William Wimsatt, Re-Engineering Philosophy for Limited Beings
PHIL 484
Table of contents
1: Myths of LaPlacean Omniscience
- Academics who claim to seek the truth want to pretend that they have always had it
- Science is fallible and testable – but we hide embarrassing papers; we don’t report negative results; errors are only okay if theyr’e not mine
- We ought to celebrate a good research plan even if it failed
- Many contradictions emerge when we idealize science and endorsement of philosophical views which see knowledge only in certainty
- Focus on truth is understandable: the sciences are the best repositoreis of truth
- Results of science are important for our technology and public validation
- The way in which we talk and teach science misrepresents the importance of cognitive technologies, material procedures, and social structures
- Analytic training gives bad reactions – we don’t understand something until we have broken it down into the smallest possible pieces.
- Mistake: belief that reductionist or analytic methods eliminate / analye away what is being analyzed or reduced – creating an opposition at different levels of organization, between ‘humanistic’ and ‘scientific’ approaches.
- Can we appreciate processes at higher levels of organization without compromising them with reductionist methods?
- Sketched by Herbert Simon
- Realism – real problems
- “With this focus, scientific processes are seen as end-directed or teleological activities, generating a cognitive engineering of science, with natural resonance with design analyses in engineering – a fruitful path so far ignored in our profession.”
Realism for Limited Beings in a Rich, Messy World
- Methodological tools appropriate to well-adapted but limited and error-prone beings
- A philosopy of sciecne which can be pursued by real people in eral situations in real time with tools we actually have
- Oppose eliminativisms and overly idealized or rationalistic accounts.
- A realism account which does not seek stark simplicities; it is based from top to bottom on heuristic principles of reasoning and practice
- Ontology for the tropical rainforest
- A philosophy for “messy systems in the real world”
- “Piecewise approximations” – we aren’t God and don’t have a God’s-eye view of the world
- Local vs. global order
- Re-engineering: cumulative
- Our social, cognitive, and cultural ways of being are no less real than the rest of the natural world
- Natural scientists suppose all knowledge flows from their investigative enterprise
Social Natures
- Philosophers were once deaf to claims that cognitive science, evolutionary biology, etc. are central to philosophical methodology
- I want to re-psychologize, re-socialize, and re-embed us in the world, where we reason about that world as well as about how we interact with and reflect upon it. Can we still be recognizably philosophical while letting the subjects of “philosophies of” shine through much more clearly and inspire new philosophies, rather than merely exporting our same old “philosophical” disputes to these new territories?
- “Empirical contingencies” are crucial to philosophy
- We are embodied socialized beings
- We steer through the world with values which are uneasy combinations of history, religion, science, etc.
- The Hobbesian myth of the state of nature is a myth
- We were social throughout our primate history.
Heuristics as Adaptations for the Real World
- Philosophers have established ways of creating order working from a few key idealizations about decision making and rational or logical inference.
- “Generatively entrenched”
- Idealizations undemand unrealistic degrees of knowledge, unlimited inferential behavior – don’t fit with our behavior
- The very succes of our cognitive and social adaptations shows that giving up idealizations is not a disaster
- See our practice as strengths for cognitive adaptations rather than as compromised attempts to pursue our ideas. Deviations are not failings but the source of strenghts in an uncertain world.
- Consider our cognitive powers, limits, social contexts, embodiments
- More realistic accounts of activities are more flexible
Nature as Backwoods Mechanic and Used-Parts Dealer
- Logical empiricism built on minimalism, axiomatization
- Computational worldview builds on algorthms – adds from axiomatization an appreciation of problems in complexity and procedures
- Algorithmic view presupposes / constructs a well-defined structure and an “abstract crispness”
- Ancestors didn’t adapt with simpler and formal deduction systems; biology is full of inductive patterns
- Biological and cultural dimensions of our reason should be heuristic
- Heuristic principles are neither axioms nor algorithms; they have interesting properties as a group; they can be retuned, remodulated, recontextualized
- Heuristic principles are not just for the individual but for the community
- Heuristics vs. axioms and algorithms
- Nature as a “reconditioned parts dealer”, a “mechanic” – fixing and redesigning old machines
- What is the role of nature here? Are we trying to ‘study’ nature?
- Heuristics are not merely a subclass of all possible algorithms
- Heuristics are prior to algorithmic abstractions
- Our reasoning has a heuristic character
- Context-free elegance is itself a certain part of our history of reasoning
- Should we substitute heuristics for axioms and proceed before? No – we also need to change with our reasoning, change the scope, become more error-tolerant, treat earror differently.
- We have to become more capable of developing and responding to errors at different levels
- Theories should explain how structures are changed
- Generative entrenchment: heuristic and dynamical reading of foundational theories
- Non-foundationalism
- Heterogeneous, multi-level tropical rainforest
- Multi-perspectival realism in the heterogeneity of piecewise complementary approaches
2: Normative Idealizations versus the Metabolism of Error
Inadequacies of Our Normative Idealizations
- Scientists use often conflicting idealizations for various ends
- Different sets of false assumptions play crucial roles in teasing apart different parts of our world’s causal structure
- Models and idealizations increasingly replace theories as foci
- Scientists treat models as simplifications of or useful counterfactual transformations of nature; a dialectic; we don’t expect reality to chang eto fit the model
- Philosophers: we want human agents to change to fit models. Normative role of idealized models of rationality and inference. Scientist as a rational agent, logicla thinker, utility maximizer.
