Making Sense of Level-Relative Causation Alastair Wilson Monash - - PowerPoint PPT Presentation

Making Sense of Level-Relative Causation

Alastair Wilson Monash University & University College, Oxford alastair.wilson@monash.edu

Plan

• The causal exclusion problem.
• The level-relativization response.
• Worries about level-relativization by fiat.
• Locating the context-sensitivity in events.
• Level-relativity and downwards exclusion.
• Comparison with level-relativity of chance.
• The source of the level parameter.

A simple causal exclusion problem

• (1) Distinctness: Mental event types are distinct

from physical event types.

• (2) Completeness: Every event has a physical

event as a cause.

• (3) Efficacy: Mental events can be causes.
• (4) No overdetermination: The effects of mental

causes are not systematically overdetermined.

• (5) Exclusion: No event has (at a particular time t)

more than one cause unless it is overdetermined.

Level-relative causation

• If causation is level-relative, then ‘A caused X’

could be true relative to level 1, and ‘B caused X’ could be true relative to level 2.

• So an action could have a mental cause

relative to one level, and a physical cause relative to another level.

• The level is – somehow or other – supplied by

conversational context.

How level-relative causation helps

• Exclusion is false: ‘A caused X’ and ‘B caused X’

can both be true, even if A and B are distinct.

• But in its place we have:

– L-Exclusion: No event has (at a particular time t and at a particular level L) more than one cause unless it is overdetermined.

• L-Exclusion arguably captures the motivations

for Exclusion, and it is compatible with 1) - 4).

Level-relative causation by fiat

• The simplest way to allow for level-relative

causation is via a brute-force approach.

• When we assert ‘A caused B’ the level is filled

in by a primitive element of context, to go along with the time and world of utterance.

• Terence Horgan, inter alia, seems to endorse

this view.

• But by itself it looks much too cheap.

Problems with level-relativization by fiat

• Brute-force relativization strategies are

worryingly generalizable.

• There is no obvious semantic mechanism by

which a level is contextually supplied.

• There is no obvious account of how levels

themselves are individuated.

• Consequently, the strategy looks ad hoc.
• Can we do better?

Context-sensitive cause-ascriptions

• According to contextualist strategies, ‘the

cause of X’ is a context-sensitive expression.

• So both of these sentences can be true, in

different contexts:

– ‘My hunger was the cause of my eating of the banana.’ – ‘The microphysical state S of my brain was the cause of my eating of the banana.’

Where is the context-sensitivity located?

• Most extant versions of the level-relativity

strategy locate the context-sensitivity in ‘causes’.

• ‘Causes’ really means ‘causes according to

model X’, ‘causes X-ly’, or something similar, with parameter X supplied by context.

• I have a different suggestion – locate the

context-sensitivity in the event descriptions.

Individuating events

• How my eating of the banana is individuated

is a context-sensitive business.

• Identity criteria for events can contextually

vary in their fineness of grain.

• This leads to contextual variation in degree of

specificity for events.

• Highly specific events have few possible

realizers; highly unspecific events have many.

Causation and degrees of specificity

• The same-level requirement: cause and effect

should match in their degree of specificity.

• Examples:

– My hunger caused my eating of the banana. – My ‘hunger neurons’ firing caused my physiologically-specific eating of the banana. – My microphysical state S caused my microphysically-specific eating of the banana.

The modesty of level-relativity

• The sort of level-relativity just described

results from contextual variation in identity- criteria for events.

• For some, this will not be a very interesting

kind of level-relativity.

• But it does make the truth-value of causation-

ascriptions contextually variable, and seems to

• ffer a response to the exclusion problem.

The downwards exclusion response

• Menzies & List [2009], amongst others, argue

that Exclusion is compatible with Efficacy.

• This is because some instances of exclusion

are in the downwards direction: the mental cause excludes any physical cause.

• This occurs if the causal relation is realization-

insensitive: if the effect would still have

• ccurred were the cause differently realized.

How a mental cause can exclude a physical cause

• This counterfactual is true:

– Had I not been hungry, I would not have eaten the banana.

• But this counterfactual is not true:

– Had I not been in (micro-physically specific) state S, I would not have eaten the banana.

• Had I not been in S, I would still have been

hungry, so would still have eaten the banana.

A case of overdetermination?

• If Menzies and List are right, then Exclusion is

compatible with the causal-explanatory autonomy of the special sciences.

• Does this render level-relativity redundant?
• I will argue that the downwards exclusion

response and the level-relativity response are in fact complementary to one another.

Consequences of downwards exclusion

• The premise of the exclusion argument which

the downwards exclusion response rejects is:

– (2) Completeness: Every event has a physical cause.

• Rejecting Completeness is supposed to be

compatible with physicalism because supervenience on the physical is maintained.

• But it looks like a potential cost.

How to uphold Completeness

• Completeness: Every event has a physical

cause.

• M-Completeness: Every event has a physical

cause at the microphysical level.

• L-Completeness: Every event has a physical

cause at every level.

• The level-relativity response requires us only

to give up on L-completeness.

