and anarchy." In other words, the Cartesian approach works when the world can be approximated into a "realistic" representation, out of this approximation (or discretisation) everything is so frightening that Cartesian people do not even want to know. The main issue here is that the world that we observe includes our own self. Citing Varela et al. again: "... The link between nihilism and Cartesian anxiety can be seen very clearly in the Society of Mind when Minsky comforts our inability to find a fully independent world. As he notes, the world is not an object, event, or process inside the world. Indeed the world is more like a background–setting of and field for all of our experience, but one that cannot be found apart from our structure, behavior, and cognition." Philosophers such as Merleau-Ponty, Husserl and Heidegger have developed phenomenology as the philosophy of human experience. Phenomenology provides a method for ontology elicitation based on human experience. Introspection is the basic mechanism of elicitation. The phenomenological approach departs from the grounded positivist approach of cognitivism. In the light of phenomenology, I would like to propose that the art of knowledge representation, supported by cognitivism, should be revised. I am not suggesting that the knowledge representation approach does not make sense any longer. It should be reformulated according to human experience. In particular, concepts such as interactivity, cooperation and emergence should be taken into account in knowledge representation especially since



Page 27

this article is initially targeted towards human-centered design of emerging life support systems, and knowledge management systems in particular. The Art of Memory is coming back to the front line because human experience is growing in our information-intensive world. The need for methods that link abstract concepts to experienced ones, objects to subjects, involves new approaches to human cognition. This is the case in situated actions, or distributed cognition, and activity theory.

9. CONCLUSION AND PERSPECTIVES

When I try to figure out the evolution of basic functions during the last millenniums, I observe very little change content-wise. We still need to eat, sleep, work for food, take care of our children and parents, or fight for our freedom. However, the nature of our artifact-based activities has drastically changed. These activities have become more remote from the physical world as the number of artifacts has increased. Delegation has become a key function for any human being to master. Human beings need to extend their abilities to process information and knowledge. If I had to give only one attribute of human intelligence, I would say adaptation. People are able to adapt to almost anything, e.g., cold and hot weather or unexpected situations. They have learned to adapt to evolving situations of life over the ages. They can make decisions using very imprecise, incomplete and uncertain information. Today, new information technology is a major challenge to human adaptation. Following the various multi-agent models that are described in this chapter, all agents should adapt to each other. In other words, co-adaptation, seen as joint evolution of artifacts and emergence of human practices, is a crucial research issue. We have seen that one of the main problems today is to cope with the rapid evolution of technology and the emergence of new practices. Observability and controllability of evolving socio-technical systems is a major enterprise that must be better understood. Socio-cognitive models currently developed and used are good candidates. In particular the work in distributed cognition is very promising to this end. Finally, knowledge management is leading to the investigation of new paradigms departing from the classical information processing assumption. If Descartes is responsible for the mathematization of our occidental world, new approaches are emerging reactivating old techniques and models from the ancient times such as the Art of Memory, human experience and currently active oriental philosophical practices.

9.1. Towards an epistemology of organizational cognition

Research in knowledge management has grown over the last decade to become a scientific field of investigation. Human and social sciences tremendously need to consolidate their investigation approaches, and articulate around the growing necessity of human-centered design. Multi-disciplinary teams are flourishing with the goal to humanize technology. In addition, the trend is to break with the traditional view of two separate camps: research and practice. Technology is developing much faster than research is able to rationalize. Today, research should take place more in the real world than in the laboratory. Disciplines such as anthropology that are developing in the information technology world are very promising. Knowledge management still needs more conceptual tools that support investigations. This chapter is based on two philosophical viewpoints: positivism that was originated by Auguste Comte, and developed by Bertrand Russel; and phenomenology that was originated by Henri Bergson, and developed by Maurice Merleau-Ponty. Positivism led to behaviorism since it is based on the fact that



