Notebook

Buckner, C. (2017): Rational inference: the lowest bounds

PDF Link

Inference

Here’s a nice summary of the positive view:

*we are thus left with the following new package: intuitive practical inferences are a) categorization procedures that are b) causally sensitive in well-understood ways to the context-sensitive intensions of the representational states that drive them and c) epistemically rational according to their ecological fit to the informational structure of the agent’s environment. We thus here reject CI entirely, but finesse Taking by noting the intensional sensitivity of these judgments for the purposes of psychological explanation, allowing us to draw a firm border between such similarity judgments on the one hand and the intensionally-insensitive causes of behavior like headaches, lexical priming, and “mere concatenation” associations on the other. In slogan form, inference begins at configural conception.* (p. 16)

  • Comment Summary

    • Summary, Page 1, Highlight:

      the judgments of nonlinguistic animals

      This seems to ignore the sense in which something counts as a judgment


    • Summary, Page 1, Underline:

      *I aim to establish the lowest bounds of rational inference, arguing that many (though not all) of these opaque judgments in nonlinguistic animals should be counted as inferential.*


    • Summary, Page 2, Highlight:

      on the basis of beliefs and desires

      I’m somewhat skeptical of the notion of “on the basis” of that’s being appealed to here - it seems like it could be automatic and non-conscious


    • Summary, Page 2, Highlight:

      processes in which psychological states non-mentally cause a behavioral outcome

      I’m a bit confused by what “non-mentally cause” is supposed to mean here. The examples don’t help since, e.g. a headache’s causing me to miss a question still seems like it is a form of mental causation.


    • Summary, Page 2, Highlight:

      causal connections are arational

      this seems to be the key – psychological states non-rationally cause a behavioral output.


    • Summary, Page 2, Underline:

      *In this paper, I evaluate two popular ideas which have been deployed to police this boundary between inferential and non-inferential psychological causation: 1) the “Taking condition”, which holds that inferring necessarily involves thinkers somehow taking their premises to support their conclusions (Boghossian, 2014; Valaris, 2016), and 2) a view about the form of practical inference that I will hereafter call “Classical Inferentialism” (CI), which holds that one’s premises (inferentially) support their conclusion when reached by syntactically fitting those premises and conclusion together in the schema of a formal rule, such as those provided by deductive logic or decision theory. Each idea is independently popular, and they are commonly paired—despite frequently acknowledged tensions. Taken together, these ideas preclude the possibility that nonlinguistic agents could draw genuine inferences, because using formal inference rules probably requires language.*


    • Summary, Page 3, Highlight:

      They thus seem to satisfy at least one core idea associated with Taking and CI: that some nonlinguistic judgments can be sensitive to the representational contents of the cognitive and conative states governing the agent’s actions.

      this is not getting what is core to the appeal to the taking condition (see e.g. boghossian 2016). Boghossian and others’ point is that there are lots of ways to be “sensitive to the representational content” of a state – what is important is delimiting what counts as appropriately rational sensitivity.


    • Summary, Page 4, Highlight:

      without requiring that in doing so they must connect fully propositional contents.

      There is a real worry here of cross-talk, since it isn’t clear what sense is to be made of an “inference” that isn’t between propositional contents. Isn’t inference just the (special kind of) transition between one contentful mental state and another?


    • Summary, Page 4, Highlight:

      First, Boghossian (2014) simply stipulates that genuine inference begins at “System 1.5 and above”, referencing the dual system/process literature in human psychology (Evans, 2008); but what exactly counts as System 1.5, and why the boundary should be located precisely there, is left unexplained.

      This is a ridiculous objection. Boghossian is very clear that he wants to distinguish a certain kind of reasoning, but not one that is exactly what Kahneman identifies as ‘system 2’.


    • Summary, Page 4, Highlight:

      ethicists and action theorists trying to mark out the lower bounds of inference tend to be more concerned with questions of responsibility and ownership than psychological explanation, and these issues may not be directly related.

