Listen to this post



1. Introduction



Parts 3-4 of this series developed and defended a reason-responsive consequentialist theory of rational inquiry for bounded agents. The next order of business is to apply this theory to shed light on bounded rationality, the Standard Picture, and the epistemology of inquiry.



2. Bounded rationality revisited



Part 2 of this series developed five characteristic claims made by the bounded approach. We saw that a good theory of bounded rationality should help us to clarify, defend and apply these claims. Vindicatory epistemology was discussed in Part 4, so four claims remain. Let me discuss one by way of illustration.



Theories of bounded rationality claim that rational inquirers often rely on a toolbox of fast-and-frugal heuristics. Three traditional arguments are given for the rationality of heuristic cognition:



First, there is often an accuracy-effort tradeoff: the strategies that produce (in expectation) more accurate judgments tend to be (in expectation) more effortful. Heuristics often perform well on the accuracy-effort tradeoff, returning reasonably accurate judgments at low cost.



Second, agents have limited cognitive abilities, as a result of which they often cannot apply complex nonheuristic methods.



Third, sometimes less is more: in many environments heuristics outperform more complex methods, even ignoring cognitive costs, by avoiding overfitting.



A good theory of bounded rationality should ground all three of these traditional arguments, and the reason-responsive consequentialist view does this handily.



The accuracy-effort tradeoff matters because on a rich axiology, accuracy is a goodmaking feature of outcomes (perhaps intrinsically, and certainly for its consequences) and effort is a badmaking feature (at least because that effort could be expended elsewhere).



Limited abilities matter because ought implies can. On a reason-responsive approach, agents cannot be rationally required to do something unless they ought to do it, so it follows that rational requirements are things agents can do. Agents then cannot be rationally required to use nonheuristic methods when these methods exceed their abilities.



Less is more effects matter because, as we saw above, accuracy is a goodmaking feature of outcomes, and it certainly cannot hurt that less effort is required to get it.



3. The Standard Picture revisited



I said in Part 2 of this series that there are two reactions to Standard Picture violations. On the one hand, we can blame the agent, retaining the Standard Picture as a normative theory and taking the agent to be irrational. On the other hand, we can blame the theory, replacing the Standard Picture with a different normative theory which treats the agent’s thought and action as rational.



I said in Part 2 that bounded rationality would be an exercise in blaming the theory. But that is not quite right. While I did develop a theory of rational inquiry that does not draw on the Standard Picture, I did not argue for any normative claims about rational attitudes, such as the falsity of Standard Picture requirements.



This reaction charts a third way between blaming agents and blaming theories. I blame theorists rather than theories, for focusing too much on substantive facts about the irrationality of attitudes in place of procedural facts about the rationality of the processes that produced them. And I blame attitudes rather than agents, holding that many irrational attitudes could only have been avoided by irrational processes of inquiry,

Listen to this post



1. Introduction



Yesterday’s post developed a reason-responsive consequentialist theory of rational inquiry for bounded agents. Today’s post gives three arguments for that view.



2. The argument from minimal criteria



A good theory of bounded rationality should satisfy at least three minimal criteria.



First, it should be tradeoff-sensitive. Bounded agents have limited resources that must be traded off within and between inquiries and other activities. We cannot ignore tradeoffs, so a good theory should tell us how to make them.



Second, it should be stakes-sensitive. For example, it should say that ceteris paribus, it is better to think quickly about less important matters and slowly about more important matters.



Third, it should explain the irrationality of many instances of stereotyping, despite a highly uncomfortable relationship between stereotypes and seemingly rational heuristics. I won’t defend an agent who concludes that a woman in a boardroom is a secretary, though I might defend an agent who concludes that a tree in my home town is a pine.



I argue that the reason-responsive consequentialist view is our best hope for satisfying the minimal criteria. Consequentialists give precise and explanatory accounts of stakes and tradeoffs, and account for the irrationality of stereotyping by citing the magnitude of risked harms.



3. The explanatory argument



Consequentialists have a consequentializing program. We want to develop detailed, plausible, and explanatorily powerful accounts of normative data that other approaches struggle to accommodate. The explanatory argument for consequentialism takes these explanatory successes as evidence for a consequentialist approach.



