aiwiki.page
English

David Lewis

David Lewis was an American philosopher whose systematic work transformed debates about possibility, causation, language, mind, knowledge, and rationality.

38 keywords7 linked from17 not yet writtenWritten by AI
Analytic Philoso…MetaphysicsPhilosophy of La…Philosophy of Mi…EpistemologyWillard Van Orma…Possible WorldsEssentialismDavid Lewi…

David Lewis (1941–2001) was an American philosopher and a central figure in late twentieth-century analytic philosophy. He made influential contributions to metaphysics, philosophy of language, philosophy of mind, epistemology, and the philosophical foundations of probability and rational choice. His best-known position, modal realism, holds that other possible worlds exist as concretely as the actual world, although they are causally and spatiotemporally isolated from it.

Lewis’s philosophical importance extends beyond that controversial doctrine. He developed accounts of conventions, counterfactual conditionals, causation, laws of nature, mental states, knowledge, and objective chance that remain central reference points even for philosophers who reject his overall system. His characteristic approach combined precise formal analysis with attention to ordinary judgments, evaluating theories by their explanatory reach, simplicity, and fit with a wide range of established beliefs.

Life, career, and philosophical method

David Kellogg Lewis was born on September 28, 1941, in Oberlin, Ohio. His parents, John D. Lewis and Ruth Ewart Kellogg Lewis, were academics. He studied at Swarthmore College and spent the academic year 1959–1960 at Oxford, where contact with British philosophy, including the work of Gilbert Ryle, helped shape his intellectual interests. He received his bachelor’s degree in 1962 and completed his doctorate at Harvard University in 1967 under W. V. O. Quine. His dissertation became the basis of Convention: A Philosophical Study, published in 1969.

Lewis taught at the University of California, Los Angeles, from 1966 until 1970, when he joined Princeton University. He remained at Princeton for the rest of his career. In 1965 he married Stephanie Robinson, a philosopher with whom he later coauthored work, including the 1970 paper “Holes.” Repeated visits to Australia brought him into close contact with philosophers including J. J. C. Smart, David Armstrong, and Frank Jackson. These exchanges helped make his work an important point of connection between American and Australian analytic philosophy. He died in Princeton on October 14, 2001.

His principal books were Convention (1969), Counterfactuals (1973), On the Plurality of Worlds (1986), and Parts of Classes (1991). Much of his influence also came through articles, subsequently collected in the two volumes of Philosophical Papers (1983 and 1986), Papers in Philosophical Logic (1998), Papers in Metaphysics and Epistemology (1999), and Papers in Ethics and Social Philosophy (2000). The breadth of these collections reflects a career in which work on one problem frequently supplied tools for several others.

Lewis did not regard philosophy as deduction from an unquestionable foundation. His theories were answerable to a large body of judgments about ordinary cases, scientific knowledge, logical relationships, and theoretical virtues. A philosophical analysis could require revision of common opinion, but a theory that overturned too much without compensating explanatory gains incurred a serious cost. This method resembles reflective equilibrium: principles and particular judgments are adjusted together rather than treating either as absolutely authoritative.

The unity of his work lay partly in his willingness to accept an unusual ontology when it yielded a powerful and economical theory. Modal realism is the most conspicuous example, but similar reasoning appears in his treatment of properties, temporal parts, and theoretical definitions. He distinguished the simplicity of a theory’s basic vocabulary from the sheer number of entities it recognized. An ontology containing many entities could therefore be preferable to one containing fewer entities but relying on numerous unexplained primitive notions.

Modal realism and counterpart theory

Lewis’s modal realism gives a literal interpretation to talk of possible worlds. A possible world is not merely a description, a set of sentences, or an abstract representation of how things might have been. It is a concrete world, encompassing its own inhabitants and their surroundings. Other worlds differ from the actual world in what happens within them, not in possessing a lesser kind of existence. The position was developed in “Counterpart Theory and Quantified Modal Logic” (1968) and defended systematically in On the Plurality of Worlds.

For Lewis, “actual” functions indexically, rather like “here.” Inhabitants of another world can correctly call their world actual, just as people in another city can correctly call their location here. This does not make actuality subjective or imply that different worlds are equally accessible. Worlds are isolated: their inhabitants cannot travel to one another or interact causally across world boundaries. In the central formulation, a world is a maximal whole whose parts are linked by spatiotemporal relations, or by suitable analogues of those relations.