      This is somewhat bizarre given that the most prominent ethical theories in question—constitutivists—are going to what to say that motivation (psychological explanation) and normative issues, including responsibility, have to be directly linked.


    • Summary, Page 5, Underline:

      *In this paper, I explore an alternative, broadly Lockean strategy for articulating the lower bounds of inference. This strategy takes the relevant rationalizing connection to be intensional, but intuitive rather than formal in nature.*


    • Summary, Page 5, Highlight:

      classical empiricists like Locke and Hume saw no tension between association and inference

      this is because classical empiricists were either incoherent in their conception of inference or denied that there was any necessary connection between states or contents.


    • Summary, Page 5, Highlight:

      The problem is that flexible thought precedes learning these forms, which are mastered only with effort.

      If this is disparaging of syllogism then Descartes was equally if not more so. However, Descartes did not thereby think that that association was a kind of inferential process. So maybe the emphasis on “Classical Inferentialism” is mistaken?


    • Summary, Page 5, Highlight:

      I here answer this charge by offering a psychologically-plausible model of the relevant faculty of nonlinguistic judgment. On this model, nonlinguistic inferences can be causally sensitive to particular modes of presentation of an agent’s mental states, though the details of the rationalizing connection may be opaque to introspection.

      I don’t understand how the beginning and end of this sentence are compatible.


    • Summary, Page 5, Underline:

      *this model rejects CI and finesses Taking—the idea being that nonlinguistic practical inference can be understood as a form of learned, similarity-based categorization judgment, operating over complex, configural associative structures, in which one behavioral option is selected, on intensionally-sensitive grounds, as matching the type of action which is desired.*


    • Summary, Page 6, Underline:

      *my goal here is only to mark the lowest bounds of practical inference, with the hope that more sophisticated forms of inference can be bootstrapped up from this base case by adding additional components.*


    • Summary, Page 6, Highlight:

      I can reason about their relative locations in a way which goes beyond anything I observed directly in my personal experience.

      is this supposed to be relevant to what inference is?


    • Summary, Page 6, Highlight:

      the point stands that maps can ground substantial inferences even when they do not replicate logical structure.

      this seems too quick. the issue of whether maps instantiate logical structure is directly relevant to whether they can figure in inferential transitions. if it turns out that maps have no, or too little logical structure then this would show that they cannot stand in inferences. likewise, if we have good evidence that map-like representations figure in inferences, then one thing this might show is that they have logical structure.


    • Summary, Page 7, Highlight:

      It is hard to deny that something relevant to Taking has occurred here, especially when such maps are represented entirely “in the head”.

      I’m not sure why the defender of the taking condition would concede that there is logical inference goin on here rather than some other form of reasoning


    • Summary, Page 7, Highlight:

      taking the structure of the map to correspond to the structure of the world

      but this explicit form of taking seems very unlikely to be done either in non-human animals or (mentally) in humans.


    • Summary, Page 9, Underline:

      *To evaluate the relevance of configural learning to inference, I suggest that instead of starting with classical logic or decision theory and revising downward—by trying to extract protological structure from configurations or idealize these judgements as somehow ranking disparate options in an abstract, fungible currency like utility—we should try a more biologically-inspired, bottom-up approach. Specifically, we should focus on the kinds of practical decisions animals make on a day-to-day basis*


    • Summary, Page 9, Highlight:

      This kind of specialized configural selection is the kind of judgment where we should expect to see a Taking-like inferential sensitivity take shape, where the individual’s own epistemic character begins to guide decision-making.

      I think more needs to be done to explain what a “taking-like” inference is supposed to be.


    • Summary, Page 10, Underline:

      *The picture of nonlinguistic practical judgment that I am offering here is that most animals have a relatively small set of motivational modes divided by basic biological needs and drives that serve as primitive reasons for acting (Hurley, 2003), reasons that can motivate a range of behavioral responses that can perhaps be flexibly chained.*


    • Summary, Page 13, Underline:

      *While I agree that some form of Taking is a non-negotiable component of rationalizing explanation, I argue below that procedures operating on other non-language-like representations like configural patterns are sufficiently intensional to satisfy this demand*


    • Summary, Page 14, Underline:

      *Thus the task in question can be understood as identifying the action to-be-performed as an instance of a type of action towards which the agent has a pro-attitude. Practical inference can then be understood as a categorization judgment; and once the judgment is so broadly construed, models of inference and rationality other than those provided by CI become available for discussion.*


    • Summary, Page 14, Highlight:

      eligible

      eligible for what?