- These assumptions are often invisible, not treated as if they were part of the analysis.
- Models are kinds of abstract structures and assumptions for either descriptive or normative use.
- A model treated as normative is not usually compromised by our failure to act accordingly. We can choose to follow a normative rule or not.
- But… falsely describing human behavior is a weak test for normative status.
- Is it rational to be perfectly rational in a world of limited and fallible agents with imperfect, noisy, incomplete information? Herbert Simon: it is not; urge ‘satisficing’ rather than maximizing rational choice
- Naturalistic fallacy?
- The normative status of philosophical idealizations does not allow for adaptations to our cognitve biases
- “The shifty utilitarian” – you can retroactively justify all these behaviors as rational
- Is this an objection with rationality or an objection to rationality’s failure to be sufficiently rational? Meta-rational
- How do we measure the goodness of choices?
- Our world is too complex, and our abilities are too limited
- Philosophers build normative edifices on denial and don’t consider the possibility that they are false
Satisficing, Heuristics, and Possible Behavior for Real Agents
- Herbert Simon, satisficing: agents have levels of aspiration set and modified through experiene; alternatives are generated through time. Decisions are made according to heuristic rules
- We don’t need to compare or generate all choices
- Satisficing integrates learning
- No need to determine an exhaustive and mutually exclusive set of choices
- ‘Reasonable goal setting’ contributes to better performance.
- We need to recognize limitations on our powers of thought.
- “Human engineering”
- Reasoning idealizations enter surreptitously into our technology and cuase real problems
- “What is wrong is the design of the technology that requires humans to behave in machine-centered ways, ways ofr hwich people are not well-suited”
- “What kinds of machines are we supposed to be to run this technology?” – question of AI
- Idealizations of our cogntivie powers ignore not only our humanity but also our biology
- Look at nature’s designs
- Cultural tools for inference and problem solving
The Productive Use of Error-Prone Procedures
- Heursitic principles don’t guarantee results
- By error, we can misuse a truth-preserving rule of inference
- One can be a skeptic about any of our inferences
- Senses don’t give us infallible knowledge of the external world
- Philosophers get hung up on certainty; since Decartes, it is the best way to avoid errors.
- What if a search for reliable knowledge is not best pursued as a search for guarantees?
- Maybe errors are okay if they’re not too frequent, and we have good ways of remembering, detecting, and recovering from common ones.
- “More remarkable than our occasional failures is the fact that these common methods work so well as often as they do.”
- The most general maxim for those who study functionally organized systems is that we come to understand how things work by studying how, when, and where they break down.
- In the real world, knowledge does not come packaged with its axioms.
- Localizing and fixing the faults that occur. This works for the mind no less than for any of our other tools. With that understanding, we can analyze, calibrate, and debug both our reasons and our reasoning.
- We can’t idealize deviations and errors out of existence in our normative theories because they are central to our methodology – we are error-prone and error-tolerant
- We learn more when things break down than when they work write
- We metabolize mistakes
- We are particularly tuned to detecting and working with violated expectations, and more generally, with differences.
- Our normative models should reflect how we learn
- We need to avoid significant unbackable errors, not errors in general
- Positive view of error
- Models don’t lose their normative force by being descriptively accurate
3: Toward a Philosophy for Limited Beings
- An adequate philosophy of science should have normative force: it should help us do science or avoid sourcs of error
- The ‘special sciences’ can’t be too special
- Philosophers should be prepared to do field work, to speak the language
- Philosophers should be activists
- Be humbler about the advice we give to science; we are not ‘keepers of the logic of science’
- Like statisticians among mathematicians: philosophers should be theorists and therapists of reason
- We need a new organization of methodological knowledge around families of heuristics
Ceteris Paribus, Complexity, and Philosophical Method
- Studies of scientific methodology can be philosophically innovated
- In complex systems you shuold expect all interesting generalizations to be of the ceteris paribus kind
- Multiple realizability and multiple exceptions are inevitable
- We fail to appropriately correct for the incompleteness of our knowledge
- ‘Theory of practice’? – do we exchange ‘theory’ with ‘meta-theory’ too much?
- COgnitive omniscence
- Philosophers tend to presume that one could have true and interesting things to say about this in general, even in advance of actually seeing the theory. I don’t believe it!
- General meta-theories are particularly vulnerable to discovery of new ways to reframe the question.
- We should be more careful and modest about meta-theoretical claims
- I am by choice a conceptual engineer, not a pure theoretician
- We can get too enamored by tidy conceptualizations and claim lgoically necessary consequences
Our Present and Future Naturalistic Philosophical Methods
- We have used these tools well when they work well, but now we push them too widely.
- Too many problem areas are not well illuminated by our philosophical lampposts-at least where they are currently placed.
- Two topics in a new epistemology: heuristics and objects / boundaries
- Functional localization fallacy
- When a system can be described at a variety of levels of organization, or from a variety of perspectives, how do you recognize when a property is attributed to it at the wrong level, or from the wrong perspective? And how do inferences go wrong when this happens?
- A rain forest is a rich place after all, still far richer than we know.