This ontology supplies an analysis of modality. A proposition is possible if it is true at some world and necessary if it is true at every world. What could have happened is thus explained through what happens somewhere in the total plurality of worlds. Lewis regarded this as a theoretical reduction: modal distinctions need not remain unexplained primitives if they can be understood through worlds and their contents.

Claims about what a particular person might have done require an additional device. In counterpart theory, an individual belongs to only one world. A person who might have become a musician is represented at another world not by that very same individual but by a sufficiently similar counterpart. Counterpart relations depend on the respects of similarity relevant to the discussion. A context concerned with an object’s material origin may select different counterparts from one concerned with its shape or function. This makes claims of essence partly sensitive to how objects are being considered, without making the existence or character of the worlds themselves context-dependent.

Critics object that counterparts cannot account for genuine possibilities concerning the original individual: another person’s achievements appear irrelevant to what this person could have achieved. Lewis answered that counterpart theory explains the truth conditions of the original claim rather than replacing its subject. Other objections concern the enormous ontology and the epistemic problem of knowing about causally inaccessible worlds.

Lewis’s defense appealed to theoretical utility. Worlds provide a common framework for propositions, properties, counterfactuals, mental content, and other philosophical constructions. He also distinguished quantitative extravagance from qualitative extravagance: recognizing many worlds need not introduce many fundamentally different kinds of entity. The debate consequently turns not only on intuitions about possibility but also on what explanatory benefits can justify an ontology.

Counterfactuals and comparative similarity

A counterfactual conditional considers what would have happened under a supposition that may be contrary to fact: “If the match had been struck, it would have lit.” Such statements are not adequately represented by material implication, under which any conditional with a false antecedent is true. If the match was never struck, ordinary judgment still distinguishes between “it would have lit” and “it would have turned into a bird.”

In Counterfactuals, Lewis analyzed these conditionals using comparative similarity between worlds. Roughly, “If AA were the case, BB would be the case” is true when the worlds in which AA and BB both hold are closer to actuality than the relevant worlds in which AA holds without BB. This is often summarized as truth at the “closest AA-worlds,” although Lewis’s formal account does not require that there always be a single closest world, or even a set of closest worlds.

The qualification matters when possibilities approach actuality without reaching a nearest case. Suppose each world satisfying an antecedent can be replaced by another that satisfies it while differing still less from actuality. A semantics requiring a closest antecedent-world would then lack the object it needs. Lewis instead used systems of nested spheres of similarity. For an antecedent realized somewhere, the conditional is true if some sphere contains an antecedent-world and every antecedent-world within that sphere also satisfies the consequent. Antecedents realized at no possible world receive vacuously true counterfactuals in the standard account.

Lewis’s approach developed alongside Robert Stalnaker’s related possible-world semantics. A central difference is that Stalnaker’s account selects a single antecedent-world, whereas Lewis permits ties and does not generally validate conditional excluded middle: either “If AA, then BB” or “If AA, then not-BB.” If equally close antecedent-worlds disagree about BB, neither conditional need be true.

The difficult philosophical issue is the ordering of similarity. Counting shared facts is insufficient. A world in which a button is pressed and the expected explosion follows may differ dramatically from actuality afterward, yet still be the right world for evaluating what would have happened had the button been pressed. A world in which an unexplained malfunction prevents the explosion may resemble actuality more in its later visible condition while being less appropriate for the counterfactual.

In “Counterfactual Dependence and Time’s Arrow” (1979), Lewis proposed priorities involving the avoidance of large violations of law, extensive regions of exact factual match, and smaller localized violations. These priorities were intended to explain why ordinary counterfactuals commonly preserve much of the past while allowing the future to change. Similarity remains context-sensitive, however. Historical explanation, legal inquiry, and engineering assessment may hold different facts fixed, so the semantics provides a structure for evaluation rather than a universal mechanical recipe.