In particular, I consider three data:



Norms of clutter avoidance say that agents are rationally required to avoid forming junk beliefs and to do what they can to purge existing junk beliefs from long-term memory. I show how my view recovers all three classic arguments for clutter avoidance made by Gilbert Harman: agents need to avoid (a) wasting effort, and (b) overwhelming limited storage and (c) retrieval capacities in memory.



Norms of friendship hold that agents are rationally required to show some partiality in thinking about friends. I argue that this is best treated as a datum about inquiry rather than belief, and show how a consequentialist approach gives a more moderate account of duties of friendship in inquiry than some competing approaches.



Norms of logical non-omniscience say that agents are rationally required to have some degree of logical omniscience, but not perfect logical omniscience. Most theorists have struggled to give precise and plausible accounts of the degree of logical omniscience that is required. I draw on existing discussions of bounded rationality in the context of chess, a game which is logically trivial but computationally intractable, to show how a consequentialist account can give precise and plausible constraints on the degree of logical omniscience that is required.



4. The vindicatory argument



We saw in Part 2 of this series that theories of bounded rationality are vindicatory, showing how many seeming irrationalities are nothing of the sort. A good theory of bounded rationality should be able to deliver vindicatory explanations that others cannot. I focus on two case studies.



The first is anchoring and adjustment. If I ask you to name the date of George Washington’s birth, you might anchor on a salient value (say,

Listen to this post



1. Introduction



Yesterday’s post developed five characteristic claims of the bounded rationality approach. This revealed the need for a theory of rational inquiry to defend, clarify and apply those claims.



Today’s post develops a reason-responsive consequentialist theory of rational inquiry for bounded agents.  This approach has three main components and three subsidiary components.



2. Three main commitments



The reason-responsive consequentialist view begins with (1) a consequentialist theory of rightness. On this view, acts are right just in case they are best. Acts are best just in case they promote the most value. Inquiry is an activity, and the fact that inquiry is often a mental activity should not change its rightness conditions. Inquiries are therefore right just in case they are best, and best just in case they promote the most value.



We can extend a consequentialist theory of rightness to a theory of rationality through (2) a reason-responsiveness theory of rationality. On this approach, rationality consists in doing what we have most reason to do in response to the reasons which favor it. But actions are right just in case we have most reason to perform them, so an equivalent way to state the reason-responsiveness view is that acts are rational just in case they are (a) right and (b) taken in response to the reasons which make them right.



On this view, acting rationally is strictly more demanding than acting rightly. The rightness condition (a) is supplied by a consequentialist theory of rightness. The responsiveness condition (b) requires that agents not act randomly or for bad reasons, but rather in response to the reasons which make their actions right.



From what perspective should rightness and reasons be assessed? Traditionally, two answers have been given: agents should do what actually promotes the most value (objective consequentialism) or what they believe or expect will promote the most value (subjective consequentialism). I argue that a recently-popular third option is more successful for bounded agents. (3) An information-sensitive account of deontic modals evaluates deontic modals, and probably also reasons, relative to a body of evidence. For example, we might say that acts are right just in case they promote, in expectation, the most value, where the probabilities used to compute expectations are derived from the agent’s total evidence. 



3. Three further commitments



For the reason-responsive consequentialist view to be plausible, we need three further commitments.



Consequentialists have traditionally urged (4) a strict level separation between normative questions about processes of deliberation and the attitudes they produce. To say that a process of inquiry is right, rational or virtuous is not to say that the attitudes it produces are right, rational or virtuous, and the same is true in the other direction.



Level separation allows us to defend a consequentialist theory of right and rational inquiry without thereby implying that right and rational belief are sensitive to the consequences of belief. Given level separation, the reason-responsive consequentialist view  is fully compatible with most leading theories of rational belief, including evidentialism.



Consequentialists are also not (all) hedonists, welfarists, or pragmatists. Consequentialists can adopt as rich and varied an axiology as anyone else. I adopt (5) a rich axiology on which many things, such as health, wealth, knowledge and understanding may bear final value. This allows a consequentialist view to give due consideration to traditional mainstays of epi...