Page 28

consciousness is not observable, and thus investigations of consciousness are close to metaphysics. Positivism is interested in anything that is observable. It leads to empiricism. Phenomenology is interested in introspection and human experience. The main question is: how to save positivism without loosing the meaning of knowledge? There is no answer to this question without a model or a theory deeply grounded into real-world experience. This is why theories of human cognition are very useful as mediating tools enabling the observation of knowledge management issues. They obviously need to be extended. Every model evolves when it is used. A model is a set of elements that partially reproduce another set of richer elements that are commonly called "reality". Several models may reproduce the same reality. These models may be more or less concrete or abstract. An airplane may be modeled by a physical mockup, a drawing, or a mathematical model. A good model is a useful model. In addition, a model enables its user to perform a simulation. A model is useful when operations performed on it make sense compared to operations performed in the equivalent real world. This requires that the elements that constitute the model and the relations among these elements be chosen such as their functions image isomorphic functions in the real world. The difficult problem is that cognitive functions are extremely difficult to identify in the real world because the real world is not separable into pieces, i.e., although this provides satisfactory results in limited, well-defined contexts, we cannot in general cut the world into pieces and link them again to re-build an equivalent world. One of the main issues is then to define the context of validity of the cognitive functions that are modeled. Any model should be built using explicit rules. Expressionist and cubist painters represent nature differently. Even if these rules are difficult to write, they exist and define a community of practice. When such rules persist then the community becomes more structured and solid. Today, the notion of context has emerged as a very important element in various theories of human cognition. In addition, it has put forward the notion of societies of interacting agents. Interaction that involves communication, co-operation, co-ordination, trust, delegation and other appropriate functions, is the main element of investigation. In knowledge management, interaction is multidimensional, in space and time, and across various types of agents (actors). Tuomi, who recently described the autopoiesis (self creation) model proposed by Maturana and Varela (1980), distinguished the organization and the structure. "Organization denotes those relations that must exist among the components of a system for it to be a member of a specific class. Structure, in contrast, denotes the components and relations that actually constitute a particular unity and make its organization real. Organization, therefore, defines a class of systems, whereas structure is a particular implementation of a system in the organizational class." Tuomi compares the homeostatic system from the cybernetic control theory with an autopoietic system. Tuomi argues that "[Wiener's] description of living systems... focuses on regulation mechanisms, whereas Maturana and Varela focus on production mechanisms." An homeostat does not reproduce itself as an autopoietic system does. In some cases, the homeostatic model will work elegantly. In other cases, the autopoietic model will be preferred. To summarize, models are very useful tools that are built to connect observable cues to meaning. In other words, they try to reconcile positivism and phenomenology. Tuomi proposes the 5-A model of knowledge generation that encapsulates these notions. It includes five main functions: anticipation, appropriation, articulation, accumulation and action (Figure 8).



Page 29

Articulate

“Learn” Act

AnticipateAccumulate

Appropriate

Figure 8. Tuomiʼs 5-A model of knowledge generation. Knowledge may be produced from a breakdown in human adaptation to the world. This is not possible without action. For example, we learn from differences between anticipated behavior of the environment and the actually perceived situation. "Knowledge can also be produced by appropriating knowledge that exists in the society." For example, we learn from others who are more knowledgeable than us, as well as from errors or accidents. Existing knowledge may be articulated and reconfigured in order to create new knowledge. Knowledge is accumulated in a memory store following an incremental process similar to Piaget's accommodation mechanism.

9.2. Towards a concurrent approach of human and organizational learning, and human-centered design

There is a need for studying the future of organized knowledge supported by new information technology. The integration of new information technology in the life support systems, such as human/organizational learning and human-centered design, should enhance its role of preserving cultural heritage, improving knowledge transfer, social integration, and safety. Human and organizational learning tries to adapt people to the world. Human-centered design tries to adapt technology to people. This co-adaptation process should contribute to: • the development of autonomy and subsequent mandatory co-ordination, as well as collective and

individual learning; • the removal of barriers caused by social or geographical isolation; • the openness to the external world and facilitation of synergy with local resources. The following perspective of co-adaptation of people, organizations and technology integrates my overall knowledge on the various theories of human cognition and the observation of the real world in knowledge management. I strongly believe that good individual and collective human knowledge management starts at school and in family circles. Learning is a life-long experience. Our children will really live in and develop the information-based society. I propose five attributes of learning that come from my own experience: • motivation; • learning how to learn; • efficiency; • allowing for errors in order to learn from them;



