A  field  guide  to  mechanisms:  part  I       Holly  Andersen   Simon  Fraser  University   [email protected]       Abstract:  In  this  field  guide,  I  distinguish  five  separate  senses  with  which  the  term   ‘mechanism’  is  used  in  contemporary  philosophy  of  science.  Many  of  these  senses   have  overlapping  areas  of  application  but  involve  distinct  philosophical  claims  and   characterize  the  target  mechanisms  in  relevantly  different  ways.  This  field  guide  will   clarify  the  key  features  of  each  sense  and  introduce  some  main  debates,   distinguishing  those  that  transpire  within  a  given  sense  from  those  that  are  best   understood  as  concerning  distinct  senses.  The  ‘new  mechanisms’  sense  is  at  the   center  of  most  of  these  contemporary  debates,  and  will  be  treated  at  greater  length;   subsequent  senses  of  mechanism  will  be  primarily  distinguished  from  this  one.  In   part  I  of  this  paper,  I  distinguish  two  senses  of  the  term  ‘mechanism’,  both  of  which   are  explicitly  hierarchical  and  nested  in  character,  such  that  any  given  mechanism  is   comprised  of  smaller  sub-­‐mechanisms,  in  turn  comprised  of  yet  smaller  sub-­‐sub-­‐ mechanisms,  and  so  on.  While  both  of  the  senses  discussed  here  are  anti-­‐reductive,   they  differ  in  their  focus  on  scientific  practice  versus  metaphysics,  in  the  degree  of   regularity  they  attribute  to  mechanisms,  and  in  terms  of  their  relationships  to  the   discussions  of  mechanisms  in  the  history  of  philosophy  and  science.       Keywords:  mechanisms;  causation;  explanation;  reduction;  methodology         0)  Introduction       Talk  of  mechanisms  is  central  to  a  variety  of  recent  debates  in  philosophy  of  science.   For  those  working  outside  of  these  debates,  it  can  be  difficult  to  get  a  clear  sense  of   what  all  the  talk  of  ‘new  mechanisms’  is  about.  This  two-­‐part  field  guide  is  intended   to  provide  an  overview  of  the  main  new  sense  of  mechanism  that  has  been  the   primary  focus  of  discussion,  and  to  distinguish  this  new  sense  from  other  related   but  relevantly  different  senses  in  which  the  term  mechanism  appears  in  philosophy   of  science.  This  field  guide  begins  by  a  concise  overview  of  the  literature  on  ‘new   mechanisms’,  and  then  turns  to  distinguishing  and  comparing  other  notions  of   mechanism  to  this  one.     1 This  field  guide  offers  taxonomic  classifications  of  species  of  mechanism,  with  key   distinguishing  markings  of  each  species.  These  taxonomic  distinctions  have  not  been   utilized  in  debate  involving  ‘mechanisms’  but  are  motivated  out  of  the  kinds  of   characterizations  and  debates  that  have  multiplied  on  this  topic  recently.   In  elaborating  each  distinct  sense,  I  am  thereby  arguing  that  each  involves  a  unique   set  of  ontological,  methodological,  explanatory,  and  anti-­‐/reductive  commitments.   Claims  that  are  true  of  one  sense  will  fail  of  others;  as  such,  the  senses  are  not   interchangeable,  even  though  they  are  often  presumed  to  be  (especially  mechanism   1 and  mechanism ).     2   The  division  between  parts  I  and  II  classifies  types  of  mechanisms  in  terms  of   whether  or  not  they  are  explicitly  anti-­‐reductive.  Here  in  part  I,  I  focus  on  two   senses  that  include  an  explicitly  tiered  ontology  as  an  alternative  to  reduction.  These   approaches  to  mechanisms  concern  ways  to  individuate  and  connect  levels,   especially  but  not  limited  to  levels  as  they  are  studied  in  the  sciences,  while   rejecting  that  this  constitutes  a  reduction  of  higher  levels  to  lower  levels.  This  anti-­‐ reductive  approach  to  mechanisms  aims  to  provide  explanatory  autonomy  to  the  so-­‐ called  special  sciences,  while  still  grounding  their  objects  of  investigation  firmly  in   the  physical  world.  