[Newell, Rosenbloom, Laird, Symbolic Architectures for Cognition]
p.99 Requirements on the Cognitive Architecture...
1. Behave flexibly as a function of the environment
2. Exhibit adaptive (rational, goal-oriented) behavior
3. Operate in real time.
4. Operate in a rich, complex, detailed environment
a. perceive an immense amount of changing detail
b. use vast amounts of knowledge
c. control a motor system of many degrees of freedom
5. Use symbols and abstractions
6. Use language, both natural and artificial
7. Learn from the environment
8. Acquire capabilities through development
9. Live autonomously within a social community
10. Exhibit self-awareness and a sense of self
[Alan Allport, Visual Attention]
p.631Issues of attention bear on every area of cognitive science... What
would a computational theory that was adequate to the functions of attention be like? What is the overall purpose
(or what are the overall purposes) of attention, and what are the determining and enabling constraints on
attentional processes?
p.632-633 The fundamental constraint that underlies all the operations of
attention, imposing their essentially selective character, is the limited information-processing capacity of the
brain... the basic function or purpose of attentional mechanisms is to protect the brain's limited capacity system
(or systems) from informational overload.
p.633 nonselected sensory inputs were no longer supposed to be strictly
excluded from the limited capacity systems, but only attenuated in signal strength... the basic concept - of limited central
processing capacity as the fundamental causal constraint that imposed the need for selectivity of processing - remained unchanged.
p.636 The efficiency of selective cueing is appropriately indexed, for example,
by the presence or absence of response interference resulting from the presentation of task-irrelevant distractors.
p.636 Efficient selective cueing, however, defined in terms of the effective
designation (and segregation) of information for the control of a particular action, carries no necessary implications about
the level of processing or nonprocessing of the information that has not been thus designated, and still less, any implications
about the level or domain of processing in which capacity is limited.
p.637 If there is a limitation on processing capacity that is responsible
for the selectivity of attention, what is its computational basis?
p.639 Shiffrin and Schneider further assumed that all attentional
selectivity and all forms of distractor interference resulted from competition for (and capture of) the same limited control
(or the same controller).
p.647 spatial selectivity... appears to be implemented through a process
of positive priority assignment rather than through suppression or attenuation of nonselected locations.
p.647 Spatial attentional selectivity in this respect evidently acts to
suppress extraneous actions, not to attenuate the quality of sensory input.
p.647 Within systems responsible for the spatial direction of attention,
moreover, there appears... clear indication of a process of prioritization (selective enhancement).
p.648 Many authors have suggested that the selectivity of attention is in
some way related to or dependent on the need for coherent control of action.
p.649 The necessary control of action of course includes decisions not
to perform a given action. The sensory input is clearly able to govern decision processes affecting action at many different
levels, including decisions about what category of action and whether (as well as when) it is appropriate to release
it. Selection among competing sources of control parameters in the sensory input is equally essential for the coherence of
action preparation, or action planning, even though no overt action is in fact released or is released at a later time...
The attentional selectivity with which we are concerned is therefore selection for potential control of action.
p.649 one activity must be given priority while the other is postponed or
abandoned.
p.650 For effective visual-motor control therefore visual selection-for-action
must be capable of being focused selectively on any one coherent source of visual information (potential control parameters),
however that information might be spatially distributed in the visual array. Very often the appropriate source of control
parameters can be characterized as a separable visual 'object' - that is, an entity that can be individually named, categorized,
avoided, or otherwise independently acted on... This constraint implies that (possibly very complex) processes of perceptual
grouping and segmentation may have logically to precede the effective focusing of visual selection-for-action...
Many factors are known to influence perceptual segmentation. It is therefore to be expected that these factors will interact
very powerfully with the efficiency of attentional selection.
p.651-652 The following basic ecological constraints impose a number of
strong requirements on any attentional motivational system.
Unpredictability and time constraint First, we are concerned
with systems that have to operate in an environment that is, at best, only partly or incompletely predictable...
Multiple goals Second, we are concerned with systems with
a very wide range of potential goals of action. Priority (importance, urgency) regarding the implementation of different goals
must be continuously adjusted, moment by moment, as potential threats or opportunities arise in the environment and as internal
conditions... as well as cognitive variables (plans and subplans, 'insights,' and so on) change or develop. Priority assignment
is necessary...
Multi-functional systems Third, we are concerned with multi-functional
organisms (human and other species, natural or artificial). By this is meant organisms (or robots) whose subcomponents (sense
organs, effectors, cognitive subsystems) are not in general uniquely dedicated to particular goals or to particular categories
of action. Subcomponents must therefore be selectively engaged and coordinated to implement particular activities and particular
goals.
p.652-653 The critical problem for any attentional system (in multifunctional
organisms) is therefore how to satisfy two conflicting requirements: the need for continuity of attentional engagement, against
the need for its interruptibility. Failure to shift attentional engagement when faced by environmental threats (or opportunities)
can of course be fatal to an organism's survival or physical integrity. At the other extreme constant shifting of fragmentation
of attentional engagement, triggered by every sensory event of environmental affordance, would make sustained, purposeful
activity impossible and result only in behavioral chaos. Between these two equally disastrous extremes lies a range of more
or less viable solutions. All of them depend on some means of evaluating, or at least estimating, the relative motivational
importance and temporal urgency of the potential threats and affordances outside the current attentional engagement,
relative to the estimated importance and urgency of the current activity or activities.
p.653 Humans (and other species) appear to have adopted a combination of
several different partial solutions to this problem, which can be broadly grouped into three categories: (1) internally generated,
predictive control of attention shifting, depending on a range of heuristic processes of widely varying levels of
sophistication: (2) externally elicited shifts of attentional engagement, cued by the detection of more or less complex
triggering conditions; and (3) active combinations of (1) and (2), as, for example, in many forms of exploration and search.
I should therefore add, as a further indispensable attentional function, continuous (parallel) monitoring of the environment
(internal as well as external) for changes relevant to current and long-term goals.
p.653 Environmental monitoring may be expected to operate at many different
levels. These should include fast, relatively crude or approximate systems, operating on rule-of-thumb (that is,
associative) criteria, both learned and unlearned, as well as possibly slower and more sophisticated systems, supported also
by processes of (discontinuous) predictive search under goal-directed control. Detection of a critical external or internal
event must be capable of causing temporary interruption of inhibition of the current attentional engagement and a (rapid)
shift of sensory priority assignment (selection-for-action) toward the information source responsible for the interruption
- in other words it must be capable of triggering attentional orienting. Equally important, sensory-motor orienting
must be capable of being inhibited by the competing, higher priority of the current attentional engagement.
Computation of this critical balance of priorities appears to be
one of the central problems for any system of intelligent attentional control.
p.654 In these models... executive control [JLJ - of attention] is exercised
only indirectly, typically through a mechanism of competitive priority assignment
p.655 Reaction time to a secondary probe stimulus (probe RT) has often been
presented as a method of assessing the momentary processing capacity demanded by a concurrent primary task.