    • Summary, Page 15, Highlight:

      our idea of the prototypical police officer for rural Texas is quite different from our idea of the prototypical police officer in Trafalgar Square.

      in fact - it may be sufficiently different that the prototypes end up with different extensions (e.g. cops with guns vs. cops without guns). so now I’m not sure how to reconcile the example with the previous statement about differing intensions having the same extension.


    • Summary, Page 15, Highlight:

      This observation importantly implies a concession to particularism about practical judgment—what counts as relevantly similar will depend upon the category’s intension in that particular context—but it is not clear why this is inappropriate, especially for the kinds of nonlinguistic judgments we are considering.

      isn’t its inappropriateness related to the issue of concept identity? different extensions mean different concepts


    • Summary, Page 16, Highlight:

      From this perspective, the evaluative component of practical rationality concerns the “bet” that agents make about the informational structure of their environment, and the bet is justified if it achieves a good fit with environmental circumstance.

      one of the most obvious criticisms of this is that it isn’t clear that the animal makes the bet rather than some subsystem within it.


    • Summary, Page 16, Highlight:

      we must give up on the idea that one can be guaranteed to act rationally just by “following the rules”.

      it isn’t clear that this is a widely-held view, especially if one finds new evil-demon type worries persuasive


    • Summary, Page 16, Underline:

      *we are thus left with the following new package: intuitive practical inferences are a) categorization procedures that are b) causally sensitive in well-understood ways to the context-sensitive intensions of the representational states that drive them and c) epistemically rational according to their ecological fit to the informational structure of the agent’s environment. We thus here reject CI entirely, but finesse Taking by noting the intensional sensitivity of these judgments for the purposes of psychological explanation, allowing us to draw a firm border between such similarity judgments on the one hand and the intensionally-insensitive causes of behavior like headaches, lexical priming, and “mere concatenation” associations on the other. In slogan form, inference begins at configural conception.*


    • Summary, Page 16, Underline:

      *this view comprises a novel hybrid of traditional internalism and externalism: judgments are internally rational given an intensional match between a target category and an exemplar, but externally rational according to an ecological match to informational context.*


    • Summary, Page 16, Highlight:

      a novel hybrid of traditional internalism and externalism

      this is not a novel hybrid. internalist explanatory rationality and externalist justificatory rationality combined together has been a view at least since Alston 1998’s “an internalist externalism”


    • Summary, Page 16, Highlight:

      Too often these two questions have been conflated

      ???


    • Summary, Page 19, Highlight:

      Whether we call these judgments perceptual or not, there would remain a significant sense, relevant to Taking, in which they would simply count as inferential, because if a different configuration of cues were associated with the same goal-category, it would lead the animal to different categorizations.

      I’m not sure about this – why is the fact that there is counterfactual dependence sufficient for classification as inference?


    • Summary, Page 22, Highlight:

      what might be successfully argued in the passage from Valaris above is that one cannot be held epistemically responsible for taking a rapid configuration of facial cues to suggest a particular emotional state; but contra Valaris’ apparent conclusion, when executive resources are available, one does have control over whether one acts on that take.