These  two  senses  can  be  distinguished  by  differing  in  3  major   characteristics:  the  first  emphasizes  the  role  of  mechanisms  in  various  aspects  of   scientific  practice,  while  the  second  emphasizes  a  metaphysical  role  for   mechanisms;  the  first  attributes  some  degree  of  regularity  to  mechanisms,  while  the   second  allows  any  unique  causal  chain  to  count  as  a  mechanism;  and  while  the  first   offers  something  novel  to  the  ongoing  philosophical  discussion  of  mechanisms,  the   second  is  more  closely  aligned  with  the  historical  notion  of  mechanism.  These   characteristics  are  mutually  supportive;  for  instance,  an  emphasis  on  scientific   practice  leads  clearly  to  a  focus  on  regularities  as  targets  for  experimentation,  while   an  emphasis  on  metaphysics  connects  much  more  closely  to  the  historical   discussion  of  mechanisms  in  natural  philosophy.       2 Each  section  of  this  field  guide  will  begin  by  briefly  characterizing  the  most  salient   philosophical  features  of  the  relevant  species  of  mechanism,  including  the  kind  of   discussion  in  which  it  most  frequently  appears.  I  will  then  briefly  discuss  the   ontological  commitments  of  each  sense,  their  methodological  implications,  and  their   role  in  explanation,  respectively.  Each  section  includes  a  comparison  distinguishing   it  from  other  related  senses  of  mechanism.         1) Mechanism :  Mechanisms  as  integral  to  scientific  practice   1   This  is  the  ‘new  mechanism’  approach  that  is  arguably  the  primary  sense  of  the  term   mechanism  in  recent  discussion.  This  species  of  mechanism  is  very  commonly  found   in  discussions  of  explanation  in  the  sciences,  as  distinct,  for  instance,  from   explanation  in  history  or  of  action.  It  comprises  several  claims.  One  is  the   ontological  claim  that  the  phenomena  studied  by  many  sciences,  especially  so-­‐called   higher-­‐level  sciences  like  biology,  have  a  particular  hierarchical  structure  comprised   of  nested  levels.  Each  level  involves  entities  organized  in  various  ways  and   connected  via  causal  interactions  or  activities,  constituting  or  giving  rise  to  a  pattern   that  recurs  under  specific  identifiable  circumstances.  These  patterns  are  the   regularities  that  constitute  the  phenomena  studied  by  these  sciences.  Such   mechanisms  provide  non-­‐arbitrary  ways  to  divide  parts  of  the  world  into  levels  of   physical  size  and  organization,  and  to  understand  the  connections  between  different   levels,  without  having  to  reduce  or  eliminate  higher  levels  or  treat  them  as  merely   apparent  or  epiphenomenal.       A  second  claim  is  that  the  methodology  of  many  sciences  reflects  this  ontological   structure  of  their  target  phenomena,  by  isolating  consistent  patterns  or  regularities   in  the  world  and  then  decomposing  them  into  constituent  entities,  causal   connections,  and  spatio-­‐temporal  organization.  The  emphasis  is  on  the  practice  of   science.  A  third  claim,  central  to  this  sense,  concerns  explanation  in  such  sciences,     3 namely,  that  the  explanations  they  offer  reflect  this  ontology.  Explanations  are  of   regularly  recurring  phenomena  in  the  physical  world  that  are  the  end  product  of,  or   are  constituted  by,  the  operation  of  such  mechanisms;  such  regularities  are  to  be   explained  by  providing  details  about  the  mechanism(s)  responsible  for  producing   them.       The  trajectory  leading  up  to  this  sense  of  mechanism  is  familiar.  As  logical   empiricism  waned  in  the  middle  of  the  20th  century,  philosophers  of  science  focused   on  deduction  from  universal  laws  as  the  primary  form  of  explanation  in  science  (e.g.   Hempel  and  Oppenheim  1948,  Hempel  1963).  On  this  view,  an  explanation  involves   a  general  law  plus  a  set  of  initial  conditions,  from  which  the  explanandum  could  be   deduced.1  A  variety  of  problems  plagued  the  deductive-­‐nomological  account  of   explanation  (see,  for  instance,  Salmon  1978).  