Relationally structured beliefs--论文代写范文精选

可以说,解耦的证据表示可以正确地解释。这似乎显示解耦表示，在正确的选择情况下。值得注意的是,大鼠海马损伤也导致这一问题。下面的essay代写范文进行详述。

Abstract
If decoupled representation appears in the early evolution of nervous systems this puts into focus the question of whether the definition of decoupled representation adequately captures belief-like representation. As described earlier, the justification for this claim lies with the structure of the overall account and its empirical support. The important point here is that there is no independent support for the claim that decoupled representations are belief-like, other than the overall architecture and the evidence on which it is based. The idea that very simple neural phenomena fit the definition for decoupled representation is disorienting, and there is a temptation to think that these can’t really be decoupled representations because they aren’t very belief-like. But this would get the cart before the horse. It’s a hypothesis that decoupled representation is a good definition of belief-like representation, and if such simple phenomena satisfy the definition this is counter-evidence to the hypothesis. Rejecting these cases as instances of decoupled representation because they aren’t belief-like would beg the question.

Yet Sterelny reviewed research on navigation, tool use, ecological knowledge and social cognition, and found no clear evidence for decoupled representation, although he thought it plausible that great ape social cognition would involve decoupled representation. It is strange if decoupled representation can be found in very simple circumstances, and not in the case of more complex behaviors where it might be expected to play a stronger role. Arguably, however, evidence for decoupled representation can be found in these kinds of cases, when interpreted properly.

For instance, Dusek and Eichenbaum (1997) developed a transitive inference task for rats using spices buried in sand to provide distinct odors. The rats could dig in the sand and would in some cases find cereal. Training involved presenting rats with pairs of odors and rewarding according to the scheme A > B, B > C, C > D, D > E, where A > B means that the rat should choose odor A over odor B to find the food. In the test trial the rats were presented with the pair BD, which they had not previously encountered. To choose correctly they should pick B. In the experiment 88% of normal rats performed the transitive inference correctly. This appears to show decoupled representation: the new behavior is the correct choice in the novel situation, the pairing of B and D. The representation is the relational scheme extracted from prior training. The most interesting result from the experiment was that rats with a disconnected hippocampus performed at chance on the transitive inference probe, even though they learned the odor pairings presented during training just as quickly as the normal rats. To put it another way, for rats with an intact hippocampus the BD pair was an extension of the problems they had been encountering recently, whereas for rats without a hippocampus the BD pair was a new problem for which they could only choose at random.

Notably, rats with hippocampus damage are also impaired at solving water maze tasks compared with normal rats. In a typical Morris water maze task the rat is placed in a pool in which there is a platform below the surface, which the rat can’t see because the water is opaque. The pool is placed in a room with landmarks, and the rat must swim around until it finds the platform. The rats are released from varying starting positions during training. Normal rats are able to solve the problem, evidently learning to locate the platform using the configuration of landmarks, but rats with a damaged hippocampus can only learn to locate the platform if they are released from a fixed starting point, apparently able to associate the platform location with a single landmark, but unable to flexibly locate the platform within a landmark array (Eichenbaum 2000).

Taken together, this evidence suggests that the hippocampus is involved in extracting stable relational information that allows flexible problem solving, whilst rats without a hippocampus can only learn fixed responses. The relational information is a kind of decoupled representation, and it contributes to flexible behavior control. Above it was argued that the perceptual discriminations that participate in associative learning count as decoupled representations because they aren’t coupled prior to learning, and can be recoupled with ongoing learning. The decoupled aspect is easy to overlook because, once learned, the effect of the associations on behavior control is inflexible.9 That is, the association is something like an acquired detection system. The kind of learning demonstrated here shows a more powerful kind of decoupled representation which can flexibly influence dynamic behavior control. Moreover, the hippocampus is associated with declarative memory, and the results are consistent with Eichenbaum’s theory that declarative memory is constructed by extracting relational information across multiple experiences (Eichenbaum 2001).