      I’m not sure that this appropriately responds to Valaris


    • Summary, Page 22, Underline:

      *My take on the current state of the empirical literature is that we should lean towards the more inclusive demarcation, because we should expect executive control to come in degrees and be modulated by a great many general cognitive resources that we already know vary widely across the Animal kingdom, like working memory capacity, intelligence, and attention.*


    • Summary, Page 23, Underline:

      *This essay offered a novel model of nonlinguistic practical inference, based in the relevant empirical psychology and philosophical tradition. This model defends something that much philosophical orthodoxy has regarded as incoherent: that opaque inferences can be internally rational in the sense relevant to Taking. The key to this solution was to treat the questions of intensional fit and epistemic evaluation separately, offering an internalist solution to the former and an externalist solution to the latter. Rational inference, it was suggested, bridges these two forms of rationality by allowing intensional takes to guide adaptive decision-making—in short, it is the process whereby agents exert some control in placing their own ecological bets.*


  • Interesting

    • Interesting, Page 1, Highlight:

      We thus need to decide when such judgments should be pushed down to the non-inferential level by pointing out their automaticity and lack of integration with other epistemic states (e.g. Gendler, 2008), upgraded to inferential status by emphasizing their sensitivity to evidence and inferential promiscuity (e.g.  Mandelbaum, 2013), or declared a sui generis middle ground that fits into neither of these better-understood categories (e.g. Levy, 2015).


    • Interesting, Page 1, Highlight:

      there is a trivial sense of ‘inference’, common in cognitive psychology, on which any information processing counts as inferential; this is not the sense of inference explored here


    • Interesting, Page 2, Highlight:

      any theory here must articulate criteria to distinguish inferences—drawn by agents on the basis of reasons —from processes in which psychological states non-mentally cause a behavioral outcome.


    • Interesting, Page 2, Highlight:

      A theory of inference must minimally solve this “demarcation problem” by identifying a shared character that inferences possess and non-inferential judgments lack.


    • Interesting, Page 3, Highlight:

      By denying inferential status to nonlinguistic judgments, the combination of Taking and CI stands in tension with trends in cognitive science, where researchers increasingly describe the judgments of nonlinguistic animals and infants in inferential terms


    • Interesting, Page 3, Highlight:

      they explain these judgments by appeal to the representational contents of those agents’ cognitive and conative states, especially their tracking of abstract, higher-order, spatial, causal, or psychological relations in their environment (Buckner, 2015). They thus seem to satisfy at least one core idea associated with Taking and CI: that some nonlinguistic judgments can be sensitive to the representational contents of the cognitive and conative states governing the agent’s actions.


    • Interesting, Page 3, Highlight:

      Mandelbaum’s strategy offers a powerful argument for the conclusion that many opaque judgments in adult humans involve logical form, but it is not the right instrument for the demarcation cut currently needed in nonlinguistic agents. The problem is that the argument relies on a dichotomy between the simplest forms of association on the one hand (which should i.e. be eradicated by counterconditioning and extinction) and logical structure on the other; but we already know that many of the candidates for inference in animals and infants do not fit neatly into either of these two categories.


    • Interesting, Page 4, Text:

      I’m a bit confused concerning how the author construes “association” – classical and operant conditioning are are, e.g., “associative” in only a very attenuated sense, since they don’t require postulating any representational states in the organism at all.


    • Interesting, Page 4, Highlight:

      Occasion-setting could notably explain failures of extinction or counterconditioning without appealing to logical structure, for it could allow subjects to treat the original training situation and counterconditioning/extinction situation as different contexts.


    • Interesting, Page 4, Highlight:

      if inferential “upgrading” strategies are to be applicable to animal judgments, they must move beyond old dichotomies and more clearly articulate when associative inferential connections become complex enough to merit inferential explanation, but without requiring that in doing so they must connect fully propositional contents.


    • Interesting, Page 5, Highlight:

      Bermudez’ approach—which is closest in methodological aims to the present inquiry—marks the boundary at the presence of “protological” structures involving nonlinguistic analogues of logical operators like the negation and conditional. Yet, I argue below that a great many flexible, representation-driven judgments that lack even protological structure should count as inferences, so this still places the lower boundary too high.