One  clear  issue  was  the  apparent  lack   of  suitable  laws  in  biology  to  serve  in  a  deduction.  This  posed  an  unfortunate   dilemma:  either  biology  and  related  sciences  do  not  really  explain,  or  the  D-­‐N   account  applies  only  narrowly  to  some  explanations,  mainly  in  physics.  Laws  can  be   reconstrued  in  ways  that  render  them  more  applicable  in  biological  sciences   (Mitchell  1997),  but  many  still  found  the  D-­‐N  model  of  explanation  unsatisfactory  in   biological  contexts.     This  ‘new  mechanism’  approach  arose  in  response  to  many  of  the  issues  raised  with   the  D-­‐N  model  of  explanation.  It  differed  from  the  D-­‐N  account  in  terms  of  what  it   took  the  sciences  to  be  investigating,  how  those  investigations  generally  proceed,   and  the  explanatory  results  of  those  investigations.  Ontologically,  it  shifted  the   emphasis  away  from  universal  laws  and  nomic  necessity,  towards  recurrent   patterns  of  causal  structure,  with  a  much  more  limited  scope  of  generalizability.   Methodologically,  it  took  the  sciences  to  be  engaged  in  finding  more  and  more  detail,   by  targeting  and  decomposing  the  mechanisms  that  support  limited  regularities,                                                                                                                   1  On  this  view,  the  deductive  arguments  that  rely  on  universal  laws  to  show  the   necessity  of  the  explanandum  also  have  an  inductive  counterpart,  relying  on   statistical  generalizations  that  raise  the  probability  of  the  explanandum.     4 rather  than  seeking  ever  broader  and  more  abstract  generalizations.  These   ontological  and  methodological  views  support  an  account  of  explanation  that   eschews  propositionally  structured  laws  and  logically  valid  arguments,  instead   relying  on  mechanism  models,  of  varying  levels  of  descriptive  specificity,  involving   real  entities,  their  activities,  and  their  organization  into  coherent  processes  that   recur  under  appropriate  circumstances.       Finally,  this  new  sense  of  mechanism  is  deeply  anti-­‐reductionist:  science  may   uncover  explanatory  or  ontological  connections  between  higher  and  lower  levels,   but  does  not  thereby  either  eliminate  or  reduce  the  higher  levels  thus  connected   (Wimsatt  1994).  The  hierarchically  structured,  anti-­‐reductionist  approach  involved   in  mechanisms  holds  that  science  provides  genuine  explanations  that  “bottom  out”   1 without  having  to  be  grounded  in  some  ultimate  fundamental  level  (Machamer,   Darden,  and  Craver  2000).  Higher-­‐level  explanations  are  not  viewed  as  placeholders   for  some  ‘real’,  microphysical,  explanation;  the  focus  is  instead  on  integrating  and   connecting  levels  (Glennan  1995,  2002,  2005;  Darden  and  Craver  2009).  Even  if   mechanisms  in  biology  are  ultimately  composed  of  sub-­‐sub-­‐submechanisms  in   physics,  this  does  not  meant  that  biological  phenomena  explained  by  such   mechanisms  are  thereby  reduced  to  the  lower-­‐level  mechanisms.       This  anti-­‐reductive  stance  has  ontological  consequences.  Methodologies  for   investigating  levels  of  recurrent  mechanisms  can  proceed  in  relative  autonomy,   neither  affirming  nor  denying  any  ultimate  ontological  relationship  between   different  branches  of  science,  scientific  theories,  or  phenomena.  Mechanism  thus   1 imposes  some  distance  between,  on  the  one  hand,  normative  epistemological  or   methodological  claims  about  the  most  effective  ways  to  investigate  the  world,  and,   on  the  other  hand,  the  underlying  nature  of  that  world.  It  is  not  explicitly  anti-­‐ metaphysical,  but  rather  metaphysically  agnostic.  The  anti-­‐reductive  character  of   mechanism allows  us  to  make  methodological  recommendations  about   1   investigating  the  world  (see  especially  Bechtel  and  Richardson  1992  and  Darden     5 2002)  without  thereby  committing  ourselves  to  a  single  account  of  what  that  world   is  like.       