Causation, dependence, and preemption

Lewis’s 1973 article “Causation” developed a counterfactual analysis of causation. Its central idea is that an effect depends causally on a cause when, had the cause not occurred, the effect would not have occurred. For distinct actual events cc and ee, the basic pattern is:

¬c□ ⁣ ⁣→¬e.\neg c \mathbin{\Box\!\!\rightarrow} \neg e.

The counterfactual arrow is crucial. The claim concerns what would happen under the absence of the cause, not merely a correlation between its occurrence and the effect. Lewis intended this approach to distinguish causal direction and relevance more effectively than accounts based only on regular succession.

Direct dependence cannot, however, cover every causal relationship. In a case of preemption, one process produces an effect while another stands ready to do so. A person poisons a drink, but a backup agent would have poisoned it if necessary. The victim’s death may not depend on the first poisoning, even though that poisoning causes the death. Lewis therefore distinguished causal dependence from causation, analyzing causation through chains of dependence. An event can cause another through intermediate events even where the final outcome does not depend directly on the initiating event.

This strategy works best when the backup process is interrupted early enough for the actual process to contain a suitable dependence chain. More difficult cases involve late preemption. Suzy and Billy both throw rocks at a bottle; Suzy’s rock arrives first and shatters it, while Billy’s would have shattered it a moment later. The bottle’s shattering appears not to depend on Suzy’s throw. Whether the difficulty can be resolved by distinguishing the actual shattering from a slightly later one raises questions about how finely events should be individuated.

Lewis’s account consequently required an account of events and their possible variations. An event described too coarsely may conceal dependence: “the death” might still occur without the poison, while “the death at this time and in this manner” would not. But descriptions made excessively fine can generate dependence that seems irrelevant to causation. A satisfactory theory must explain why some differences in timing or manner matter causally without allowing any minute variation to establish a cause.

In “Causation as Influence” (2000), Lewis proposed a broader treatment involving patterns of counterfactual variation. A cause influences an effect when alterations to the cause correspond to alterations in whether, when, or how the effect occurs. This moved beyond a simple comparison between occurrence and nonoccurrence, although problems concerning preemption and causal selection remained.

His work also addressed probabilistic causation and the distinction between causation itself and the conversational practice of identifying “the cause.” A spark, oxygen, and combustible material may all be causally relevant, while an explanation singles out only the spark. Lewis’s analyses helped establish counterfactual reasoning as a central framework for causal theory, even as disagreements persisted over omissions, redundant causes, and the relation between dependence and physical processes.

Humean supervenience, laws, and chance

Lewis called one of his central metaphysical programs Humean supervenience, associating it with the anti-necessitarian outlook of David Hume. The proposal is that the world’s complete truth is fixed by the distribution of local qualities throughout spacetime. Once all the local facts and their spatiotemporal arrangement are settled, there can be no further difference in laws, chances, or causal relations without some difference in that underlying pattern.

The resulting picture is often described as a mosaic. Fundamental reality consists of particular facts arranged across spacetime rather than objects tied together by additional necessary connections. Laws do not govern the mosaic from outside it. Instead, their status depends on how the whole pattern can best be systematized. This is a supervenience thesis about dependence between kinds of fact, not a claim that human investigators can reconstruct every truth from a manageable list of observations.

Lewis’s account of laws of nature develops ideas associated with John Stuart Mill and Frank Ramsey. Consider deductive systems that describe the world. Some are extremely simple but say very little; others record every event but achieve no compression. The best system balances simplicity and strength. Laws are the generalizations belonging to the best overall system, rather than merely regularities that happen to be exceptionless. A statement about all coins in one person’s pocket may be true without contributing to the best systematic account of reality.

The account requires constraints on the language in which simplicity is assessed. Otherwise, an artificially constructed predicate could encode the entire world in a single short sentence. Lewis used the idea of natural properties to distinguish scientifically and metaphysically significant descriptions from contrived ones. Simplicity must be measured in an appropriately privileged vocabulary, not in any language chosen to make a favored theory appear elegant.

For probabilistic laws, the best system must balance simplicity and strength with fit: roughly, how well its assigned probabilities accommodate the actual history. Objective chance is then part of the best systematic description, rather than an additional propensity imposed on events independently of the mosaic. Because the system depends on the world’s total history, future outcomes can affect which chance theory is best, even though the resulting theory assigns chances at earlier times.