    • Interesting, Page 5, Highlight:

      A difficulty that such “intuitive” models of inference must overcome is that the way in which these decisions are intensionally sensitive is opaque to the agent’s introspection: the particular features assessed may not be directly available for reflection and scrutiny


    • Interesting, Page 8, Highlight:

      Configural learning is associative in the sense that the acquisition of the patterns depends upon statistical tracking of spatiotemporal contiguities between cues and that cue configurations lack explicit logical form, but is more advanced than basic associative learning in that multiple cue associations can occur in parallel and can be relational and context-sensitive in nature.


    • Interesting, Page 8, Highlight:

      the associative influence of any particular feature can depend upon the value of many other features in spatial, temporal, or conceptual context.


    • Interesting, Page 8, Highlight:

      This kind of configuration cannot be reduced to a mere concatenation of stimulus generalizations along individually conditioned stimulusresponse links.


    • Interesting, Page 10, Highlight:

      Advanced forms of associative learning enhance the practical agency of animals by modulating and guiding natural appetitive responses towards functional ends—rather than appearing to de-rationalize behavior through automatic linking of arbitrary stimuli to functionless responses, an attitude encouraged by an exclusive focus on inflexible elemental associations and artificial laboratory environments.


    • Interesting, Page 12, Highlight:

      While animals may often judge in ways consistent with classical logic and decision theory in limited domains—a mode of thinking that Brunswik (1955) dubbed “ratiomorphic”—they lack the mastery of syncategorematic and mathematical relations required to apply such rules in a domain-general way (Hurley, 2003; Watanabe & Huber, 2006). Moreover, it is unlikely that any animals are capable of explicit metarepresentational awareness of the contents of their mental states or the inferential connections between them.


    • Interesting, Page 12, Highlight:

      An important insight about inference emerges from the role played by Taking in these views, however: that intensionality is central to the distinction between rational inferences and non-rational forms of behavioral causation.


    • Interesting, Page 12, Highlight:

      For a rationalizing explanation of behavior to identify real causes of behavior, it cannot be phrased in terms of what the agent could or should have recognized. The agent must have actually performed the action because of those particular contents, as those contents are presented in its own psychology.


    • Interesting, Page 13, Highlight:

      It is worth remembering at this point that there are venerable reasons—related to Lewis Carroll’s story of Achilles and the tortoise (Carroll, 1895)—why we should hesitate to require that the agent’s understanding of the fit between reasons and action take the form of an explicit metarepresentational belief or further rule, at least in the base case of inference.


    • Interesting, Page 13, Highlight:

      The moral that many draw from this worry is that an inferential agent’s understanding of the fit between its reasons and action must ultimately bottom out in “a certain complicated disposition or competence or practical capacity” (Stroud, 1979), and a capacity for assessing configural fit might be just the right kind of disposition for this job.


    • Interesting, Page 14, Highlight:

      A distinction is commonly drawn in psychology between rule-based and similaritybased categorization (Juslin & Olsson, 2004).


    • Interesting, Page 14, Highlight:

      Our ability to apply reason-based explanation to such judgments will thus depend upon our ability to discern, for the task, which similarities count as relevant; but this is precisely where the agent’s individualized modes of presentation become paramount. Animals with different configural patterns for the same goal-categories (i.e. predator or prey) may categorize the same exemplar differently, because they will focus on different similarities.


    • Interesting, Page 14, Highlight:

      First, we should bring to bear our models of the creature’s attentional demands, perceptual abilities, working memory constraints, and representational capabilities


    • Interesting, Page 14, Highlight:

      it is not clear that the lioness need possess anything like a causal belief in this case, that approaching a giraffe in the grassland would cause it to be kicked, but approaching in the riverbed would cause it to obtain a meal. Minimally required for such a causal belief is some sense in which the animal distinguishes causation from mere correlation, and it is not clear that the lioness needs to draw any such distinction. Some animals appear sensitive to this difference (Blaisdell, Sawa, Leising, & Waldmann, 2006; Taylor, Miller, & Gray, 2012), and the addition of protological structure to configural judgments may bootstrap them to more advanced forms of reasoning; but these capacities go beyond the basic forms of configural pattern-matching I have discussed above.