Mechanism  is  often  elaborated  as  part  of  a  detailed  look  at  the  actual  investigatory   1 practices  of  the  sciences,  especially  biological  sciences.  Bechtel  and  Richardson   (1993;  see  also  Bechtel  and  Abrahamsen  2005)  characterize  scientific  discovery  in   the  biological  sciences  in  terms  of  localizing  and  decomposing  mechanisms.  Instead   of  searching  for  universal  laws,  many  scientists  are  engaged  in  the  task  of   identifying  regular  phenomena  and  the  precise  circumstances  under  which  they   transpire,  and  the  locus  of  control  by  which  mechanisms  can  be  manipulated.  Some   target  regularity  must  first  be  delineated  –  what  is  the  phenomenon  to  be   explained?  This  often  proceeds  by  investigating  the  precise  spatio-­‐temporal   boundaries  of  the  regularity,  including  parameter  values  that  modulate  how  it   occurs  and  supporting  environmental  factors  outside  of  the  mechanism  itself.   Localized  mechanisms  can  be  decomposed  into  constituent  elements  or   components,  their  interactions,  and  the  organization  of  those  elements  into  a   coherent  and  re-­‐identifiable  process.  Once  localized  and  decomposed,  the   delineated  mechanism  explains  why  the  identified  regularity  occurs  as  it  does,   under  the  conditions  that  it  does.       The  way  in  which  mechanisms  are  investigated  shapes  the  kinds  of  explanations  in   1 which  those  mechanisms  figure.  Explanations  can  involve  a  greater  or  lesser  degree   of  specificity,  depending  on  the  state  of  scientific  knowledge  or  the  context  in  which   a  given  explanation  is  provided.  For  example,  Craver  and  Darden  (2001),  Darden   (2002),  Craver  (2007),  and  Darden  (2008)  offer  mechanism  sketches,  mechanism   schemata,  and  fully  developed  mechanisms  as  identifiable  stages  involved  in  the   discovery  of  mechanisms.       Mechanism  schemata  are  abstract  descriptions  of  mechanisms  that  can  be   instantiated  to  yield  descriptions  of  particular  mechanisms.  The  term   mechanism  schemata  is  fitting  because  their  components  are  placeholders  that     6 can  be  filled  in  with  detailed  stages  between  the  setup  and  termination.  …   Mechanism  sketches,  in  contrast  to  mechanism  schemata,  are  abstract   descriptions  of  mechanisms  that  cannot  yet  be  filled  in  (Craver  and  Darden   2001,  120).       Darden  calls  these  distinctions  “advisory”:  neither  simply  descriptive,  nor  purely   prescriptive,  they  provide  a  set  of  terms  and  concepts  for  scientists  that  can  be   genuinely  useful  in  structuring  research,  while  recognizing  that  these  terms  and   concepts  have  been  developed  by  considering  what  many  scientists  are  already   doing.       Schemata  and  sketches  vary  in  terms  of  the  amount  of  detail  they  fill  in  about  the   mechanism(s)  in  questions  and  the  amount  of  possibility  they  leave  open  for  how   that  mechanism  might  be  instantiated.  The  idea  is  that  a  mechanism  sketch  places   constraints  on  the  kind  of  mechanism  that  could  instantiate  it.  The  first  stage  of   discovery  of  a  mechanism  may  consist  in  such  a  sketch  that  guides  further  research   by  narrowing  the  range  of  possible  kinds  of  mechanisms  to  be  investigated.  A  fuller   set  of  constraints  may  be  provided  by  mechanism  schemata,  where  components  and   their  connecting  activities  have  sufficient  constraints  placed  on  them  that   researchers  can  begin  filling  in  physical  details  for  particular  mechanisms.  A  single   schema  may  be  instantiated  by  several  distinct  mechanisms  that  fill  in  the   placeholder  components  with  different  entities  or  activities  that  fulfill  broadly   similar  functions.     This  highlights  the  need  to  distinguish  between  mechanisms ,  which  are  in  the   1 world,  and  models  of  mechanisms  (e.g.  Glennan  2005,  Illari  and  Williamson  2010).   1 This  distinction  figures  in  the  debate  of  what  it  is  in  virtue  of  which  mechanisms   1 are  explanatory.  