Listen to this post



1. Introduction



What does it mean to theorize about bounded rationality? Today’s post situates theories of bounded rationality against a competing Standard Picture that came to prominence during the middle of the twentieth century.



2. The Standard Picture



On what Edward Stein (1996) has termed the Standard Picture of rationality, rationality is exhausted by requirements of consistency or coherence. For example, agents may be required to have logically consistent beliefs, probabilistically coherent credences, and preferences satisfying the Savage axioms.



The middle of the twentieth century was a profoundly optimistic period in the study of human rationality. A newly rigorized science of economics took Standard Picture requirements on preference to describe, in rough outline, how humans actually think and act. And although some theorists resisted normative construals of the Standard Picture, increasingly many theorists took Standard Picture axioms as normative requirements. For these theorists, rationality requires agents to satisfy the Standard Picture, and that is more or less what we do.



Mid-century optimism came under assault by a wealth of descriptive evidence suggesting that humans systematically deviate from Standard Picture requirements. For example, we neglect base rates, judge conjunctions to be more probable than conjuncts, and shift judgments and decisions across different framings of the same problem.



We can react to these findings in one of two ways. On the one hand, we can blame the agent. We can retain the Standard Picture as a normative theory and hold that agents who violate Standard Picture requirements are thinking or acting irrationally. Blaming the agent was a common reaction during the turbulent 1970s and 1980s, as optimism gave way to pessimism about human rationality.



On the other hand, we can blame the theory. We can use descriptive violations of the Standard Picture as a guide to identifying normatively relevant considerations which the Standard Picture does not model. In keeping with a wave of recent optimism about human rationality, I suggest that we should sometimes blame the theory.



To see where to start, note that the Standard Picture is architecturally neutral. It requires the same thing of toddlers, teenagers, thirty-somethings and toads. That is surprising because these agents have quite different cognitive architectures. No one would deny that an agent’s physical architecture affects the physical actions that it is rational for her to perform, for example by making some physical actions impossible and changing the effects that others will have. Theories of bounded rationality suggest that we should also take an agent’s cognitive architecture to affect the mental actions that it is rational for her to perform,  for example by making some mental actions impossible and changing the effects that others will have.



3. Bounded rationality



I suggested above that the right reaction to some Standard Picture violations is to blame the theory,  searching for normatively relevant bounds that the Standard Picture does not model. This is a task for theories of bounded rationality.



Traditional theories of bounded rationality make at least five characteristic claims.



First, bounds matter. Paradigmatic cognitive bounds such as limited abilities and the cost of exercising them bear on how it is rational for agents to cognize, just as they bear on the rationality of physical action.



Second,

Listen to this post



1. Introduction



This is the first in a five-part series introducing my book, Inquiry under bounds. The book is available here under an open access model from Oxford University Press.



2. My project



Humans are bounded agents. Some of our bounds are internal. For example, we have limited cognitive abilities and incur costs in exercising our abilities. Other bounds are external. For example, we find ourselves in environments that we did not choose, which structure the problems we face and the likely outcomes of cognitive strategies that we can apply to them.



Theories of bounded rationality ask what rationality requires of bounded agents, who are bounded in these and other ways.



Herbert Simon held that the fundamental turn in the study of bounded rationality is the turn from substantive to procedural rationality. Whereas theories of substantive rationality ask normative questions about the attitudes produced by inquiry, theories of procedural rationality ask normative questions about the processes of inquiry that produce them. Taking the procedural turn means spending less time asking normative questions about attitudes such as rational belief, credence, preference, and intention, and more time asking normative questions about the processes of inquiry that produce them.



If Simon is right, then a key part of the study of human rationality is a theory of rational inquiry for bounded agents. We need, that is, a theory of inquiry under bounds.



3. Outline of book and series



Inquiry under bounds aims to deliver that account. The book has four parts, which will be covered in the remaining four posts in this series.



