A system-level model of spatial working memory is described, using the au- thor's computer science and logical modeling approach. The mental image which is remembered is located in a lateral parietal area, and is part of a larger distributed rep- resentation including object identities and object appearances. The system is driven by plans which are stored in a separate module corresponding to ventral prefrontal and then copied to a planning module corresponding to dorsal prefrontal, where they are executed and sequenced. The system is integrated by an event memory module corresponding to the hippocampus, which gathers event information from most of the modules and con- structs a representation of the current mental event and the current mental episode. This information is used by the mental imagery modules to construct a unied representation of the current scene. Mental images are maintained by explicit messages from a currently executed plan. Upon a cue to use a saved mental image for motor output or eye move- ment, a message from the plan causes the image to be temporarily reinstated allowing it to be used for motor control. The system generates predicted MRI image sequences. We developed an energy estimation function based on the dierent kinds of information processing being carried out and we graphed energy usage in the main modules over time during a spatial working memory experiment. We also made a study of lesioning the model to produce spatial working memory decits. We

We describe a declarative language, called BAD (brain architecture description language), which we have developed for describing and then running brain models. Models are at the system-level of description, so that modules correspond to brain areas. Each module has a process and the set of modules runs in parallel and communicates via channels corresponding to observed brain connectivity. Processes are described using a parallel set of left-to-right first-order logical rules in clause form, but with additional activity in a rule body described by Prolog code. Data items are represented by logical literals. Both data and rules use certainty values. The overall system described by the user consists of more than one agent each controlled by a brain model, and behaving in a 3D virtual environment, which is described by logical literals. Interaction with this environment is described by Prolog code representing sensors and actuators. Brain models have been developed for social interaction, problem-solving, and episodic memory, routine memory and spatial working memory.

We developed a distributed modular architecture based on distribution of processing and storage according to data type, inspired by an analysis of the primate cortex. Data items and rules have sets of weights, and these are updated, attenuated and combined in various ways to allow data storage management and rule competition. There are many motivations for using weights, including modeling strengths of different primate behaviors, storage management so that data items are stored and removed from store as needed, rule selection by competition based on computed weight, rule selection stability by confirmation feedback among modules, time smearing to prevent propagation delay problems within the distributed architecture, and the desire to eventually find corresponding neural models. 1 Motivation from neuroscience We used a modeling approach of a modular distributed computational architecture and an abstract logical description of data and control [1] [2] [5], for which we have also analyzed the correspondence to the primate neocortex [6]. We analyzed the neuroanatomy of the cortex, which is divided into neural areas with genetically determined connections among areas [6]. We reviewed experiments which indicated the functions that each neural area appeared to be involved in, and we construed these results as each neural area’s action being to construct data items of data types characteristic of that area. We also defined neural regions made up of small numbers of areas, as there are about 50 cortical areas whose functions are sometimes not clearly differentiated. The end result of our analysis is summarized in Figure 1. In order to design a model of the cortex, we abstracted from our review some biological information-processing principles: 1. Each neural area stores and processes data items of given types characteristic of that neural area; data items are of bounded size. 2. To form systems, neural areas are connected in a fixed network with dedicated point-to-point channels. 3. The set of neural areas is organized as a perception-action hierarchy.