Downwards exclusion and level- relativity as complementary

• At each level of specificity, an effect has a

single cause (unless it is overdetermined).

• But which matching cause-effect pair is picked
• ut by a true sentence of the form ‘A caused

B’ is a context-sensitive matter.

• This combination allows us to preserve both

the causal closure of the physical and the causal autonomy of the mental.

What caused my eating of the banana?

The ‘exclusion problem’ for chance

• Chance-Exclusion: No event has (at a

particular time t) multiple distinct objective chances.

• Chance-Exclusion entails that if fundamental

physics were deterministic, special sciences couldn’t project non-trivial objective chances.

• But chances are ubiquitous in special sciences.
• This invites a level-relativization manoeuvre.

The Principal Principle

“Let C be any reasonable initial credence

• function. Let t be any time. Let x be any real

number in the unit interval. Let X be the proposition that the chance, at time t, of A’s holding equals x. Let E be any proposition compatible with X that is admissible at time t. Then C(A/XE) = x.” Lewis [1980]

Characterizing admissibility

“Admissibility: Propositions that are admissible with respect to outcome-specifying propositions Ai contain only the sort of information whose impact on reasonable credence about outcomes Ai, if any, comes entirely by way of impact on credence about the chances of those outcomes.” Hoefer [2007]

Lewis on admissibility

• Lewis [1980]: all historical information, plus

information about the laws, is admissible.

• This prevents probabilities in special-science

theories from counting as chances:

– Probabilities in classical statistical mechanics. – Probabilities in Bohmian quantum mechanics. – Probabilities of being dealt a specific hand in a poker game.

Relativizing admissibility

• Admissibility is relativized to degree of

fineness of grain of description:

– In statistical mechanics, information about the exact microstate of the system is inadmissible – In Bohmian quantum mechanics, information about the positions of corpuscles is inadmissible. – In poker, information about the order of cards in the deck is inadmissible.

Level-relative causes from chances?

• Causation might inherit a level parameter

from chance, if chance features (directly or indirectly) in the analysis of causation.

• Or the level parameter might derive from

resources featuring in the analysis of both causation and of chance.

• Either way, a unified account of level-relativity

for chance and causation would be appealing.

A level parameter from chances – directly?

• The most straightforward way to get level-

relative causation from level-relative chances is via a probabilistic analysis of causation.

• Appropriate probabilistic analyses of causation

might include those of:

– Kvart [2004] – Glynn [2010]

A level parameter from chances – indirectly?

• Various theories incorporate objective

chances into the semantics of counterfactuals.

• Given such theories, counterfactuals could

inherit a level parameter from chances.

• And given counterfactual difference-making

theories of causation, cause ascriptions in turn could inherit a level parameter from counterfactuals.

The source of the level parameter?

• Jonathan Schaffer, inter alia, suggests that the

parameter is a pair of contrast classes.

• Peter Menzies, inter alia, suggests that the

parameter is a model.

• Christopher Hitchcock, inter alia, suggests that

the parameter is a default state.

• Which of these options can supply a level

parameter both for chance and for causation?

A suggestion of Schaffer’s

“One further option to consider—drawing on Ludlow’s (2008) idea of the dynamic lexicon—would be to hold that event descriptions have contextually shifting

• interpretations. For instance, “the switch’s getting set

to local” might in some contexts rule in local, rule out broken, and leave express unresolved, while in another context it might rule in local, rule out express, and leave broken unresolved. This would in effect be a lexical account of contrastivity, but one that arises due to underdetermination of lexical meaning rather than due to ambiguity, and one that targets the event descriptions rather than targeting ‘causes’.” Schaffer [MS]

The causal reference class problem

• Hájek [2007] distinguishes metaphysical and

epistemological reference class problems for the theory of probability; a similar pair of problems apply to causation.

• The level-relativization manoeuvre resolves

the metaphysical problem.

• The epistemological problem remains

unresolved: the way in which context selects a level is not epistemically transparent.

Conclusions

• We can make sober sense of level-relativity of

causation via degree of specificity of events.

• This complements the downwards exclusion

response to the exclusion problem.

• The same move could potentially account for

level-relativity of objective chance.

• Which semantic mechanism contextually

selects a level remains an open question.

References

• Menzies & List *2009+. ‘Non-reductive physicalism and the limits of

the exclusion principle’, Journal of Philosophy 106(9).

• Glynn, L. *2010+. ‘A Probabilistic Analysis of Causation’, British

Journal for Philosophy of Science.

• Hájek, A *2007+. ‘The reference class problem is your problem too’,

Synthese 156.

• Hoefer, C. *2007+. ‘The third way on objective probability’. Mind.
• Kvart, I. *2004+. ‘Causation: Probabilistic and Counterfactual

Analyses’, in Collins, Hall, and Paul (eds.), Causation and Counterfactuals

• Lewis, D. *1980+. ‘A subjectivist’s guide to objective chance’.
• Ludlow, P. *2008+. ‘Cheap Contextualism’. Philosophical Issues 18.
• Schaffer, J. *MS+. ‘Causal Contextualisms’.