Sloman, A. 1998. Architectural requirements for humanlike agents both natural and artificial. In: Human cognition and social agent technology. v.19, p. 163-195. John Benjamin's Publishing Company.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....states as affective states , as I do not want to get involved in a discussion as to whether robots can experience emotions. The states I will describe do, however, come close to what is sometimes called primary emotions (such as startled, terrified, sexually stimulated, etc. see, for example, [16]) In the following I will first spell out my claim in more detail and motivate the experimental setup, i.e. the computer simulation used to argue it. Then I shall describe the design of the simulation, in particular, that of the involved agents in detail, concluding with a summary of the ....

Sloman, A.: Architectural Requirements for Human-like Agents Both Natural and Artificial. (1999) (to appear in Human Cognition and Social Agent Technology, Ed. Kerstin Dautenhahn)

....done in the area of user modelling with research on affective reasoning. In particular, I will enhance a user modelling component with affective information on stereotypes. This knowledge will be used by an affective reasoner component, which includes a reactive primary input appraisal component [6] and a hierarchical presentation planning mechanism [1] 2] Especially this knowledge will be used to control the selection of dialogue and presentation strategies. The basic flow of information within this system starts with the analysis of the user s verbal input by the speech component. The ....

Sloman, A. (1999). Architectural Requirements for Human-like Agents Both Natural and Artificial. (What sorts of machines can love?). To appear in K. Dautenhahn (Ed.) Human Cognition And Social Agent Technology, John Benjamins Publishing.

....architecture developed at DFKI [2] with a number of new features most notably enhanced input and output modalities for affective communication, and an Affective Reasoning Engine for affective state recognition and generation. In line with recent research in affective computing [4] 16] and [22], we use two affective information processing channels (see Fig. 2) Primary emotions (i.e. being startled, frozen with terror, or sexually stimulated) are generated using simple reactive heuristics, whereas Secondary emotions are generated by the deliberative Affective Reasoning Engine according ....

Sloman, A. (1999). Architectural Requirements for Human-like Agents Both Natural and Artificial. (What sorts of machines can love?). To appear in K. Dautenhahn (Ed.) Human Cognition And Social Agent Technology, John Benjamins Publishing.

....[21] we use two affective information processing channels (see Fig. 6) Our two affective channels are used to model primary and secondary emotional states a cognitive classification scheme based on the information level processes involved in the generation of the emotion types (see [4] and [27]) Primary emotions (i.e. being startled, frozen with terror, or sexually stimulated) are generated by innate neural machinery in the reactive cognitive layer centred around the human limbic system. Secondary emotions can either arise: a) through learned associations between categories of ....

.... associations between categories of objects and situations attended to by deliberative thought processes on the one hand, and primary emotions, on the other [4] or (b) in response to the deliberative planning process itself (when relevant risks are noticed, progress assessed, and success detected) [27]. Secondary emotions therefore require a deliberative cognitive layer in addition to the reactive machinery of primary emotions. As emotion types (happiness, sadness, envy, etc. often exhibit very different characteristics (i.e. varying degrees of cognitive richness and or expressiveness) ....

[Article contains additional citation context not shown here]

Sloman, A. (1999). Architectural Requirements for Human-like Agents Both Natural and Artificial. (What sorts of machines can love?). To appear in K. Dautenhahn (Ed.) Human Cognition And Social Agent Technology, John Benjamins Publishing. Integrating Models of Personality and Emotions into Lifelike Characters

.... DAM Symposium: How to Design a Functioning Mind 17 18 April 2000 Introduction: Models of Models of Mind Aaron Sloman School of Computer Science, The University of Birmingham http: www.cs.bham.ac.uk axs Abstract Designing a Mind abbreviated as DAM is easier to type than the full title of the symposium. Many people are working on architectures of various kinds for intelligent agents. However different ....

Aaron Sloman. Architectural requirements for humanlike agents both natural and artificial. (what sorts of machines can love?). In Kerstin Dautenhahn, editor, Human Cognition And Social Agent Technology, Advances in Consciousness Research, pages 163--195. John Benjamins, Amsterdam, 2000.

....For example, we may execute some steps in a plan essentially unconsciously, while closely controlling other steps, or parts of steps. More work is also required to elaborate the detailed behaviour of the attention filter and scheduler. This is the subject of current research. Acknowledgements Aaron Sloman, Luc Beaudoin and Ian Wright read an earlier version of this paper and made many helpful comments. ....