A System-level Brain Model of Spatial working Memory Alan H. Bond (alan.bond@exso.com) National Institute of Standards and Technology, MS 8263, Bldg. 220, Gaithersburg, Maryland 20899, and Institute for Neuroscience, UCLA, Los Angeles, California 90095 Abstract scription of data and control [Bond, 1996] [Bond, 1999] [Bond, 2004a], for which we have also analyzed its cor- respondence to the cortex [Bond, 2004b]. Modeling spatial working memory involved develop- ing mechanisms for a frontal area containing a mainte- nance process, a posterior area containing mental im- ages, and an integration mechanism involving a simple model of episodic memory, corresponding to the hip- pocampal complex. We implemented the model as a computer system and studied normal behavior and then abnormal behavior by introducing different types of component deficit. We specifically obtained a match to the work of Conklin et al [Conklin et al., 2005] on schizophrenic and schizoptypal subjects. For short delays such as 0.5 seconds they per- formed normally, but for long delays such as 7 seconds they exhibited a clear deficit. A system-level model of spatial working memory is de- scribed, using the author’s computer science and logical modeling approach. The mental image which is remem- bered is located in a lateral parietal area, and is part of a larger distributed representation including object iden- tities and object appearances. The system is driven by plans which are stored in a separate module correspond- ing to ventral prefrontal and then copied to a planning module corresponding to dorsal prefrontal, where they are executed and sequenced. Mental images are main- tained by explicit messages from a currently executed plan. Upon a cue to use a saved mental image for mo- tor output or eye movement, a message from the plan causes the image to be temporarily reinstated allowing it to be used for motor control. We describe a gradual approach to deficits in which a goal attenuates, causing the plan to fall below a threshold. This approach al- lowed us to match the experimental results of Conklin et al. in which schizophrenic and schizoptypal subjects could carry out a spatial working memory task with a 0.5 second delay but not when there was a 7 second delay. Spatial working memory Introduction We can perhaps define spatial working memory by de- scribing the basic experiment. Different experiments in spatial working memory have been reviewed by Cur- tis and D’Esposito [Curtis and D’Esposito, 2003] and by Rowe et al [Rowe et al., 2002]. The subject first fixates a central fixation point, then an additional image of a small object appears at a certain spatial location in the periphery, then there is a delay with just the fixation point visible, and then a cue is given and the subject moves either their eye gaze or their hand to the spatial location where they think the small object was. Thus, the idea is that the subject has to remember a certain spatial location for a short time of the order of a few seconds. In our research, we used a basic experimental de- sign from a recent standard paper by Conklin et al [Conklin et al., 2005]. This has four steps. 1. fixate - there is a cross at the origin and the subject has to fixate it, duration 2000 milliseconds. 2. note image - there is now also an asterix at a pe- ripheral location, duration 200 milliseconds. 3. delay, maintain image - back to just the cross, du- ration either 500 milliseconds or 7000 milliseconds. 4. cue, move hand to where the asterix was, duration 5000 milliseconds. In order to more precisely characterize psychiatric disor- ders such as schizophrenia, bipolar disorder, autism and ADHD, in recent years there has arisen the concept of cognitive endophenotype, which is a particular cognitive ability that may have deficits linked to certain genes. We need to understand these cognitive abilities and their re- lation to the brain, and we need to understand which underlying processes are shared in common among cog- nitive functions. Glahn et al [Glahn et al., 2003] correlated several dif- ferent components of spatial working memory such as encoding, maintenance, manipulation, time-tagging of visual spatial information, storage capacity and com- plex motor response, against genetic predisposition to schizophrenia. They concluded that encoding and stor- age aspects of spatial working memory may be effective endophenotypic markers for schizophrenia. In this study, we attempted to analyze and under- stand the underlying mechanisms involved in one cog- nitive ability, namely spatial working memory. This has been shown to be clearly impaired in schizophrenia. In order to understand it, we developed a system-level model of spatial working memory. We used a modeling approach of a modular distributed computational architecture and an abstract logical de- Our modeling approach Our general brain modeling approach. In the

We argue that frontal areas select and control routine action which originates in the basal ganglia, providing detailed interleaving with the creative knowledge elaboration activity of the cortex.The basal ganglia form loops and learn associations from various source areas to target areas of the cortex. We relate these connections to the cortical perception-action hierarchy.We outline examples of routinization and interleaving for the Tower of Hanoi problem, for routinization of motor control, of problem solving action, and of eye movement.We suggest a role for the ventral group of thalamic nuclei in controlling the flow of routine action. (c) 2006 Elsevier B.V. All rights reserved.

