A revised theoretical picture--论文代写范文精选

Sterelny只识别生态泛化,并遵循社会智力假设，高度重视社会世界的复杂性。考虑到建筑的材料和功能性,许多感性的问题是很困难的,因为身体问题从一系列复杂的环境中提取相关信息。下面的essay代写范文进行详述。

Abstract
Sterelny’s highest grade of representational complexity is achieved by the early Cambrian, which means that the subsequent half billion years of cognitive evolution in animals – i.e., almost the entire course of multicellular animal evolution – has involved more complex phenomena than the theory can account for. The cognitive taxonomy must be modified and enriched, but the selection pressures too may be stronger and more complex than envisaged. It was suggested in section 3 that Godfrey-Smith’s flexible control account of selection for cognition should be modified in favor of a generalized formulation in terms of behavior targeting, including both flexible control and the targeting of individual behaviors. The concept of translucency in effect makes this shift, since it is a problem applying to particular behaviors, but it could be clarified to more explicitly encompass behavior targeting in general. Such a reformulation might specify both an umbrella conception of behavior targeting problems, and particular targeting problems that favor particular kinds of representation and control. An important element of this would be recognition of multiconditionalization as a behavior targeting strategy.

Failing to recognize multiconditionalization is one of the most significant weaknesses in Sterelny’s account; as argued above, we should expect that multiconditionalization evolves early and plays a profound role in cognitive evolution.11 The analysis of the sources of translucency also needs to be amended. Sterelny only identifies ecological generalization and hostility as factors contributing to translucency, and following the social intelligence hypothesis he places great emphasis on the complexity of the social worlds of the great apes. Perception of the inanimate and nonhostile biological world is argued to tend towards transparency. This makes almost no allowance for ‘physical’ problems of perception and behavior. 

Given the material and functional constraints of building perception from neural systems, many perceptual problems are difficult because of the physical problems of extracting the relevant information from a complex array of ambient signals. Similarly, there are many difficult ‘physical’ challenges in motor control and problem solving, such as solving problems with multiple related factors. These kinds of problems are surely major spurs and brakes on cognitive evolution; even humans are easily overwhelmed by problems with more than a handful of features, whether they are social or not.12 In addition to the environment, the complexity of the organism itself makes a significant contribution to the problems of behavior control. It makes behavior targeting more difficult because for a more complex organism there are more options to be controlled, and more variation that affects perception and behavior success. The role of self-generated movement in making perception more difficult is a simple example. Importantly, this will select for enhanced perceptual and control abilities, as illustrated by the evolution of nystagmus.

With regard to architecture, Sterelny’s account was characterized as a two-stream view of cognitive complexification (representation/control + motivation) in section 3.1.2, and contrasted with a three-stream view (representation + control + motivation). It was flagged there that control complexification isn’t merely an issue that Sterelny doesn’t address; he can’t coherently recognize it as an independent stream. The reason for this becomes clear when we note that the postulated late appearance of decoupled representation is also a late separation of indicative and imperative function. If the reduction of control uncertainty and indication uncertainty are generally welded together there is no room for independent control complexification, because control is automatically furnished by perception. 

On the other hand, if indication and control uncertainty separate early, as the evidence indicates, then there is a functional need for specialized control because an integration step is needed to bring perceptual information to bear on behavior. Increasing behavioral complexity will also favor control specialization, as behavior management mechanisms such as depicted in figure 7c are needed to handle response conflict and other behavior coordination problems. In addition, as relations between perception and control become more complex increasingly sophisticated valuation mechanisms will be needed to bridge the gap.

A relatively simple taxonomy covering the various lines of evidence discussed to this point identifies three kinds of control (figure 19). Fixed pattern control generates a stereotyped pattern that is not sensitive to learning. ‘Cache-based’ learning flexibly modifies the pattern of response over time, but is inflexible in immediate control: options are assigned a ‘cached’ value that is not recalculated at the time of choice. Model-based control is flexible both over time and in immediate control, where the value of options and the pattern of behavior is decided ‘on the fly’ based on models that capture aspects of the problem structure.13

In principle a given architecture might incorporate one, two, or all three forms of control (figure 20). When more than one form is present interactions between them can occur. For instance, learning can modulate fixed pattern controllers, modifying output. To avoid definitional confusion here we must note that what is fixed for a fixed pattern controller is the patterning generated by the controller itself. External modulation of the controller may result in altered patterning of the controller output, and if the fixed pattern controller is only one of several contributing to a particular behavior then the behavior may show labile aspects despite the fixity of the contribution from the fixed pattern controller. A controller itself can also have both fixed and flexible aspects. Cache-based and modelbased controllers may cooperate or compete with each other and with fixed pattern controllers in the control of a particular behavior (see e.g. Daw et al. 2005, Yin and Knowlton 2006).