Aaron Sloman. Architectural requirements for humanlike agents both natural and artificial. (what sorts of machines can love?). In Kerstin Dautenhahn, editor, Human Cognition And Social Agent Technology, Advances in Consciousness Research, pages 163--195. John Benjamins, Amsterdam, 2000.

....This is referred to as a reactive form of behaviour modelling. For a detailed discussion of Brooks approach and its relationship to theorem proving see [Bun94] Recent years have seen an attempt to reconcile the deliberative and the reactive approaches in single agent architectures [Slo99] This is partly motivated by looking at the human way of acting and reasoning which can be better explained as a combination of the two cases rather than by any one of them alone. Also, practical issues play an important role: in certain cases reactive behaviour is computationally more ....

....behaviour, resources are very common in this field. Important in our context is the development from planning, as described for instance in Strips [FN71] over an approach of merely reactive behaviour, as advocated by [Bro91] to very sophisticated multi level architectures, as presented by [Slo99] From an architectural point of view our system can be seen as an adaptation of the configuration for multiple planning cells of the multi agent planning architecture (as described by [WM98] to the proof planning domain. MPA is a distributed architecture which can run a number of agents in ....

Aaron Sloman. Architectural requirements for human-like agents -- both natural and artificial. (what sorts of machines can love?). In Kerstin Dautenhahn, editor, Human Cognition and Social Agent Technology. John Benjamins Publishing, 1999.

....This is referred to as a reactive form of behaviour modelling. For a detailed discussion of Brooks approach and its relationship to theorem proving see [Bun94] Recent years have seen an attempt to reconcile the deliberative and the reactive approaches in single agent architectures [Slo99] This is partly motivated by looking at the human way of acting and reasoning which can be better explained as a combination of the two cases rather than by any one of them alone. Also, practical issues play an important r ole: in certain cases reactive behaviour is computationally more ecient, ....

....behaviour, resources are very common in this eld. Important in our context is the development from planning, as described for instance in Strips [FN71] over an approach of merely reactive behaviour, as advocated by [Bro91] to very sophisticated multi level architectures, as presented by [Slo99] From an architectural point of view our system can be seen as an adaptation of the con guration for multiple planning cells of the multi agent planning architecture (as described by [WM98] to the proof planning domain. MPA is a distributed architecture which can run a number of agents in ....

Aaron Sloman. Architectural requirements for human-like agents { both natural and articial. (what sorts of machines can love?). In Kerstin Dautenhahn, editor, Human Cognition and Social Agent Technology. John Benjamins Publishing, 1999.

....restarts can also be viewed as a typical characteristic of reactive systems. For a detailed discussion of Brooks approach and its relationship to theorem proving see [9] Recent years have seen an attempt to reconcile the deliberative and the reactive approaches in single agent architectures [26]. This is partly motivated by looking at the human way of acting and reasoning which can be better explained as a combination of the two cases rather than by any one of them alone. Also, practical issues play an important role: in certain cases reactive behaviour is computationally more efficient, ....

A. Sloman. Architectural requirements for human-like agents -- both natural and artificial. (What sorts of machines can love?). In K. Dautenhahn, editor, Human Cognition and Social Agent Technology. John Benjamins Publishing, 1999.

....sets of capabilities, states and processes, and these different clusters characterise different concepts. For instance, the fullest instantiations of the CogAff schema account for at least three classes of emotions: primary, secondary and tertiary emotions, extending previous classifications. [6, 14, 23, 28]. An architecture based analysis can lead to further refinements in the classification of affective states. 25] Likewise, different concepts of seeing relate to visual pathways through different subsystems in a larger architecture. Blindsight [31] could arise from damage to connections ....

A. Sloman. Architectural requirements for human-like agents both natural and artificial. (what sorts of machines can love?). In K. Dautenhahn, editor, Human Cognition And Social Agent Technology, Advances in Consciousness Research, pages 163--195. John Benjamins, Amsterdam, 2000.

....sets of capabilities, states and processes, and these different clusters characterise different concepts. For instance, the fullest instantiations of the CogAff schema account for at least three classes of emotions: primary, secondary and tertiary emotions, extending previous classifications. [6, 14, 23, 28]. Likewise, different concepts of seeing relate to visual pathways through different subsystems in a larger architecture. Blindsight [31] is explained by damage to connections between meta management and intermediate high level perceptual buffers, while lower level pathways remain intact. ....