Drawing  from  Salmon  (1984),  the  ontic  conception  of  mechanisms   holds  that  mechanisms  are  represented  in  models  and  theories  in  the  sciences,  but   are  not  identical  with  those  representational  devices.  They  are  instead  ‘out  there’  in   the  world.  This  makes  an  important  difference  when  considered  in  terms  of     7 explanation:  do  semantic  or  syntactic  relationships  between  elements  of  a  model-­‐ theoretic  representation  do  the  explanatory  work  in  their  own  right,  or  do  they   merely  point  to  or  exhibit  the  causal  relationships  in  the  world  that  do  the   explanatory  work?  Wright  (2012)  argues  against  this  ontic  conception,  holding   instead  that  explanation  is  a  communicative  process  for  the  purpose  of   understanding  which  must  thus  involve  models  of  mechanisms.  Many  models  of   mechanisms  are  not  propositionally  structured,  but  are  instead  conveyed  using   1 diagrammatic  or  visual  forms  of  representation  (Darden  and  Craver  2002,  Perini   2005a,  2005b).  This  debate  concerns  the  role  that  mechanism  plays  in  explanation:   1 do  mechanism  explanations  involve  physical  mechanisms  themselves,  or  models  of   those  mechanisms?  It  has  consequences  that  reach  beyond  the  discussion  simply  of   mecahnisms ,  however,  by  undermining  general  accounts  of  explanation  that  rely   1 on,  for  instance,  logical  entailment  relationships  between  premises  in  arguments  to   characterize  explanation  (Strevens  2008).  If  mechanism  models  figure  in   1 explanations,  diagrammatic  or  other  visual  representations  of  mechanisms  lack  the   1 requisite  propositional  structure  to  figure  in  such  logical  relationships;  if  physical   mechanisms  themselves  figure,  they  lack  such  features  even  more  obviously.   1   The  extent  to  which  mechanisms  constrain  ontological  commitments  is  also  a  point   1 of  dispute.  Some  of  these  debates  are  best  understood  in  terms  of  a  univocal  sense   of  mechanism  (namely,  mechanism ),  while  others  involve  tension  between   1 different  senses  of  mechanism.  Two  examples  of  intra-­‐sense  debate  concern  how  to   best  characterize  the  internal  structure  of  mechanisms,  and  the  extent  of  ontological   commitments  entailed  by  mechanism .  For  the  first  example,  there  are   1 disagreements  about  how  to  best  describe  the  process  of  a  mechanism  in  action.   Machamer,  Darden,  and  Craver  (2000)  utilize  start-­‐up  conditions  that  trigger  the   mechanism,  and  termination  conditions  that  are  the  last  stage  within  the   mechanism.  Using  the  example  of  circadian  rhythms,  Bechtel  and  Abrahamsen   (2010)  argue  instead  that  mechanisms  are  often  cyclical,  and  therefore  lack  clear   start-­‐up  or  termination  conditions.       8 The  second  example  involves  disagreement  about  how  much  ontological   commitment  should  accompany  mechanism .  While  some  authors  who  defend  this   1 new  mechanism  approach  do  so  in  an  ontologically  agnostic  way  (for  instance,   Bechtel  and  Richardson  1993),  mechanism  is  clearly  well-­‐suited  to  certain  kinds  of   1 ontological  commitments.2  For  instance,  Machamer,  Darden,  and  Craver  (2000)   argue  for  a  dualistic  entity-­‐activity  ontology  to  ground  the  mechanistic  account  of   1 explanation.  They  characterize  mechanisms  as  “entities  and  activities  organized   such  that  they  are  productive  of  regular  changes  from  start  or  set-­‐up  conditions  to   finish  or  termination  conditions”  (2000,  3).  Mechanisms  are  organized  chains  of   1 entities  connected  via  productive  activities  of  those  entities;  the  organized  activities   provide  both  continuity  to  the  mechanism  and  regularity  of  operation  (ibid.).   Phenomena  to  be  explained  via  mechanisms  are  either  the  final  product  that  results   once  termination  conditions  are  reached  or  a  higher-­‐level  regularity  to  which  the   mechanism  gives  rise.  Thus,  mechanisms  can  explain  (at  least)  two  ontological  kinds   of  target  phenomenon.  