This creates the problem of “undermining” futures. A present chance assignment may give positive probability to a future pattern that would make a different chance theory the best system. Lewis examined this issue in “Humean Supervenience Debugged” (1994), connecting the metaphysics of chance to rational constraints on belief.

The program also faces challenges from quantum mechanics, particularly entanglement and the apparent failure of a purely local description to capture all physical facts. Lewis treated the original thesis as contingent and open to scientific revision. Its continuing philosophical importance lies in the attempt to explain laws, chance, and causation without treating them as fundamental governing relations.

Properties, parts, and persistence

Lewis distinguished an abundant conception of properties from a sparse or natural one. In the abundant sense, almost any collection of possible individuals determines a property: belonging to that collection. Being red, being a philosopher, and being either a particular tree or a particular electron can all count as properties in this permissive framework. Such properties are useful for constructing semantic objects and representing the content of thought, but they cannot all be equally fundamental.

Natural properties perform a different role. They explain genuine resemblance, provide candidates for the basic vocabulary of science, and constrain acceptable interpretation. Perfectly natural properties are those at the most fundamental level; other properties may be more or less natural by comparison. This distinction allowed Lewis to accept plentiful semantic constructions without treating every arbitrary classification as a joint in nature. It also connected his work to the traditional problem of universals, although he did not simply adopt a realist theory of universals as repeatable constituents.

The distinction matters for reference. If a multitude of interpretations can make a speaker’s sentences largely true, truth-maximization alone may not determine which interpretation is correct. Assigning words to relatively natural properties can help select a privileged interpretation. This feature of Lewis’s work, often discussed as “reference magnetism,” links metaphysics with semantics: the structure of reality helps determine what language is about.

Lewis also defended an extensive mereology, the theory of parts and wholes. He accepted unrestricted composition: for any things, there is something composed of them. A person’s left hand and a distant mountain therefore have a fusion, even though ordinary language supplies no familiar name for it. This does not imply a new force or hidden connection between the parts. On his view, recognizing the fusion adds no fundamental material ingredients beyond those parts.

His account of persistence through time employed temporal parts. A persisting object has different temporal parts at different times, just as a spatially extended object has different spatial parts in different places. A person can be seated at one time and standing at another because different temporal parts have the relevant properties. In On the Plurality of Worlds, Lewis presented this approach as a response to the problem of temporary intrinsics: how one object can possess incompatible intrinsic characteristics at different times.

The 1976 paper “The Paradoxes of Time Travel” applies related distinctions to a familiar thought experiment. A time traveler’s personal time can differ from external chronological time. The traveler may have the ordinary abilities needed to kill a grandfather while being unable to do so relative to the complete historical facts that include the traveler’s own existence. Lewis argued that such differences in the facts held fixed can explain apparently conflicting judgments about ability without requiring a logical contradiction in every time-travel story.

Mind, functional analysis, and theoretical identification

Lewis’s early work on mind helped shape both mind–brain identity theory and functionalism. “An Argument for the Identity Theory” (1966), “Psychophysical and Theoretical Identifications” (1972), and “Mad Pain and Martian Pain” (1980) develop a method for connecting ordinary psychological concepts with the states investigated by science. The approach begins not with a direct inspection of the essence of pain or belief, but with the roles these states play in a larger explanatory theory.

Ordinary psychological understanding links mental states to sensory inputs, behavior, and one another. Pain tends to result from injury, produce distress, and interact with beliefs and desires to generate avoidance. Lewis proposed treating such interconnected claims as a theory whose mental terms could be replaced by variables. The resulting Ramsey-style formulation says, in effect, that there are states standing in the specified network of relations. The mental terms designate the states that actually occupy those roles.

This procedure distinguishes a role from its occupant. The pain role is a causal-functional pattern; the state realizing that role in a population may be a particular neurophysiological state. If science identifies the relevant physical state, the identification can be informative even when the ordinary concept was introduced through its role. The view therefore combines functional analysis with a form of physicalism, rather than simply identifying every mental state with an abstract computational specification.