A. Sloman. Architectural requirements for human-like agents both natural and artificial. (what sorts of machines can love?). In K. Dautenhahn, editor, Human Cognition And Social Agent Technology, Advances in Consciousness Research, pages 163--195. John Benjamins, Amsterdam, 2000.

....against which instances of a class of designs can be compared. In simple cases we can use a fitness function , giving a numerical result. In general the relation between a design and a niche is best thought of as a complex qualitative description. 6 as outlined in (Sloman, 1994; Sloman, 1998b; Sloman, 2000b) We understand a particular architecture better if we know what differences would arise out of various sorts of design changes: which capabilities would be lost and which would be added. We also have a deeper understanding of the architecture if we can see what sorts of pressures and trade offs ....

....which architecture to propose is very difficult and is not in general constrained by experimental observations, though they certainly provide clues and tests. We can, however, constrain our theories by combining a number of considerations which I have discussed a greater length in (Sloman, 1998b; Sloman, 2000a) such as: 1) trade offs that can influence evolutionary developments, 2) what is known about our evolutionary history, 3) what is known about human and animal brains and the effects of brain damage, 4) what we have learnt in AI about the scope and limitations of various information ....

[Article contains additional citation context not shown here]

Sloman, A. (2000a). Architectural requirements for human-like agents both natural and artificial. (what sorts of machines can love?). In Dautenhahn, K., editor, Human Cognition And Social Agent Technology, Advances in Consciousness Research, pages 163--195. John Benjamins, Amsterdam.

....neurons as part of doing its job, there will not be much interference with its performance if an additional connection is made to a monitoring network getting information from many other parts of the system. Compare the Alarm mechanisms depicted in our recent papers on the CogAff architecture (Sloman 2000a; Sloman and Logan 2000; Sloman 2000b; to appear) The features of modern computing systems listed above can all be seen as continuations of the two trends of development that started in previous centuries and were accelerated by the advent of electronic mechanisms that replaced mechanical and ....

....will not be much interference with its performance if an additional connection is made to a monitoring network getting information from many other parts of the system. Compare the Alarm mechanisms depicted in our recent papers on the CogAff architecture (Sloman 2000a; Sloman and Logan 2000; Sloman 2000b; to appear) The features of modern computing systems listed above can all be seen as continuations of the two trends of development that started in previous centuries and were accelerated by the advent of electronic mechanisms that replaced mechanical and hydraulic storage mechanisms, sensors, ....

A. Sloman. Architectural requirements for human-like agents both natural and artificial. (what sorts of machines can love?). In K. Dautenhahn, editor, Human Cognition And Social Agent Technology, Advances in Consciousness Research, pages 163--195. John Benjamins, Amsterdam, 2000.

.... without an external diagram must also involve assembling possibilities for change in thinking about a 3 I conjectured in (Sloman, 1989) that some autistics lack the perceptual ability to move up levels of abstraction in perception, also described in more recent papers (Sloman and Logan, 2000; Sloman, 2000b) 7 solution to the problem. Practice somehow develops fluency in doing this: How I learnt a great deal by playing with meccano sets, as a child. Different visualisation skills are developed by mathematical or other sorts of training. What changes during such learning Experienced software ....

....the possibility of looking for and using easy short cuts requires a more sophisticated processing architecture than a typical problem solver or planner. It requires an architecture which supports mechanisms for observing, analysing, evaluating, and noticing patterns in internal processes ((Sloman, 2000a) However, having an architecture supporting such meta level abilities does not guarantee general meta level competence. It seems that humans have to learn to be reflective in different domains. e.g. someone who is good at noticing opportunities for improving his software designs may fail to ....

Sloman, A. (2000a). Architectural requirements for human-like agents both natural and artificial. (what sorts of machines can love?). In Dautenhahn, K., editor, Human Cognition And Social Agent Technology, Advances in Consciousness Research, pages 163--195. John Benjamins, Amsterdam.