We discuss the problem of finding neuroscientific and psychologically plausible representations of the memories of events and episodes. In order to do this we need to take into account the neuroanatomical connectivity between the cortex and the hippocampal complex, and also the cognitive psychology of episodic memory. We then need to develop a model of the cortex and hippocampal complex and to find representations of events that are consistent with biological information-processing constraints. We conclude that events and episodes can be represented by certain codes which are stored in associative memories.

We describe how aircraft are designed in a large organization. We discuss the interactions among the different specialist departments during the design process, and the models used by each department. We observe that the main design choices are refinement operations on the design. The overall structure of the organized design process is one of coordinated refinement of models. We describe how the design negotiation process is controlled by an organizationally agreed sequence of commitment steps. We briefly discuss design elaboration, for wing design, within one commitment step.

Working at the systems level of analysis, we will use the term functional architecture to concern what processing components exist, how they are interconnected, and what information-processing functions each is involved in. In this paper, experimental evidence for the primate neocortex is analysed for conclusions concerning the existence of neural areas, for corticocortical connectivity among neural areas, and for the involvement of each cortical neural area in the functioning of the brain. We characterize the information-processing function for each neural area in terms of the types of information it is associated with, and conceive of its activity as processing, storage and transmission of data of the corresponding types for that area. We also adapt concepts of goal, plan, sequence, event and context for the description of information processing in the neocortex. This analysis shows that the primate neocortex consists in the main of a perception hierarchy, an action hierarchy and connections between them. In other words, from an information-processing point of view, the primate neocortex has a hierarchical perception–action architecture.

Experimental evidence has shown that the primate neocortex consists in the main of a set of cortical regions which form a perception hierarchy, an action hierarchy and connections between them. By using a computer science analysis, we develop a computational architecture for the brain in which each cortical region is represented by a computational module with processing and storage abilities. Modules are interconnected according to the connectivity of the corresponding cortical regions. We develop computational principles for designing such a hierarchical and parallel computing system. We demonstrate this approach by proposing a causal functioning model of the brain. We report on results obtained with an implementation of this model. We conclude with a brief discussion of some consequences and predictions of our work.

We describe a psycholinguistically and neurolinguistically plausible model of natural-language processing by the human brain. This model is based on the work of Gerard Kempen and coworkers at Leiden and Nijmegen who have developed computational models of language generation and of language recognition. We show how to use our own brain modeling approach to develop a neurolinguistically plausible model based on the Kempen psycholinguistic model. Our model is implemented as a set of inter-communicating brain modules that run in parallel. These brain modules have the same structure and control regime as other nonlinguistic brain modules. They approximately correspond to Broca's and temporal lobe areas.

We discuss a computer science and artificial intelligence approach to modeling the brain, and how it can illuminate the study of consciousness. After a brief review of the study of consciousness, we align ourselves with the research strategy of trying to understand individual properties of consciousness. We propose a computer science approach to developing a brain model, which we expect to illuminate control properties of mental dynamics, that is describing the brain by a set of processes with certain control relations. We then give a treatment of our model which consists of a set of processing modules corresponding to brain areas, and which are interconnected according to the connec-tivity of the corresponding brain areas. The action of each module is given by a set of predicate logic rules which are executed bottom up to quiescence. This gives an abstract description of the computation involved which can then be used to specify an implementation as a neural net. We explain how our model is implemented on a computer and how we demonstrated social interactions among model primates, then problem solving of the Tower of Hanoi problem and recognitionn of natural language sentences. We discuss the dynamics of our model and point out that it has some of the properties of consciousness. It is distributed but maintains a coherent state by the exchange of data. It constantly seeks viable states in which prioritized goals are pursued consistently with the currently perceived environment, and with any social interactions involved. We finally discuss the concept of cortical motivation. The action of our model leads to coherence and integrity of the distributed state of the cortex. 1 Background 1.1 Attitudes to consciousness There are many different attitudes to thinking about consciousness, from consciousness as too mysterious to comprehend [Flanagan, 1992], to consciousness as a syntactic artifact [Dennett, 1991], to consciousness as concerned with an integrated state According to WHM [WHM, 2002], there is some implicit agreement about three major levels of consciousness, namely, Type I-basic awareness and awakeness, Type II-primary consciousness, the ability to detect, represent and adapt to shifting states of the world, and Type III-consciousness of self, in which the self is detected and observed as an object in its own right. Wilber has a more complex classification [Wilber et al., 1986]. 1.2 Neural explanations of consciousness When it comes to neural explanations of consciousness, reviewed by [Rees et al., 2002] [Farah, 2001] …