One  kind  is  the  effects  of  causal  relationships,  where  the   explanandum  is  the  final  stage  in  an  often  complex  causal  chain  at  the  same  level  as   the  rest  of  the  entities  and  activities  in  the  mechanism.  The  second  kind  is   phenomena  at  a  higher  level  than  the  explanans,  where  causal  activities  connect  the   entities  at  a  lower  level,  and  the  entire  mechanism’s  operation  constitutes  the   higher-­‐level  phenomenon.       Machamer,  Darden,  and  Craver  (2000)  explicitly  aim  to  offer  a  view  of  mechanisms   1 that  is  compatible  with  Salmon’s  (1984)  process  account  of  causation.  In  contrast  to   their  commitment  to  the  process  account,  the  causal  relationships  binding   mechanisms  together  could  be  construed  in  terms  of  counterfactuals,  or   1 dispositions,  or  capacities  (e.g.  Ilari  and  Williamson  2010).  Accordingly,  we  see  that   mechanism  is  not  itself  an  account  of  causation  (this  will  be  more  fully  explored  in   1                                                                                                                 2  This  is  not  to  say  that  mechanisms  are  compatible  with  just  any  ontology:  the   1 epistemological  or  methodological  features  of  the  new  mechanisms  approach  place   compatibility  constraints  on  ontological  commitments.  Rather,  the  point  here  is  that   these  constraints  are  not  enough  to  dictate  a  single  unique  ontology.     9 the  next  section),  however  much  claims  about  causation  figures  in  ongoing  debates   within  the  mechanisms  literature.  Williamson’s  (2011)  criterion  for  being  a   mechanistic  theory  of  causality,  that  “two  events  are  causally  connected  if  and  only   if  they  are  connected  by  an  underlying  physical  mechanism  of  the  appropriate  sort”   (421),  is  not  met  by  mechanisms .     1   The  anti-­‐reductive  aspect  of  mechanisms  generates  questions  about  interlevel   1 causation;  this  is  a  debate  concerning  causal  relationships  within  mechanisms ,  not   1 about  the  nature  of  causation  per  se.  The  nested,  hierarchical  structure  of   mechanisms  allows  for  non-­‐arbitrary  ways  to  individuate  levels  in  the  world,  in  a   local  and  context-­‐sensitive  rather  than  broad  or  generally  applicable  way.  It  also   raises  the  question  of  whether  direct  causal  relationships  between  entities  or   processes  at  different  levels  are  possible.3  Craver  and  Bechtel  (2007)  reject  the   possibility  of  interlevel  causation.  They  argue  for  a  differentiation  of  levels  such  that   it  is  only  meaningful  to  speak  of  higher  and  lower  levels  within  a  single  given   mechanism.  The  levels  in  a  mechanism  form  a  hierarchy  structured  by  the   constitutive  relationships  between  phenomena  and  the  mechanisms  that  give  rise  to   them.  Craver  and  Bechtel  note  a  common  stricture  on  causal  relata,  such  that  parts   of  a  whole  cannot  directly  enter  into  causal  relationships  with  that  whole.  They   conclude  that  there  is  no  meaningful  way  to  speak  of  causal  relationships  between   different  levels  within  a  mechanism.  Either  those  relationships  must  be  non-­‐causal   and  constitutive,  or  they  must  not  be  between  genuinely  different  levels,  since  they   occur  between  different  mechanisms  and  levels  are  only  defined  within  individual   mechanisms.       This  conclusion,  however,  is  in  tension  with  much  scientific  practice,  where  levels   are  often  treated  as  robust  differentiations  across  multiple  mechanisms.  Mitchell   (2008)  provides  a  number  of  examples  of  what  she  calls  multilevel  causation  and                                                                                                                   3  There  are  many  ways  to  construe  levels;  Craver  (2007)  offers  a  different  kind  of   field  guide  to  many  of  those  ways.  Here,  I  focus  on  levels  within  a  mechanism,  which   are  defined  in  terms  of  nested  mechanisms.       10