....is no general overview of the space of interesting or important architectures, or the different types of requirements against which they can be evaluated, though (Dennett 1996) makes a good start. In short, there are no adequate surveys of design space and niche space and their relationships (Sloman 2000b) As a partial remedy, we offer the CogAff schema depicted in figures 1(a) b) and 2(a) b) 1.1 Conceptual confusions There is much terminological confusion, despite the confidence with which people use phrases referring to mental phenomena, including, belief , desire , intention , ....

....Humans seem to have all three. See the text and figure 2 for details. concepts. For instance, the fullest instantiations of the CogAff schema account for at least three classes of emotions: primary, secondary and tertiary emotions, extending previous classifications. Damasio 1994; Picard 1997; Sloman 2000a; Sloman and Logan 2000) Architectures differ not only between species, but also while an individual develops, and after various kinds of brain damage or disease. The resulting diversity requires even more conceptual differentiation. What it is like to be a bat (Nagel 1981) may be no more ....

A. Sloman. Architectural requirements for human-like agents both natural and artificial. (what sorts of machines can love?). In K. Dautenhahn, editor, Human Cognition And Social Agent Technology, Advances in Consciousness Research, pages 163--195. John Benjamins, Amsterdam, 2000.

....involving trajectories in two di erent but related spaces, design space and niche space. Co evolution involves multiple parallel trajectories, with complex feedback loops. This paper attempts to explain these ideas and explore some of their implications, building on previous partial expositions ([18, 20, 22, 24, 28 30]) 1.1 What is a niche A biological niche is something that can produce pressure for evolutionary change. A niche is sometimes thought of as a geographical location, but since two types of organisms in the same location (e.g. a bee and a wasp ying close together) can have di erent niches, a ....

....generations will be called e trajectories (evolutionary trajectories) Examples include development of humans and other animals from much simpler organisms and modi cations of software structures by genetic algorithms. Conjectured e trajectories leading to human minds are discussed in [5] and [28]. If identical individuals inhabit slightly di erent niches, reproductive success will be favoured by di erent traits. e.g. in a farming community physical strength may be more important than intelligence, whereas in a nearby industrialised region intelligence is more useful for acquiring ....

[Article contains additional citation context not shown here]

Sloman, A.: Architectural requirements for human-like agents both natural and articial. (what sorts of machines can love?). In Kerstin Dautenhahn, editor, Human Cognition And Social Agent Technology, Advances in Consciousness Research, 163-195. John Benjamins, Amsterdam, 2000.

.... components of the architecture: firstly the distinction between perceptual, central and action components, and secondly a distinction between types of components which evolved at different stages and provide increasingly abstract and flexible processing mechanisms within the virtual machine (Sloman, 2000; Sloman and Logan, 2000; Sloman, to appear) By analysing some of the types of states and processes that can occur within different variants of the architecture schema we find that our intuitive notions of affect, emotion, perception, belief, and other mental states and processes, correspond, ....

....of the state, along with internal high level reactions triggered by that, begin to reach the sort of complexity involved in many human pains and pleasures. Other authors Damasio (1994) Goleman (1996) Picard (1997) have distinguished primary and secondary emotions. I have tried to show elsewhere Sloman (1998, 2000, 1999) Sloman and Logan (2000) that those ideas can be both explained and generalised by relating primary emotions to the capabilities of the reactive layer of H Cogaff (also found in simpler architectures) relating secondary emotions to disturbances triggered by events in the deliberative ....

[Article contains additional citation context not shown here]

A. Sloman. Architectural requirements for human-like agents both natural and artificial. (what sorts of machines can love?). In K. Dautenhahn, editor, Human Cognition And Social Agent Technology, Advances in Consciousness Research, pages 163--195. John Benjamins, Amsterdam, 2000.

....in advance, though there are individual differences in such capabilities. The human deliberative system can also consider hypothetical past or future situations not reachable by chains of actions from the current situation, and can reason about their implications. As explained elsewhere (e.g. (Sloman 2000)) physically implementable mechanisms required for such sophistication, including a long term associative memory and especially a re usable short term memory, will cause the deliberative system to be discrete and serial, and to proceed in much slower steps than a reactive system can. A ....

.... to notice that their planning processes are wasteful or that it might be better to abandon the current goal in the light of 1 Partly inspired by Simon (1967) and elaborated in our previous papers (Sloman Croucher 1981, Beaudoin 1994, Sloman 1994, Sloman Poli 1996, Sloman 1997, Sloman 1998, Sloman 2000, Sloman Logan 1999) 5 RELATED THEORIES 7 some new information. Various forms of reactive mechanisms are found in all organisms and some of them must have developed very early in biological evolution. Deliberative mechanisms evolved later and are found in fewer organisms, though we do not know ....