Page 30

• memory retention. Learning should be motivated. It seems that too rigid education systems do not encourage motivation to learning. We need to be motivated to learn. In our occidental societies, children are motivated to buy things and to watch television for instance. They are motivated to learn also. Adults too. But learning, reading, discovering scientific matters or discovering music should not be made unnecessarily difficult and boring. How can we keep people discover the pleasure of learning? Information technology brings new ways of motivating learners. These ways should be further investigated. Learning should be learning how to learn. Did we learn how to learn? Do our children learn how to learn? Occasionally. In many countries, it is implicitly assumed that learning is part of our intellectual capabilities, and learning how to learn seems to be not necessary. Teachers should try to encourage learning how to learn by providing appropriate artifacts that help students to learn. Most of these artifacts are based on experience. They are very rarely discussed and transferred. They deal with pragmatics. We say that this teacher is good because he/she knows how to present the right artifact at the right time according to students’ needs and reactions. Students need to learn how to state problems. "The essence of intelligence is to act appropriately when there is no simple pre-definition of the problem or the space of states in which to search for a solution." As Winograd and Flores put it. Learning should be efficient. Most people stop learning when they find that it is not efficient. They usually think that they are wasting their time because they do not see much improvement in the quantity and quality of the results. Usually, students stop learning because they are not motivated and become discouraged when confronted with unnecessarily complexified matters. Learning should be made simple enough to afford rapid understanding. Learning should be more situational. Again, students should be pleased with themselves after completing successfully an exercise for instance. Of course, there are complex things to learn. Many things are boring to learn, but are necessary to learn. These should be made more attractive using metaphorical artifacts that break down the complexity. Learning should allow making errors and learning from errors. Many students rarely make errors because they learn what they are told to learn and repeat it in the right way (like robots). They obtain diplomas and eventually become part of the national elite. The other students make errors and feel bad about themselves. I claim that errors are good for people. Experiencing errors is enriching and should be better investigated. There are errors that need to be made and recovery strategies that are good to practice. Life is not linear. People will experience problems all throughout their life. They should be armed to solve them. For this reason, errors are good for contextualizing learning. As a metaphor, standing is not a passive static error-prone status, but is an active dynamic process of balance-error corrections. Children learn how to stand and naturally maintain this capability when they become older. Learning should improve memory retention. What should or should not be remembered? How is this possible and effective? Today, information is extremely volatile. The quantity of information that we receive every day on television, for instance, is too much to be remembered in depth. In turn, television has learned to present quick and shallow information spots. People do not make efforts to remember everything because not everything is relevant. In contrast, there is important information that needs to be remembered. For example, many people studied a foreign language for several years at school, and, ten years later, they are unable to speak it. Memory needs to be reactivated in order to persist. Activation can be improved using appropriate methods such as the Art of Memory.



Page 31

My last suggestion is about the design of new information technology. We have developed what was possible to do so far, but generally not often what was necessary. Some well-intentioned people say that the goal of science and technology is "to make the impossible possible". Sometimes, some possibles should be made impossible, and the focus should be on issues that are necessary to be addressed with respect to ethics, culture, biology and environment for example. This is tremendously difficult for two reasons: • it is difficult to anticipate the consequences of the development of artifacts that will be "necessary"