We describe an approach to social action and social relationship among socially intelligent agents [4], based on mutual planning and mutual control of action. We describe social behaviors, and the creation and maintenance of social relationships, obtained with an implementation of a biologically inspired parallel and modular agent architecture. We define voluntary action and social situatedness, and we discuss how mutual planning and mutual control of action emerge from this architecture.

We describe a psycholinguistically and neurolinguistically plausible model of natural-language processing by the human brain. This model is based on the work of Gerard Kempen and coworkers at Leiden and Nijmegen who have developed computational models of language generation and of language recognition. We show how to use our own brain modeling approach to develop a neurolinguistically plausible model based on the Kempen psycholinguistic model. Our model is implemented as a set of intercommunicating brain modules that run in parallel. These brain modules have the same structure and control regime as other nonlinguistic brain modules. They approximately correspond to Broca's and temporal lobe areas.

A system model of the primate neocortex is presented, based mainly on the neuroanatomy of the rhesus macaque monkey and consisting of a set of processing modules arranged as a perception-action hierarchy. These modules correspond to regions of the neocortex and their connectivity to that of the neocortex. A computational approach based on predicate logic is explained, and the results of a computer implementation of the model are reported, which demonstrate social behaviors involving affiliation and social conflict. The behavioral states of primates involved in these behaviors can be represented by the states of the system model, which have a logical representation and a diagrammatic form. It is shown how the behavioral states in goal-directed behaviors can be represented and also their short term moment-to-moment development in time. It is then shown how the state of social interaction among two or more primates can be represented, using their individual behavioral states, with interindividual action and perception. The causal dynamics of behavioral states is explained and also a control mechanism, namely, the use of confirmation signals, which stabilizes behavioral states and their dynamics. Stabilized behavioral states are seen as corresponding to coherent activations of the system, resulting from successful selection of module activations and intermodule communication with confirmation. Am. J. Primatol. 49:315–338, 1999. © 1999 Wiley-Liss, Inc.

Experimental evidence has shown that the primate neocortex consists in the main of a set of cortical regions which form a perception hierarchy, an action hierarchy and connections between them. We develop a computational architecture for the brain in which each cortical region is represented by a computational module with processing and storage abilities. Modules are interconnected according to the connectivity of the corresponding cortical regions. We explain our computational principles for designing such a hierarchical and parallel computing system, and report on results obtained with an implementation of this causal functioning model of the brain. (C) 1999 Elsevier Science B.V. All rights reserved.

We propose that a control system will address the causal dynamics of the neural network that Depue & Collins regard as underlying extraversion. We briefly describe a control system approach and articulate the notion of integration. The integration of goals and regards is achieved by subcortical assessment of reward in the nucleus accumbens and VTA (ventral tegmental area) transmission of this information largely by dopaminergic systems and representation of reward in the MOC (medial orbital cortex). Thus reward information is collected, integrated, and evaluated in the MOC. Such control decisions rely on constraining processes, a functional property of the MOC mediated largely by serotonergic neurons.