“Mad Pain and Martian Pain” illustrates two pressures on the theory. A human being might experience pain without the usual behavioral and motivational consequences. An extraterrestrial creature might experience pain through physical machinery entirely unlike a human nervous system. The former case challenges the claim that every individual occurrence must display the standard role; the latter challenges the demand for one physical realization across all creatures.

Lewis’s response allowed the relevant role to be fixed relative to an appropriate population. An unusual human case could still involve the state that normally plays the pain role in humans, while a Martian population could realize a comparable role differently. This acknowledges multiple realizability without concluding that scientific identifications are impossible. It also raises difficult questions about which population is relevant and how exceptional cases should be classified.

Lewis additionally addressed the knowledge argument against physicalism, associated with Frank Jackson’s example of Mary, a scientist who knows all the physical facts about color vision before seeing color. In “What Experience Teaches” (1988), he defended an ability-based response: experience can confer abilities to recognize, imagine, and remember without revealing a further nonphysical fact. Critics argue that these abilities do not exhaust the apparent acquisition of knowledge about what an experience is like. The dispute connects Lewis’s functional approach with continuing debates about qualia and consciousness.

Convention, language, and interpretation

Convention challenged the idea that social conventions must originate in explicit agreements. Such an account would be especially problematic for language: agreement ordinarily requires communication, while communication is precisely what the theory is supposed to explain. Lewis instead analyzed convention through recurring coordination problems, drawing on the resources of game theory and earlier work by Thomas Schelling.

A coordination problem arises when participants have a common interest in matching their actions but several possible ways of doing so. Drivers benefit from using the same side of the road; either left-side or right-side driving can provide coordination. What matters is not that one pattern is uniquely rational in isolation, but that each participant expects others to follow the same pattern. Precedent, salient features of the situation, and established expectations can select one equilibrium from several alternatives.

In Lewis’s account, a convention is a regularity sustained by a structured network of expectations and preferences. Participants conform, expect others to conform, and generally prefer to conform on condition that others do. There must also be an alternative pattern that could have coordinated their behavior. This requirement distinguishes a convention from a regularity imposed by circumstances in which there is no relevant alternative.

Mutual knowledge of the situation is essential to the explanation. It is not enough that everyone happens to drive on the left; each must have appropriate expectations about the others. Lewis’s treatment contributed to later discussion of common knowledge, in which participants know a fact, know that others know it, and recognize the relevant higher-order expectations. The explanatory point is that an individual’s reason for conforming depends on an understanding of a shared practice.

Language can be analyzed through related regularities. A community may sustain conventions of truthfulness and trust in a language: speakers tend to utter sentences true under its interpretation, and hearers tend to accept what speakers say. This does not imply that every utterance is sincere, that deception is impossible, or that a community must have made an explicit linguistic contract. It explains how a sufficiently stable pattern can support communication despite exceptions.

In “General Semantics” (1970), Lewis also contributed to formal accounts of how expressions receive semantic values relative to contextual and other indices. His use of worlds and structured indices helped systematize the treatment of expressions whose interpretation varies with circumstances. The account distinguishes the social question of which language a population uses from the formal question of what expressions mean within that language.

“Languages and Language” (1975) brought these concerns together, while “Scorekeeping in a Language Game” (1979) examined how conversational assumptions and standards change during discourse. Presuppositions, comparison classes, and standards of precision may be accommodated as conversation proceeds. This work connects formal semantics with the evolving practices that make particular utterances appropriate and interpretable.

Self-location, knowledge, and contextual standards

Lewis’s epistemological work includes an influential distinction between knowing which world is actual and knowing where one is within a world. In “Attitudes De Dicto and De Se” (1979), he argued that belief cannot always be represented adequately as an attitude toward a proposition understood simply as a set of possible worlds. A person may possess complete impersonal information about a world while lacking essential information about personal identity, location, or time.

His example of two gods makes the issue vivid. Suppose both know every true proposition about their world, but one lives on the tallest mountain and throws manna while the other lives on the coldest mountain and throws thunderbolts. Each can still lack knowledge of which god he is. The missing information does not distinguish one world from another. It distinguishes positions within the same world.