Part 1, Rationality at the crossroads, situates the bounded rationality approach against a competing Standard Picture on which rationality is exhausted by requirements of coherence or consistency. I develop five characteristic claims of the bounded rationality approach and show why a theory of rational inquiry for bounded agents is needed to defend, clarify and apply them.



Part 2, Norms of inquiry, develops a novel account of rational inquiry for bounded agents: the reason-responsive consequentialist view. This approach combines a consequentialist theory of rightness with a reason-responsive theory of rationality and an information-sensitive reading of deontic modals.



Part 3, Justifying the account, gives three arguments for the theory developed in Part 2: the argument from minimal criteria, the explanatory argument, and the argument from vindicatory epistemology.



Part 4, Applying the account, uses the reason-responsive consequentialist view to clarify and defend the characteristic claims introduced in Part 1, revisit the relationship between bounded rationality and the Standard Picture, and draw lessons for the epistemology of inquiry. I also suggest that completing the procedural turn in the study of bounded rationality requires a second zetetic turn in practical philosophy, from the study of practical attitudes to the study of practical inquiry. 

There were some really good papers last month. The three I picked to summarize all involve error-based learning on fast time-scales. One involves the cerebellum in monkeys, the other involves the songbird system in…songbirds. One reason I like these examples is because they illustrate how deeply error-sensitivity is knitted into basic sensorimotor loops in non-human species. This is not really news in neuroscience, but superficially contradicts some philosophers who seem to be providing transcendental arguments that discursive linguistic practices are a necessary condition for the possibility of error.  Anyway, let’s check out the science.
Singin’ in the dopamine! Birdsong plasticity

Hisey, E, Kearney, MG, and Mooney, R (2018) A common neural circuit mechanism for internally guided and externally reinforced forms of motor learning. Nat Neurosci 21: 589-597. [Pubmed]
Xiao, L, Chattree, G, Oscos, FG, Cao, M, Wanat, MJ, and Roberts, TF (2018) A Basal Ganglia Circuit Sufficient to Guide Birdsong Learning. Neuron 98: 208-221. [Pubmed]
Here we have two elegant papers showing that birds can learn to adjust individual notes in their songs in response to brief pulses of dopamine. While many of us tend to think of the dopaminergic system as an extremely course, slow reinforcement signal, these papers suggest that it can act very quickly to reinforce specific actions embedded in complex behavioral contexts.
Note if you don’t like words, there is a nice video explaining the basic results posted at the bottom of this summary.
For decades, the birdsong system has been a workhorse for the study of socially acquired vocal behavior. This vocal learning occurs in two main stages: first, during the sensory learning phase, a fledgling male memorizes a tutor song from a conspecific adult male. Then, in the sensorimotor learning phase, he will slowly come to reproduce that tutor song himself. He will start by generating discordant, uncoordinated songs, and slowly shaping his vocalizations until the song closely resembles the tutor song. A typical song will have a rich internal structure like that in the figure.
Sensorimotor learning requires auditory feedback, as the animal shapes its behavior by comparing its its current song to the memorized tutor song (Mooney, 2009). Consider what it is like when you are singing a tune and need to hit B-flat. You can hear how close you are to the target. For instance, if you are coming in a bit sharp, you will try to push your pitch down a bit for that particular note.
Similarly, birds are extremely sensitive to errors in individual notes in their songs. If you add annoying white noise (WN) when one of their notes is below a certain pitch, songbirds adjust their pitch upward on that note in order to escape the noise. In the figure below, ‘Pitch-dependent auditory feedback’, you can see this in action. When note ‘d’ is above a certain pitch, that is an escape trial, and the bird is left alone. But when it is below that threshold, we have a ‘hit’ trial and WN is played. Over the course of two days (WN1 and WN2, compared to baseline day B1), the animal slowly increases the pitch of that individual note, to avoid the annoying WN pips.

Dopamine as an internal reinforcer for song plasticity
What is the underlying neural mechanism for such spectral plasticity? One nice thing about the birdsong system is that it has been relatively well mapped anatomically. For instance, there is a dopaminergic region, the ventral tegmental area (VTA), and neurons in this region are modulated by how close notes are to their target notes (Gada...