A ‘three system’ control architecture (figure 20c) has an especially powerful and flexible way of managing control problems. Over evolutionary time fixed pattern control is adjusted to regularities stable across these timescales. In ontogenetic time cache-based control adjusts to capture regularities stable over extended periods of time. Some control problems, however, have variability in surface behavior-outcome relations but deeper structural regularities, and model-based control provides a mechanism for solving these kinds of problems. With regard to model-based representation, the example discussed in section 4 was relationally structured memory dependent on the hippocampus, but there is likely to be more than one kind. 

Model-based representation is advantageous anytime one or more immediate perceptual signals are ambiguous and this ambiguity can be resolved using stored structured information. Locally ambiguous and incomplete information is endemic to perception; for instance objects are only seen from one perspective at a time, and are often partially obscured. Geon theory – the idea that objects are represented in terms of a basic set of geometric shapes (Biederman 1987) – illustrates the kind of role that modelbased representation can play in perception, but more generally all sorts of frequently encountered perceptual relations might be captured in models. Model-based representation also plays a role in motor control, where forward and inverse models supplement the incomplete control information provided by perception (Wolpert et al. 1995).

Decision problems (as defined above) provide a starting point for understanding the kind of selection that would favor model-based representation and control. A decision problem can be thought of as a kind of translucency in which the mapping between correct choice and available options is variable across choices, yet still exhibits regularity. This is effectively what Shiffrin and Schneider (1977) referred to as a ‘variable mapping problem’ and used to study executive control processes in humans. Together these additional factors provide a richer framework for understanding cognitive complexification. This is a starting point for understanding both complexification in early animal evolution, and why rats should turn out to have relatively sophisticated kinds of representation, control, and motivation. A complementary issue is why cognitive evolution should be so protracted. It is being suggested here that rats have versions of the cognitive abilities central to human agency, yet the agency of rats is manifestly nothing like as sophisticated as the agency of humans. 

The basic answer is probably just that the problems are demanding, and the elaboration of increasingly refined abilities is an extended process. There is a tendency to understand the evolution of cognition in terms of the simple presence or absence of abilities, but this is clearly not the right picture for other kinds of traits. Around 380 million years of evolution separates the earliest terrestrial tetrapods – not unlike salamanders – and the cheetah, and it is clear that there are many challenges that must be overcome to get from a basic capacity for terrestrial locomotion on four limbs to the running abilities of a cheetah. A long progressive refinement of abilities may be as important in cognitive evolution as it has been in tetrapod locomotion. Concomitantly, trying to explain the distinctiveness of human agency in terms of the appearance of the basic agentic capacities involved might be a bit like trying to explain the running abilities of cheetahs in terms of the evolutionary appearance of legs.

Conclusion 
It’s been argued here that a number of aspects of Sterelny’s account are not correct, but some important core ideas are right. Translucency, robust tracking and decoupled representation are all valuable concepts, and the amendments and extensions suggested here build on this. These include a generalized conception of behavior targeting, a concept of model-based representation, and a three-stream picture of cognitive complexification. Ultimately, for this kind of theory to prove its worth it must provide a unifying framework for empirical comparative cognition research, and I conclude by briefly suggesting how the ideas canvassed here could help integrate several common, unintegrated conceptions of the evolution of cognition. In the introduction it was said that ideas on the evolution of cognition are diverse. 

This is true, but there are some core ideas that are widespread, including (i) cognition is information processing more complex than can be accounted for by simple associative learning, (ii) cognition involves mental models, and (iii) cognition involves a transition from domain-specific to domain-general abilities. The taxonomy of figure 19 provides a basis for putting these ideas into a common conceptual framework. A simple transition from one kind of control to another is unlikely to capture very well either the appearance of cognition per se, or the appearance of advanced, human-like cognition. An architectural approach characterizing the elaboration of particular forms of control and interactions between multiple kinds of control would provide a more nuanced picture. A basic overall theoretical understanding would include an account of the core properties of each kind of control, an account of the properties of multi-system architectures, and an account of the factors that drive various kinds of elaboration of multi-system architectures. If the distinctive capacities of human agency lie with a particular kind of integrated architecture, as suggested above, then such theory will be an important part of a fundamental understanding of human agency.（essay代写)

51Due网站原创范文除特殊说明外一切图文著作权归51Due所有；未经51Due官方授权谢绝任何用途转载或刊发于媒体。如发生侵犯著作权现象，51Due保留一切法律追诉权。
更多essay代写范文欢迎访问我们主页 www.51due.com 当然有essay代写需求可以和我们24小时在线客服 QQ:800020041 联系交流。-X（essay代写)