In psychological research declarative memory has been distinguished from implicit memory (Cohen and Squire 1980), and is associated with conscious control and reasoning. It is thus, plausibly, the cognitive basis for what we usually think of as beliefs. If the evidence described is correct then Sterelny’s intuition that there is some important form of representation supporting beliefs is right, and it is correct that this kind of representation has a distinctively flexible role in behavior control. However the definition of decoupled representation doesn’t effectively pick out this kind of representation. Eichenbaum’s theory of declarative memory is one alternative cognitive model of ‘belieflike representation’.

The overall picture is thus not as strange as it might have seemed. It would have been very odd if decoupled representations appeared in simple perception and behavior control, but were absent in more advanced behavior. However decoupled representation can be found in more complex animal behavior, and in fact we can distinguish multiple kinds of decoupled representation. There is the basic form of decoupled representation that participates in simple associative learning, and a more sophisticated, relationally structured kind of decoupled representation involved in flexible behavior control. This will be termed ‘model-based representation’ here because the extraction of stable relationships is effectively model building.

Evidence for preferences in rats 
Evidence for a relatively sophisticated form of decoupled representation in rats has implications for Sterelny’s arguments concerning rat preferences, because decoupled representation and preferences should arise in roughly the same circumstances, as discussed in section 2.2.2. This link is apparent in the reasons Sterelny gives for doubting that rats have preferences, where he says that the problems faced by rats don’t require “broad responses” (associated with decoupled representation). Sterelny says that the rat world may be complex in the sense that knowing when intervention is needed is a difficult problem, however “the interventions are few in number, simple, and structureless” (p. 92). It’s not entirely clear how to interpret these claims, but it looks like the transitive inference experiments described above are counter-evidence. The rats have ‘broad responses’ in the sense that they can appropriately adjust their behavior to a new problem that is related to previous problems they have encountered. The behaviors are structured in the sense that they are sensitive to problem structure.

In section 2.2.2 it was said that humans make decisions – they choose particular actions taking into account structured features of the situation – and this involves cognitive valuation. In light of what we’ve just considered it can also be noted that decisions, in this sense, require model-based representation. Further, we can draw a distinction between flexible valuation based on habit learning (Dickinson 1985), that doesn’t involve decision because option values are incrementally modified based on their experienced outcomes, and cognitive valuation involving decisions based on model-based representation. If rats have model-based representation then they are likely to make decisions, in the sense specified, and they should have preferences, in the sense of cognitive representations of outcome value. It’s important, then, to consider Sterelny’s arguments against preferences in rats to see whether they tell against this conclusion.

At this point some clarification is needed. To review, on Sterelny’s taxonomy there are two forms of motivation, drives (4a) and cognitively based preferences (4b), and he specifies four conditions which should select for a 4a → 4b transition: (i) the animal’s behavioral repertoire includes many possible options, (ii) a wide range of resources is needed, which are hard to encompass with a repertoire of built-in drives, (iii) summation mechanisms for determining which drive controls behavior don't produce good results, and (iv) the sensory profile of resources doesn't stay stable over evolutionary time. Condition (iv) suggests that preference-based motivation will be advantageous in conditions where learning is advantageous, and Sterelny’s notion of drive-based motivation appears to be something like a specific hungers picture which involves minimal learning (see e.g. p. 86; compare Bouton 2007, p. 330). 

One of the major drive theories, that of Hull, incorporates learning, however. On Hull’s account motivational state multiplies with learning to produce behavior strength (Bouton 2007, p. 331), and motivational state acts as a reinforcer: behaviors that reduce need are strengthened (Bouton 2007, p. 251). For a Hullian drive system there is thus no need for the sensory profile of resources to be stable over evolutionary time. Conditions (i) and (ii) are also consistent with Hullian drive; condition (ii) because by means of secondary reinforcement an animal can learn to value resources not covered by its repertoire of primary reinforcers. Hull’s theory doesn’t involve preferences, so conditions (i), (ii) and (iv) don’t distinctively select for preference-based motivation.（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)