in the future; • people need to express their creativity and thus create new necessities. Both of these difficulties can be overcome when human-centered design is taken seriously. Group work concurrently foster anticipation and creativity. Human-centered design is innovative, incremental and participatory. Theories of human cognition are available to help develop more rationally the design process and its solution. Like the Polynesians who used dynamic cognitive maps to navigate across the Pacific more than 1,000 years ago, could we use software agents to extend our short-term and long-term memories to handle our "navigation" in our modern world? A new Art of Memory may take a larger place within this framework in the future. SUMMARY Socio-technical systems of our post-industrial era embed their own internal cognitive mechanisms and behavior. New information technology has induced new practices and human roles. The resulting co-adaptation of people and technology will be analyzed in the light of various theories of human cognition. We will analyze various aspects of human cognition embedded into artifacts. Even if they do not use the same kinds of tools and practices, all civilizations need to manage the knowledge that they produce and use. These tools can be physical or conceptual. For a very long period of time, the Art of Memory was used to manage knowledge. Knowledge transfer was essentially based on oral transmission within small groups. Printing started to extend knowledge transfer to larger groups. Descartes created a method that revolutionized knowledge management reducing most problems to mathematical equations that are possible to solve by definition. The fact that Descartes' method worked successfully in the material world, tremendously influenced the twentieth century because it was almost totally technology-oriented. It is amazing to observe that the computer, the ultimate production of Descartes' method, suddenly rehabilitates the Art of Memory because the materialistic approach to the world is no longer sufficient. The Web recreates artificial villages (communities) where people can communicate almost exactly as their ancestors communicated in their small villages. We discuss a dual problem in cognitive science that opposes a classical scientific approach to an experiential one, and some of its potential impacts on life support systems such as human/organizational learning and human-centered design. ACKNOWLEDGMENTS This chapter was produced thanks to the support of the European ESPRIT Project number 29175 "Integrating Manuals and Training" (IMAT). Many thanks to René Amalberti, Jeffrey Bradshaw, Michel Bruot, Meriem Chater, Markus Durstewitz, Jean-Gabriel Ganascia, Rachel Israel, Nathalie



Page 32

Mathé, Jean Pinet, Alain Rappaport, Stéphane Sikorski, Cheikh Sow, and Helen Wilson for many useful and vivid exchanges on the topic. BIBLIOGRAPHY 1. Boy, G.A. (1998). Cognitive function analysis. Ablex, Stamford, CT. [This book presents a socio-

cognitive and technical approach to human-centered design of safety-critical systems] 2. Dreyfus, H. (1979). What computer can’t do. New York: Harper and Row. [This represents an

influencial critical analysis of artificial intelligence]. 3. Gibson, J.J. (1977). The theory of affordances. In Perceiving, Acting and Knowing, R.E. Shaw & J.

Bradshaw (Eds.). Lawrence Erlbaum Associates, Hillsdale, NJ. [This presents an introductory contribution to the concept of affordances].

4. Newell, A., & Simon, H., (1972). Human Problem Solving. Prentice-Hall, Englewood Cliffs, N.J. [This represents one of the most influencial early contributions to cognitivist approaches and artificial intelligence].

5. Norman, D.A. (1986). Cognitive Engineering. In User-Centered System Design., D. Norman & S. Draper (Eds.), Hillsdale, NJ: Lawrence Erlbaum Associate, pp. 31-61. [This is an introductory text on cognitive engineering].

6. Rasmussen, J. (1986). Information Processing and Human-Machine Interaction - An Approach to Cognitive Engineering. North Holland Series in System Science and Engineering, A.P. Sage, Ed. [This presents a three-level model of the behavior of a human operator].

7. Tuomi, I. (1999). Corporate knowledge: Theory and practice in intelligent organizations. Helsinki: Metaxis. [This presents plausible approaches to the representation of organizational memory systems].

8. Varela, F.J., Thompson, E. & Rosch, E. (1999). The embodied mind. Seventh Printing. Cambridge, MA: The MIT Press. [This presents a critical overview of phenomenological approaches to human experience].

9. Winograd, T. & Flores, F. (1986). Understanding computers and cognition-A new foundation for design. Addison-Wesley Pub. Comp., Inc. Reading, MA. [This presents introductory valuable views on human-centered design of cognitive systems].

10.Yates, F.A. (1966). The Art of Memory. French translation by Daniel Arasse, 1975, Editions Gallimard, Paris, France. [This presents the historical and methodological foundations of what could be an history of imagination in the occidental world: the tradition of the art of memory].

Documents

THEORIES OF HUMAN COGNITION To Better Understand the Co …my.fit.edu › ~gboy › GAB › Book_chapters_files › UNESCO.Encycl... · 2016-02-26 · Boy, G.A., [2002], Theories