    • Interesting, Page 15, Highlight:

      Second, we need to carefully scrutinize our normative analysis of the cognitive task the creature is performing


    • Interesting, Page 15, Highlight:

      Thirdly, we need to understand the informational structure of the environment in which the animal makes the decision to know what sort of similarities it will have had the opportunity to learn and will have reliably enough led to success.


    • Interesting, Page 15, Highlight:

      Since Goodman’s day, psychology and philosophy have developed a variety of empirically-supported, formal models of similarity, including Tversky’s influential feature-matching model, G€ardenfors new geometric model, and popular mathematical models of prototype and exemplar-based categorization (Ashby & Maddox, 1993; Decock & Douven, 2011; G€ardenfors, 2004; Tversky, 1977)


    • Interesting, Page 15, Highlight:

      a lingering worry is that similarity judgments are not assessable by norms, and so could not support rational inferences.


    • Interesting, Page 15, Highlight:

      One of Goodman’s concerns here is that similarity assessment varies too much by context to be subject to normative assessment


    • Interesting, Page 15, Highlight:

      it cannot be contextsensitivity alone that renders similarity assessment ineligible for normative evaluation


    • Interesting, Page 15, Highlight:

      A more targeted worry about context-sensitivity might be that similarity judgments could not be stably sensitive to the intensions of representations on which they act


    • Interesting, Page 15, Highlight:

      The final objection we might consider in this line is that such particularist judgments are not assessable by norms:


    • Interesting, Page 16, Highlight:

      Here, we need only note that there are ready and appropriate evaluative criteria, but they are ecological rather than rule-based in nature


    • Interesting, Page 16, Highlight:

      these judgments will be ecologically rational if the features used to assess target category membership are highly valid in that context, where ‘validity’ measures the conditional probability that an exemplar falls under the target category given the cues assessed (Gigerenzer, Todd, & ABC Research Group, 1999).


    • Interesting, Page 16, Highlight:

      heuristics researchers have demonstrated that classical approaches prizing coherence norms like truth-preservation or utility optimization will be routinely outcompeted by frugal, ecological strategies that are more economical in their collection of evidence and avoid the dangers of optimization, especially in common environments that are “non-compensatory” (in which a more valid cue cannot be outweighed by the less valid cues even if all the latter disagree with the former) or exhibit high noise that optimizing strategies overfit (Gigerenzer & Brighton, 2009; Gigerenzer et al., 1999).


    • Interesting, Page 17, Highlight:

      not all categorization heuristics studied by psychologists of intuitive judgment will satisfy the intensional interpretation of Taking offered here.


    • Interesting, Page 17, Highlight:

      Though Gigerenzer and colleagues have stridently defended the ecological rationality of this heuristic—given that recognition is a highly valid predictor of category membership in many environments and can be computed automatically and efficiently—it is not rational in the sense relevant to Taking. The reason is that the attribute assessed (simple recognition) is causally insensitive and semantically unrelated to the mode of presentation of the agent’s goal category; it simply does not matter what the subjects think of the target category, they will just choose the option they recognize.


    • Interesting, Page 17, Highlight:

      recent work on nonlinguistic inference demonstrates that configural judgment in many animals is under a form of executive control that is sensitive to an animal’s takes on its reasons for action.


    • Interesting, Page 17, Highlight:

      A common skeptical reaction to what has come thus far might be that these associative categorization judgments are perceptual rather than inferential in nature; that if subjects somehow just “see” the right answer, then these judgments cannot be inferences. This option would only be attractive to those who suppose perception and inference to be mutually exclusive, and thus is not for more expansive neo-empiricists who take both inference and reason to be perceptual in nature


    • Interesting, Page 17, Highlight:

      it may yet be objected that a key motivation for Taking has been not yet been satisfied: that rational decision procedures must allow agents to compare different options in an intensional way


    • Interesting, Page 18, Highlight:

      Some writers will still be dissatisfied with the permissive, intensional interpretation of Taking offered above, because they believe a more significant form of awareness is required if we are to be held epistemically responsible for our practical inferences.