This article describes a new class of information-processing models for social agents. They are derived from primate brain architecture, the processing in brain regions, the interactions among brain regions, and the social behavior of primates. 1. Primate brain and behavior In another paper (Bond, 1996), we have reviewed the neuroanatomical connections and functional involvements of cortical regions. We reviewed the evidence for a hierarchical architecture in the primate brain. By examining neuroanatomical evidence for connections among neural areas, we were able to establish anatomical regions and connections. We then examined evidence for specific functional involvements of the different neural areas and found some support for hierarchical functioning, not only for the perception hierarchies but also for the planning and action hierarchy in the frontal lobes. We concluded that the overall structure, both anatomical and functional, of the primate brain is that of a set of cortical regions with spatial localization and clustered interconnectivity. Each region is specialized in data processing and data storage to a limited number of data types. Regions process data in about 20 milliseconds, and communication channels comprise about 1 million fibres each carrying a firing rate, perhaps usable as 8 bits of information. Cortical regions are connected as a perception hierarchy, a planning and action hierarchy, and connections between corresponding levels of these two hierarchies. The perception hierarchy is based on data representing increasingly general situations. The planning and action hierarchy is based on increasingly general situations, plans and control. Hierarchical organization as an organizing principle for brain architecture has been pointed out by Albus (1981). Hierarchical control architectures have been discussed by Mesarovic et. al. (1970). From a study of primate behavior, as reported for example by Cheney and Seyfarth (1990). we concluded that affiliative behavior is fundamental, and that joint plans are co-constructed (Goodwin 1981) by primates interacting socially. Humans have similar behaviors (Harcourt and De Waal 1992). It seems that most of monkey behavior can be categorized under a small number of headings, and that almost all this behavior is socially interactive. That is, behavior of any individual monkey is highly coordinated with and motivated by that of other individuals, and involves detailed interaction with others. The following list covers most types of primate behaviors: (i) grooming, (ii) mating, (iii) infant caretaking, (iv) dominance struggles, (v) foraging and feeding, and (vi) territorial definition and control. Most of these are not solitary actions but involve detailed interaction with others. Even apparent solitary action is in any case strongly and continuously conditioned by the presence of others. We should naturally view these behaviors as joint behaviors carried out by two or more monkeys. Figure 1 shows a vervet monkey. Figure 1. A vervet monkey 2. Our computational architecture To obtain a computational architecture for a social agent based on the primate brain, we take the functional hierarchy composed of neural regions, and interpret each region as a computational module. Each module will have processing and storage abilities. We take connections between neural regions to be communication channels that data may be transmitted along. Figure 2 diagrams our social agent model. The top of the hierarchy is towards the front (left in the diagram), with perception below and action above.

An approach to CAD and CAM modeling and to the design of CAD/CAM systems is presented. Models of the product and of the process are represented by logical assertions in a common logical language. CAD/CAM functions are represented by the application of logical inference rules, which correspond to the derivation of new information as well as to actions. This allows all the different kinds of model and specification used in design and manufacturing to be represented in a computer in a common form. It therefore allows the representation of constraints and rules connecting any aspects of design and manufacturing together. This approach has all the advantages of formal specifiction, namely, ease of expression, communication, standardization and abstraction. At the same time, we demonstrate its practical implementation in an efficient form, and which is industry compatible, and we report practical experience with using this approach for CAD/CAM models and for intelligent CAD/CAM functions.

In a companion paper, we introduced an information model called EDM, for representing design and engineering information. EDM defines a small set of structures capable of depicting a wide range of semantics necessary for engineering design. These structures allow the definition of specific product models that are equivalent to database schemas, a fully instantiated structure is equivalent to an engineering or CAD database. EDM was developed in response to several criteria, among them the need to support changing technologies and evaluations and the need to support integrity checking. In this paper, EDM is applied to a small but complex example, a wall in building construction. Geometric, acoustic, and thermal properties are developed for the wall, defined in EDM structures. The example is then considered in terms of the evaluation criteria.