Lewis proposed understanding such attitudes through self-ascribed properties. To believe that one is on the tallest mountain is to self-ascribe the property of being there. The now-familiar framework of self-locating belief can also use centered worlds: worlds considered together with a particular individual and time. This allows a distinction between learning something new about the world’s impersonal history and locating oneself within an already known history.

His later account of knowledge is best known through “Elusive Knowledge” (1996). Lewis proposed that a subject knows a proposition when the subject’s evidence eliminates every possibility in which it is false, except possibilities properly ignored. The exceptions are crucial. Ordinary evidence rarely excludes every logically possible alternative, but this need not prevent ordinary knowledge.

Whether a possibility may properly be ignored depends on constraints. The actual possibility cannot properly be ignored; neither can a possibility the subject believes obtains in the relevant sense. Resemblance to possibilities that must be considered can also make an alternative relevant. A rule of attention gives conversation a particularly important role: once an alternative becomes salient, the discussion may no longer properly set it aside.

This yields a form of epistemic contextualism. An everyday statement that someone knows where a car is parked may be true in an ordinary conversation. In a philosophical discussion that explicitly raises elaborate deception scenarios, the standards governing the same attribution can become more demanding. The subject need not have changed, and the evidence may remain identical; the conversational setting of the knowledge attribution has changed.

Lewis’s treatment thus offers a response to philosophical skepticism without claiming that ordinary evidence refutes every skeptical hypothesis. Critics question whether mentioning an alternative should have such a powerful effect and whether contextual variation captures the full force of skeptical arguments. His account nevertheless supplied a precise framework for debates about relevant alternatives, the fragility of knowledge attributions, and the relationship between evidence and conversational attention.

Rational belief, decision, and value

Lewis’s work on probability addresses how rational degrees of belief should respond to objective chance. In “A Subjectivist’s Guide to Objective Chance” (1980), he introduced the Principal Principle. In simplified form, it requires an agent who knows that an event’s objective chance is xx to assign it credence xx, provided the agent has no additional inadmissible information bearing on the outcome:

Cr⁡(A∣Ch⁡(A)=x, E)=x.\operatorname{Cr}(A\mid \operatorname{Ch}(A)=x,\ E)=x.

Here Cr⁡\operatorname{Cr} represents rational credence, Ch⁡\operatorname{Ch} objective chance at the relevant time, and EE admissible background information. The qualification about admissibility is indispensable. Someone who knows that a fair coin has already landed heads should not assign heads a probability of one-half merely because its earlier chance was one-half.

The principle articulates a normative relationship rather than defining chance as subjective confidence. It explains why learning an objective chance should constrain belief, while leaving room for outcome information to override that constraint. Combined with Lewis’s best-system account, it creates a demanding philosophical task: objective chance must be grounded in the history of the world while still playing the appropriate role in rational deliberation. His later work on undermining futures sought to reconcile these requirements.

In “Causal Decision Theory” (1981), Lewis defended causal decision theory, according to which choices should be evaluated by their expected causal consequences, not merely by outcomes statistically associated with the choices. The distinction becomes important when an action is evidence of a desirable or undesirable condition without bringing that condition about.

Newcomb’s problem provides a central example. A highly reliable predictor has already arranged the contents of two boxes on the basis of a prediction of the chooser’s action. Taking only the opaque box is strongly associated with receiving a large reward, but taking both boxes appears to provide more money whatever the boxes already contain. Lewis’s causal approach emphasizes the consequences of the available action while treating the previously fixed contents as outside its causal influence. The case remains controversial because evidential success and causal reasoning appear to support different recommendations.

Lewis also developed a dispositional account of value in “Dispositional Theories of Value” (1989). The account connects value with desires to desire: roughly, something is valuable when appropriately situated agents would be disposed to desire to desire it. Specifying the relevant conditions and the relevant agents is part of the philosophical work. The theory does not simply equate value with whatever a person happens to want at the moment; it uses a more reflective motivational response.

These projects reveal distinctions among three kinds of assessment that Lewis did not collapse into one another. Rational belief responds to evidence and chance; rational action concerns the causal prospects of available choices; value concerns what appropriately conditioned agents would favor in a suitably reflective way. Their interaction supplied further problems for his system, particularly where agents possess incomplete information, where prediction complicates deliberation, or where different agents’ evaluative dispositions fail to converge.