    • Interesting, Page 18, Text:

      this is a nice admission – would have been better if author was more up front about it in the opening of the paper though


    • Interesting, Page 18, Highlight:

      First, realistic expectations: the present view cannot convince all skeptics of nonlinguistic inference, especially those who consider it akin to an analytic truth that genuine inference requires logical structure or an explicit, metarepresentational awareness of one’s reasons for acting. The best that could probably be said in response to such definitional skepticism is that I have here articulated an important natural kind of judgment—probably with the kind of homeostatic property-cluster structure that characterizes nearly any interesting psychological posit, whether modules, concepts, emotions, or cognition itself (Buckner, 2015; Fodor, 1983; Griffiths, 2004; Machery, 2005)—that merely shares many properties in common with “genuine” inferences.


    • Interesting, Page 18, Highlight:

      I thus grant that the properties highlighted by these objections—the abilities to inhibit, guide, and compare initial takes on a decision problem, components of judgment often studied in psychology under the heading of “executive control”—should be considered for inclusion in this inferential cluster. However, the empirical evidence strongly suggests that many nonlinguistic animal judgments do in fact possess these forms of executive control.


    • Interesting, Page 18, Highlight:

      While I argued above for a separation of internalist explanatory rationality and externalist justificatory rationality, these objections may highlight a lingering dissatisfaction with this division that is hard to put into words.


    • Interesting, Page 19, Highlight:

      additional agent-level awareness


    • Interesting, Page 19, Highlight:

      “epistemic feelings,” such as hunches and feelings of knowing, confidence, and uncertainty


    • Interesting, Page 19, Highlight:

      Crucially, these feelings are not experienced in the format of any sensory modality, and so it would be difficult to classify them as just a further aspect of sensory perceptions.


    • Interesting, Page 19, Highlight:

      by appealing to the animal uncertainty monitoring literature here, I am not disputing well-known arguments by Carruthers or Bermudez that language is required for truly metarepresentational metacognition; I instead aim to establish only that animals have internally-generated signals that are sensitive to their takes


    • Interesting, Page 20, Highlight:

      for a time the most popular skeptical response to these experiments was to hold that the bail-out option serves as a third response that can acquire a mediated reinforcement value throughout the course of the experiment.


    • Interesting, Page 21, Highlight:

      In light of the background findings, it is reasonable to conclude that these animals have detected a clash between two different takes on the same stimulus, and with cognitive effort can inhibit their initial take in favor of a more considered one.


    • Interesting, Page 21, Highlight:

      We have already specified the epistemic criteria that should be used to evaluate these judgments, which appeal to principles of ecological validity rather than internal coherence. This kind of reasoning has the epistemic goods when it is efficient, reliable, and operates on ecologically-valid cues, and lacks them when it does not. Note that as interpreted here, this is not a raw, case-by-case consequentialism; a thinker could get lucky every time with a highly-invalid procedure without their inferences counting as ecologically rational.


    • Interesting, Page 21, Highlight:

      is a tacit sensitivity to the strength of a take sufficient to support the idea that the animals could have chosen otherwise if they had reasoned better?


    • Interesting, Page 22, Highlight:

      I suggest that at issue is the assumption that if the specific features assessed in a configural take are opaque to us, then the take cannot grant us control over any judgments caused by it.


    • Interesting, Page 22, Highlight:

      if the kind of intuitive inference I have been articulating here is a natural kind, but executive control is not present in all animals capable of configural takes, then should the lower boundary of rational inference be placed at the advent of mere configural categorization, or rather at advent of executive control, which allows these takes to flexibly guide judgment?


  • Definition

    • Definition, Page 2, Highlight:

      process of arriving at a conclusion on the basis of reasons which support it.


  • Reference

    • Reference, Page 23, Highlight:

      I thus suggest that while we should remain agnostic between these two more precise locations for the lower boundary, we should conclude that it dips well into nonlinguistic territory, perhaps much deeper than even optimistic researchers recently supposed.


Icon by Nun from The Noun Project. Website built with Org-mode, Hugo, and Netlify.