[Article contains additional citation context not shown here]

Sloman, A. (2000), Architectural requirements for human-like agents both natural and artificial. (what sorts of machines can love?), in K. Dautenhahn, ed., `Human Cognition And Social Agent Technology', Advances in Consciousness Research, John Benjamins, Amsterdam, pp. 163--195.

....Deliberative mechanisms can contemplate possible novel action sequences, and situations, compare them, evaluate them and select them. In humans, they can also consider hypothetical past or future situations. They are inherently slower, more discrete and more sequential than reactive mechanisms (Sloman 2000). A meta management system can monitor, categorise, evaluate, and sometimes redirect some (not all) internal processes. None of the three layers has total control: they can interrupt one another and all can be disrupted by global alarms (described below) Various reactive mechanisms occur in all ....

....system works. Popper 1976, p. 173) 14 Conjecture: qualia The three layers can explain different notions of consciousness, ranging from simple sentience (e.g. in insects) to full reflective self awareness and possession of qualia (e.g. attendable intermediate contents of sensory processing (Sloman 2000)) Robots with appropriate architectures would have such qualia and be aware of them. 12 15 Conclusion An architecture based ontology can bring some order into the morass of studies of affect and consciousness, explaining how many conflicting theories and inconsistent definitions arise out of ....

Sloman, A. (2000), Architectural requirements for human-like agents both natural and artificial. (what sorts of machines can love?), in K. Dautenhahn, ed., `Human Cognition And Social Agent Technology', John Benjamins, Amsterdam, pp. 163--195.

....is just as hard, e.g. if we don t even know at what physical level most of the architecture is implemented. Do neurons or molecules do most of the information processing We can best constrain our theories by combining a number of considerations which I have discussed a greater length in [23, 26], such as: 1) trade offs that can influence evolutionary developments, 2) what is known about our evolutionary history, 3) what is known about Aaron Sloman Beyond Shallow Models of Emotion 3 Central Processing Perception Action Fig. 2. A triple tower model (based on Nilsson) Intelligent ....

....functional analysis combined with observation of existing organisms. Another breakdown of information processing functionality comes from both functional and evolutionary considerations. This is the triple layer model sketched in Figure 3, and discussed at greater length in previous papers (e.g. [21, 24, 22, 26, 16]) These three levels are Meta management (reflective processes) newest) Deliberative reasoning ( what if mechanisms) older) Reactive mechanisms (oldest) Fig. 3. The triple layer model There is good reason to believe that early organisms, like some existing organisms, were totally ....

A. Sloman. Architectural requirements for human-like agents both natural and artificial. (what sorts of machines can love?). In Kerstin Dautenhahn, editor, Human Cognition And Social Agent Technology, Advances in Consciousness Research. John Benjamins, To appear.

No context found.

Sloman, A. 1998. Architectural requirements for humanlike agents both natural and artificial. In: Human cognition and social agent technology. v.19, p. 163-195. John Benjamin's Publishing Company.

No context found.

Sloman, A. (1999). Architectural requirements for human-like agents both natural and artificial. In K. Dautenhahn (Ed.), Human Cognition And Social Agent Technology, Advances in Consciousness Research. Amsterdam: John Benjamins Publishing Company.

No context found.

A. Sloman. Architectural requirements for human-like agents both natural and artificial (What sorts of machines can love?). In K. Dautenhahn, editor, Human Cognition and Social Agent Technology. Advances in Consciousness Research, pages 163--195. John Benjamins, Amsterdam, 2000.

No context found.

A. Sloman. Architectural requirements for human-like agents both natural and artificial (What sorts of machines can love?). In K. Dautenhahn, editor, Human Cognition and Social Agent Technology. Advances in Consciousness Research, pages 163--195. John Benjamins, Amsterdam, 2000.

No context found.

Sloman, A.: 1999, 'Architectural requirements for human-like agents both natural and artificial '. In: K. Dautenhahn, (ed): Human Cognition And Social Agent Technology, Advances in Consciousness Research, 19, Amsterdam, John Benjamins Publishing Company.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC