Isaac Record Writing Sample


Isaac  Record   1  

Technology  and  Epistemic  Possibility  

   Isaac  Record      

(Forthcoming  in  The  Journal  for  the  General  Philosophy  of  Science)  

Abstract  

My  aim  in  this  paper  is  to  give  a  philosophical  analysis  of  the  relationship  between  contingently  available  technology  and  

the  knowledge  that  it  makes  possible.  My  concern  is  with  what  specific  subjects  can  know  in  practice,  given  their  

particular  conditions,  especially  available  technology,  rather  than  what  can  be  known  “in  principle”  by  a  hypothetical  

entity  like  Laplace’s  Demon.  The  argument  has  two  parts.  In  the  first,  I’ll  construct  a  novel  account  of  epistemic  

possibility  that  incorporates  two  pragmatic  conditions:  responsibility  and  practicability.  For  example,  whether  subjects  

can  gain  knowledge  depends  in  some  circumstances  on  whether  they  have  the  capability  of  gathering  relevant  evidence.  

In  turn,  the  possibility  of  undertaking  such  investigative  activities  depends  in  part  on  factors  like  ethical  constraints,  

economical  realities,  and  available  technology.  In  the  second  part  of  the  paper,  I’ll  introduce  “technological  possibility”  

to  analyze  the  set  of  actions  made  possible  by  available  technology.  To  help  motivate  the  problem  and  later  test  my  

proposal,  I’ll  focus  on  a  specific  historical  case,  one  of  the  earliest  uses  of  digital  electronic  computers  in  a  scientific  

investigation.  I  conclude  that  the  epistemic  possibility  of  gaining  access  to  scientific  knowledge  about  certain  subjects  

depends  (in  some  cases)  on  the  technological  possibility  for  making  responsible  investigations.  

Keywords  

Epistemic  possibility,  epistemic  modality,  technological  possibility,  responsibility,  practicability  

Introduction  

Let  me  begin  with  a  historiographical  debate  about  how  to  understand  the  earliest  uses  of  digital  electronic  computers  in  

scientific  investigations.  Peter  Galison  argues  for  the  necessity  of  the  digital  electronic  computer  to  certain  advances  in  

scientific  knowledge  during  the  Manhattan  Project.  “Some  kind  of  numerical  modelling  was  necessary,  and  here  nothing  

could  replace  the  prototype  computer  just  coming  into  operation  in  late  1945:  the  ENIAC”  (Galison  1996,  122).  Galison’s  

claim  is  that  a  technological  change  made  it  possible  for  scientific  knowledge  to  develop.  John  Agar  argues  against  the  

necessity  of  that  technological  change  for  the  advancement  of  knowledge:  other  approaches  could  have  sufficed.  

                                                
   I.  Record,  Postdoctoral  Research  Fellow  

Faculty  of  Information,  University  of  Toronto,  140  St.  George  Street,  Toronto,  ON    M5S  3G6,  Canada  

isaac.record@utoronto.ca  



Isaac  Record   2  

“Computerization  was  usually  first  proposed  when  the  existing  practices  and  technologies  were  still  capable  of  the  

computational  task  at  hand”  (Agar  2006,  873).  Implicit  in  these  statements  are  crucial  assumptions  about  the  

relationship  between  technology  and  knowledge.  Galison  and  Agar  agree  that  certain  scientific  knowledge  became  

accessible  to  Manhattan  Project  scientists  once  they  began  to  run  “Monte  Carlo”  calculations,  named  after  the  famous  

gambling  establishment  because  of  the  method’s  use  of  random  numbers.  They  disagree  about  which  changes  in  the  

situation  of  the  scientists  made  those  investigations  possible.  Galison  thinks  access  to  digital  electronic  computers  made  

the  difference.  Agar  thinks  that  the  Monte  Carlo  method  could  have  been  implemented  using  existing  computational  

approaches.  

Both  analyses  turn  on  what  was  possible  under  the  circumstances  rather  than  on  what  actually  transpired.  The  conflict  

between  Galison’s  and  Agar’s  claims  about  the  digital  electronic  computer  arises  because  they  invoke  different  implicit  

notions  of  what  was  possible.  The  commonsense  notion  of  possibility  is  just  what  might  happen,  what  might  exist,  or  

what  might  be  true.  But  in  practice,  we  freely  constrain  these  generic  notions  of  possibility  to  reflect  narrower  concerns.  

For  example,  suppose  I  claim  that  driving  the  wrong  way  down  a  one-­‐way  street  is  impossible.  The  truth  of  that  claim  

turns  on  the  kind  of  possibility  we  invoke  in  evaluating  it.  Driving  the  wrong  way  down  a  one-­‐way  street  is  physically  

possible,  because  the  laws  of  physics  do  not  forbid  it  (making  my  claim  false  on  this  analysis).  Alternatively,  because  a  

municipal  law  does  forbid  it,  it  is  regulatively  impossible  (making  my  claim  true).  Similarly,  my  traveling  to  the  nearby  star  

Alpha  Centauri  by  2020  is  (let’s  say)  theoretically  and  physically  possible,  but  simultaneously  technologically  and  

economically  impossible.  To  return  to  the  case  at  hand,  Agar  is  saying,  roughly,  that  digital  electronic  computers  were  

not  necessary  for  completing  the  needed  work;  Galison,  that  they  were.  What  is  at  stake  in  this  disagreement  is  an  

understanding  of  the  role  that  a  particular  piece  of  technology  played  in  the  practice  of  science  at  a  particular  time.  

According  to  Galison,  the  digital  electronic  computer  brought  certain  propositions  into  the  realm  of  scientific  

knowledge—that  is,  they  changed  what  was  technologically  possible  (allowing  the  Monte  Carlo  method  to  be  put  into  

practice),  and  consequently  what  was  epistemically  possible  for  the  scientists.  Contrariwise,  Agar  thinks  this  way  of  

putting  things  gives  too  much  credit  to  the  material  means  of  accomplishing  a  task.  Instead,  a  conceptual  means  (the  

Monte  Carlo  method)  made  those  advancements  possible—and  could  have  done  so  without  being  implemented  on  a  

digital  electronic  computer.  

In  order  to  understand  and  evaluate  Galison’s  and  Agar’s  claims  about  the  difference  technology  makes  to  knowledge,  

we  need  an  account  that  explicitly  recognizes  that  technological  changes  make  it  possible  for  individuals  to  undertake  

different  actions,  and  some  of  these  actions  make  it  possible  for  those  individuals  to  gain  different  knowledge.  The  rest  

of  this  paper  is  devoted  to  this  task,  and  the  paper  is  divided  into  two  parts.  In  the  first  part,  concerning  the  relationship  

between  possible  actions  and  possible  knowledge,  I’ll  argue  that  we  need  an  account  of  epistemic  possibility  that  

captures  the  dependency  of  knowledge  on  being  able  to  take  the  appropriate  action.  It  is,  I  think,  uncontroversial  to  say  

that  being  able  to  complete  certain  actions  can  be  a  necessary  condition  for  gaining  knowledge.  In  scientific  practice,  for  

example,  gaining  knowledge  depends  on  having  relevant  evidence,  which  makes  being  able  to  gather  the  evidence  a  

condition  for  gaining  the  knowledge.  My  contibution  is  to  argue  that  because  a  scientist  is  (under  certain  conditions)  



Isaac  Record   3  

expected  to  seek  evidence  before  making  a  knowledge  claim  within  her  domain  of  expertise,  we  need  to  build  this  

expectation  into  our  account  of  knowledge—and  because  expectations  are  not  always  fulfilled,  the  appropriate  

philosophical  concept  is  epistemic  possibility.  

In  the  second  part  of  the  paper,  I  turn  to  the  relationship  between  technology  and  possible  actions.  A  number  of  

practical  factors  affect  our  ability  to  act,  including  economics  and  ethics,  but  I  will  focus  on  technology,  touching  on  the  

others  only  incidentally.  I’ll  introduce  an  analysis  of  technological  possibility,  which  depends  on  the  availability  of  

material  and  conceptual  means  to  bring  about  a  desired  state  of  affairs,  and  argue  that  the  epistemic  possibility  of  

gaining  access  to  scientific  knowledge  depends  (in  some  cases)  on  the  technological  possibility  for  carrying  out  certain  

investigations.  In  such  cases,  technological  possibility  can  be  seen  as  a  necessary  condition  for  epistemic  possibility.  

Finally,  I  will  return  to  the  disagreement  between  Galison  and  Agar  and  show  how  my  analysis  of  epistemic  and  

technological  possibility  resolves  the  conflict.  

Part  1.  Doing  and  Knowing  

My  overarching  aim  in  this  paper  is  to  give  an  account  of  the  relationship  between  contingently  available  technology  and  

the  knowledge  that  it  puts  within  ‘epistemic  reach,’  to  use  Egan’s  vivid  phrase  (2007,  8).  The  relevant  philosophical  

concept  here  is  epistemic  possibility,  which  is  meant  to  reflect  epistemic  reach  by  distinguishing  between  what  a  subject  

can  and  cannot  know  given  her  epistemic  circumstances.  The  Agar-­‐Galison  example  illustrates  that  some  knowledge  

claims  require  the  gathering  of  evidence,  which  suggests  an  understanding  of  epistemic  reach  that  is  responsive  to  the  

actions  a  subject  can  actually  accomplish.  Canonical  accounts  of  epistemic  possibility  tend  to  be  insensitive  to  this  issue,  

as  I  will  show.  I  will  develop  a  novel  account  of  epistemic  possibility  that  takes  into  consideration  practical  conditions  for  

knowing,  taking  particular  care  to  develop  my  account  in  such  a  way  that  it  can  accommodate  epistemic  responsibilities  

such  as  the  evidence-­‐seeking  duties  scientists  adopt  when  they  aim  to  produce  scientific  knowledge.  As  I  will  argue,  this  

requires  a  definition  of  epistemic  possibility  that  includes  both  a  practicability  criterion  and  a  responsibility  criterion.  My  

approach  will  be  to  begin  with  a  canonical  definition  of  epistemic  possibility1  and  then  elaborate  it.  

The  usual  starting  point  for  epistemic  possibility  is  that  it  should  somehow  reflect  a  subject’s  epistemic  position.  Thus,  for  

a  subject  S  to  claim  that  the  proposition  Φ  is  epistemically  possible  is  for  S  to  say  that  Φ  is  possible  relative  to  S’s  

                                                
1  The  main  thread  of  philosophical  accounts  of  epistemic  possibility  traces  at  least  to  Moore  (1962),  and  the  topic  has  
enjoyed  a  recent  flourishing  (see,  e.g.,  the  collected  volume  by  Egan  and  Weatherson  [2011]).  According  to  the  canonical  
account,  when  a  subject  says  that  something  is  possible,  we  should  understand  this  claim  as  being  made  relative  to  a  
person,  a  group,  or  a  set  of  information.  But  precisely  which  person,  group,  or  set  of  information?  Egan  and  Weatherson  
(2011)  identify  three  clusters  of  answers:  contextualists  think  the  context  of  utterance  somehow  specifies  an  answer  
(see  DeRose  1991;  von  Fintel  and  Gillies  2011),  relativists  look  to  the  context  of  assessment  (see  Egan  et  al.  2005;  
MacFarlane  2011),  and  expressivists  think  modal  statements  express  the  speaker’s  mental  state  of  uncertainty  (see  
Yalcin  2011).  The  discussion  in  this  paper  centers  on  the  canonical  contextualist  accounts,  though  I  will  make  note  of  
some  divergent  views.  



Isaac  Record   4  

epistemic  position.  Taking  “S’s  epistemic  position”  to  be,  in  the  simplest  case,  “what  S  knows,”  leads  straightforwardly  to  

the  canonical  definition  of  epistemic  possibility  (Hacking  [1967],  Teller  [1972],  DeRose  [1991]  and  the  individual  

contributors  to  Gendler  and  Hawthorne  [2002]  and  Egan  and  Weatherson  [2011]  all  take  this  as  their  point  of  

departure):  

(a)  Φ  is  epistemically  possible  for  S  if  S  doesn’t  know  –Φ.  

The  idea  is  that  if  S  doesn’t  know  for  certain  that  Φ  is  not  the  case,  then  S  must  consider  Φ  to  be  possible.  For  example,  if  

S  knows  Φ,  then  S  cannot  know  –Φ,  and  Φ  is  epistemically  possible.  On  the  other  hand,  if  S  knows  –Φ,  then  Φ  is  by  

definition  epistemically  impossible.  Finally,  if  S  knows  neither  Φ  nor  –Φ,  then  Φ  is  epistemically  possible  for  S  (as  is  –Φ).  

More  concretely,  if  I  have  just  checked  my  key  hook  for  my  lost  keys  (and  failed  to  locate  them  there),  then  for  me,  it  is  

not  epistemically  possible  for  my  keys  to  be  on  the  key  hook.  But  if  I  have  not  yet  looked  on  the  table,  then  it  is  

epistemically  possible  for  my  keys  to  be  there,  assuming  I  have  no  other  reasons  for  excluding  that  possibility.  

Note  that,  on  most  accounts,  epistemic  possibility  is  sensitive  to  what  S  could  know  given  S’s  epistemic  position  rather  

than  reflecting  what  S  actually  believes.2  Suppose  S  deems  Φ  possible,  forgetting  to  take  into  account  that  –Φ.  In  such  a  

case,  S  is  wrong;  S  should  know  better  than  to  think  Φ  is  possible.  That  is,  Φ  is  in  fact  not  epistemically  possible  for  S,  

even  if  S  thinks  that  Φ  is  possible.  In  such  cases,  epistemic  possibility  provides  grounds  to  blame  S  for  a  misjudgment.  

As  it  turns  out,  taking  S’s  epistemic  position  to  mean  “what  S  knows”  leads  almost  immediately  to  results  that  confound  

the  intuitions  of  some  philosophers.  (a)  presumes  that  the  only  factor  relevant  to  S’s  epistemic  position  is  what  S  knows  

at  the  time,  a  condition  that  fails  to  hold  for  any  proposition  S  hasn’t  considered  before  (at  least  for  any  view  of  

knowledge  that  includes  “belief”  as  a  necessary  condition).  Suppose  Φ  is  the  proposition  that  “4+3=9,”  something  S  

would  reject  upon  even  a  moment’s  consideration.  Nevertheless,  if  S  has  never  considered  whether  “4+3=9,”  then,  

according  to  (a),  “4+3=9”  is  epistemically  possible  for  S,  because  S  has  no  beliefs  about  it  whatever.  On  this  view,  if  S  

blurts  that  “Perhaps  ‘4+3=9’”  without  pausing  to  consider  it,  we  have  no  cause  to  say  S  is  wrong,  for  “4+3=9”  really  is  

epistemically  possible  for  S.  Yet  if  S  should  later  consider  whether  “4+3=9,”  S  would  immediately  judge  it  to  be  

impossible,  and  could  then  be  blamed  for  saying  that  “4+3=9”  is  possible  (for  further  discussion  of  cases  like  this,  see  

Huemer  [2007]  and  Yalcin  [2011]).  

If  the  goal  of  epistemic  possibility  is  to  reflect  S’s  epistemic  position,  it  seems  like  a  strange  consequence  that  we  can  

blame  S  for  failing  to  recall  Φ,  but  not  for  failing  to  consider  it.  Several  accounts  of  epistemic  possibility  attempt  to  close  

                                                
2  Expressivists  would  presumably  disagree.  For  expressivists,  epistemic  possibilities  express  a  subject’s  uncertainty  rather  
than  describing  a  relationship  between  the  subject’s  knowledge  (or  information)  and  the  state  of  the  world  (see,  e.g.  
Yalcin  2011).  I  must  set  this  view  aside  in  this  paper.  



Isaac  Record   5  

the  gap  between  failing  to  recall  and  failing  to  consider  by  expanding  “epistemic  position”  to  include  everything  “within  

epistemic  reach,”  yet  as  we  will  see,  it  has  been  difficult  to  specify  just  what  counts  as  being  within  epistemic  reach.    

If  (a)  fails  to  capture  what  is  within  epistemic  reach,  perhaps  we  can  simply  add  such  a  description  to  the  original  

definition.  For  example,  we  might  say  that:  

(b)  Φ  is  epistemically  possible  for  S  if  S  does  not  know  that  –Φ,  nor  would  careful  
reflection  establish  that  –Φ.  

Here,  S  isn’t  allowed  to  simply  blurt  out  that  “perhaps  ‘4+3=9.’”  She  must  first  carefully  reflect  upon  the  proposition.  This  

eliminates  the  problem  of  unconsidered  cases  like  “4+3=9,”  while  still  being  limited  to  the  knowledge  S  has  (plus  

inferences  from  that  knowledge).  Unfortunately,  “careful  reflection”  is  too  vague  a  requirement  to  capture  what  is  

practicably  within  a  subject’s  epistemic  reach,  which  means  that  definition  (b)  fails  to  appropriately  reflect  what  S  is  in  a  

position  to  know  in  practice.  For  example,  Goldbach’s  conjecture  states  that  every  even  integer  greater  than  two  can  be  

written  as  the  sum  of  two  primes.  It  hasn’t  been  proved  or  disproved,  but  the  axioms  of  mathematics  are  such  that  

Goldbach’s  conjecture,  if  true,  is  true  necessarily,  and  if  false,  is  false  necessarily.  Mathematicians  don’t  yet  know  its  

truth-­‐value,  and  many  hours  of  careful  reflection  have  not  resolved  the  situation.  Nevertheless,  some  amount  of  

additional  reflection  might  solve  it,  as  has  transpired  for  many  other  mathematical  conjectures.  The  point  is  that  “careful  

reflection”  doesn’t  distinguish  between  five  minutes,  five  hours,  or  five  years  of  reflection.  (Stanley’s  [2005]  suggestion  

that  S  take  into  account  “obvious  entailments”  of  what  S  knows  seems  to  me  to  do  a  little  better  than  (b),  but  it  fails  to  

respond  to  Hacking’s  criticism  of  (c),  below.)  But  is  (b)  merely  too  vague  in  describing  epistemic  reach,  or  are  we  on  the  

wrong  track  altogether?  

I  propose  to  expand  the  notion  of  epistemic  reach  to  include  practicable  responsibilities.  Careful  reflection  remains  a  

plausible  starting  point,  though  to  be  complete  we  would  need  to  say  how  much  reflection  a  subject  is  responsible  to  

perform,  and  how  much  reflection  is  practicable.  After  considering  some  other  proposals,  I  will  argue  that  the  limits  of  

practicable  responsibility  are  determined  by  context.  But  before  continuing,  let  me  make  the  case  for  responsibility,  

since  this  suggestion  will  strike  some  as  tendentious.  

Richard  Foley  suggests  that  “our  everyday  evaluations  tend  to  be  concerned  with  whether  one  has  been  responsible  in  

arriving  at  one’s  beliefs”  (2003,  9).  Let  me  give  two  examples  of  responsibly  arriving  at  one’s  beliefs.  First,  in  the  context  

of  scientific  knowledge  claims,  we  routinely  expect  these  claims  to  carry  special  weight  in  light  of  experimental  evidence  

or  theoretical  justification.  Accordingly,  we  impose  special  responsibilities,  sometimes  called  epistemic  duties,  on  

scientists  (see,  e.g.,  Kornblith  [1983]).  Lest  we  conclude  that  role  responsibilities  are  a  special  case,  I  point  out  that  

epistemic  duties  appear  in  everyday  cases  too.  When  I  call  my  office  to  ask  a  colleague  whether  a  letter  I  am  expecting  

has  arrived,  I  won’t  be  satisfied  with  the  claim  that  it  is  “possible”  that  the  letter  has  arrived—I  want  to  know  one  way  or  

the  other!  I  want  my  colleague  to  check  the  incoming  mail.  But  reasonable  expectations  have  limits:  I  won’t  blame  my  

colleague  for  not  noticing  that  the  letter  has  slipped  behind  a  desk  or  was  delivered  to  the  wrong  recipient.  The  point  is  

that  epistemic  possibility  should  not  only  reflect  what  knowledge  S  already  has,  but  should  also  take  into  consideration  



Isaac  Record   6  

S’s  responsibilities  to  gather  additional  evidence.  On  the  epistemic  responsibilities  view,  S  should  not  always  settle  for  

the  evidence  she  has  in  hand,  but  must  in  at  least  some  cases  conduct  an  inquiry  or  seek  new  evidence  before  making  a  

knowledge  claim.  These  facts  are  a  part  of  a  subject’s  epistemic  circumstances,  and  so  should  be  reflected  in  our  analysis  

of  epistemic  possibility.  

While  the  idea  behind  epistemic  responsibilities—that  sometimes  we  must  back  up  our  claims—is  fairly  straightforward,  

the  existence  and  nature  of  epistemic  responsibilities  are  controversial  in  epistemology.  I  should  mention  that  an  

alternative  to  the  epistemic  responsibility  view  states  that  although  we  do  sometimes  have  the  duty  of  seeking  new  

evidence,  that  duty  should  be  understood  as  being  moral,  not  epistemic  (see,  e.g.,  Conee  and  Feldman  [2004]).  My  

account  is  compatible  with  either  view,  but  I  shall  use  the  term  “epistemic  responsibility”  to  indicate  any  duty  that  is  a  

condition  for  making  a  knowledge  claim  (indeed,  I  shall  use  the  term  whether  or  not  the  duty  promotes  genuine  

knowledge).  I  will  also  set  aside  the  larger  question  of  whether  we  always  have  epistemic  responsibilities,  and  instead  

distinguish  between  weak  epistemic  possibility,  which  does  not  include  responsibilities,  and  strong  epistemic  possibility,  

which  does  include  responsibilities.  For  the  remainder  of  the  paper,  when  I  refer  to  “epistemic  possibility”  I  mean  strong  

epistemic  possibility.  

Because  responsibility  is  a  novel  contribution  to  the  discussion  of  epistemic  possibility,  let  me  briefly  describe  some  

features  of  what  I  take  to  be  a  plausible  account  of  epistemic  responsibility.  I  won’t  defend  such  a  view  here;  rather,  I  

merely  want  to  show  that  some  account  might  be  made  to  work  with  my  version  of  epistemic  possibility.    First,  context-­‐

relevant  risks  may  be  distinguished  from  background-­‐level  risks.  Second,  epistemic  responsibilities  need  respond  only  to  

context-­‐relevant  risks.  And  third,  background  risks  may  be  converted  into  context-­‐relevant  risks  (and  vice-­‐versa)  through  

negotiation.  

There  is  a  distinction  to  be  maintained  between  the  question  of  when  Φ  is  justified  and  when  S  has  warrant  to  claim  Φ  

(see,  e.g.,  Williams  2001).  Epistemic  responsibilities,  as  I  use  the  term  here,  have  to  do  with  claiming.  S’s  making  a  claim  

about  Φ  invokes  a  responsibility,  but  fulfilling  this  responsibility  does  not  guarantee  that  a  claim  is  justified.  What  

determines  epistemic  responsibility  is  not  the  actual  epistemic  risk  of  a  particular  claim,  but  its  perceived  (context-­‐

relevant)  risk  relative  to  a  particular  set  of  background  commitments  that  S  need  not  defend.  The  focus  on  context-­‐

relevant  risks  allows  us  to  bracket  the  epistemic  risks  associated  with  the  background  commitments  in  order  to  stay  

focused  on  foreground  issues.  To  give  an  extreme  example,  in  deciding  whether  to  accept  a  stranger’s  testimony  about  

the  local  bus  schedule,  we  tend  to  disregard  the  possibility  that  the  external  world  is  an  illusion.  The  idea  is  not  to  ignore  

those  background-­‐level  risks,  but  merely  to  focus  on  the  risks  associated  with  a  particular  claim  within  the  relevant  

context.  Focusing  on  the  contextual  claim  has  the  effect  of  “normalizing”  or  rescaling  its  risks  against  a  chosen  

background.  Background  risks  don’t  disappear;  they  are  simply  shifted  away  from  center  stage.  

Background  risks  can  be  accommodated  in  a  number  of  ways.  We  can  demand  that  they  be  traced  (or  be  traceable)  to  

basic  beliefs  (as  in  foundationalist  accounts  of  knowledge);  we  can  reduce  them  to  mere  stipulations  (as  in  some  

relativist  accounts);  or  (as  I  prefer)  we  can  recognize  that  what  counts  as  background  is  negotiable.  As  Helen  Longino  



Isaac  Record   7  

puts  it  (with  respect  to  propositional  scientific  knowledge),  “as  long  as  background  beliefs  can  be  articulated  and  

subjected  to  criticism  from  the  scientific  community,  they  can  be  defended,  modified,  or  abandoned  in  response  to  such  

criticism”  (Longino  1990,  73-­‐74).  The  effect  of  putting  background  on  the  bargaining  table  is  to  create  a  sort  of  “division  

of  labor”  for  epistemic  risks.  Even  if  some  risks  are  unaddressed  or  unknown  at  a  given  time,  they  can  be  articulated  and  

worried  over  at  a  later  date  (and  dependent  foreground  risks  can  be  recalibrated  accordingly).  This  negotiation  model  

seems  to  fit  the  way  that  science  works,  at  least  some  of  the  time.  For  example,  in  order  to  conduct  detailed  research,  a  

scientist  interested  in  molecular  physics  takes  on  board  the  risk  of  being  wrong  about  causality,  mass-­‐energy  

conservation  laws,  statistical  laws,  and  so  on.  But  setting  those  risks  aside  doesn’t  mean  accepting  them  unquestioningly;  

scientists  decide  which  risks  they  need  to  address  before  making  a  claim,  and  other  scientists  decide  whether  the  

appropriate  risks  have  been  addressed  before  accepting  the  claim.    

Part  of  what  is  being  negotiated  is  who  is  responsible  for  addressing  particular  epistemic  risks.  Responsibilities  may  be  

stronger,  weaker,  or  even  unrelated  to  the  justification  standards  for  knowledge.  Ideally,  S’s  fulfilling  the  epistemic  

duties  for  being  able  to  claim  Φ  would  be  necessary  and  sufficient  to  justify  Φ.  But  suppose  that  there  is  a  mismatch,  and  

fulfilling  responsibilities  is  insufficient  to  justify  Φ.  Nevertheless,  fulfilling  responsibilities  is  still  necessary  for  having  

knowledge  of  Φ,  because  any  claim  that  fails  to  fulfill  responsibilities  is  a  non-­‐starter  in  the  context  in  which  it  is  made.  

That  is,  fulfilling  responsibilities  is  a  necessary  but  insufficient  condition  for  Φ  being  strongly  epistemically  possible  for  S.  

It  would  be  nice  if  we  knew  which  responsibilities  are  relevant  to  justification,  but  we  simply  cannot  be  certain.  Inquiry  in  

fields  like  science  works  on  the  basis  of  their  internal  standards,  which  sometimes  produce  genuine  knowledge  and  

sometimes  not.  But  in  order  for  a  claim  to  be  eligible  for  consideration,  the  claimant  has  to  fulfill  the  relevant  epistemic  

responsibilities.    

Whether  or  not  the  preceding  sketch  of  how  epistemic  responsibilities  are  generated  is  exactly  right  in  its  details,  I  think  

it  is  plausible  that  subjects  often  have  a  responsibility  to  go  beyond  their  present  beliefs,  such  responsibilities  depend  on  

a  subject’s  knowledge  context,  and  these  responsibilities  are  relevant  to  evaluating  what  is  epistemically  possible  for  

them.  I  will  refer  to  this  as  the  “responsibility  criterion”  for  epistemic  possibility.  The  question  is  how  to  incorporate  

responsibility  into  the  definition  of  epistemic  possibility.  

One  plausible  solution  is  to  include  S’s  epistemic  community  in  the  definition,  since  this  group  negotiates  the  boundaries  

of  context-­‐relevant  responsibilities.  It  turns  out  that,  for  other  reasons,  the  inclusion  of  community  is  a  common  

proposal  among  contextualists  as  well,  so  there  is  a  robust  literature  to  work  from.  For  those  accounts,  the  usual  idea  is  

that  if  our  concern  is  that  (a)  and  (b)  don’t  adequately  reflect  what  is  in  epistemic  reach  of  the  subject  S,  we  must  make  

epistemic  possibility  sensitive  to  the  knowledge  or  information  available  to  the  entire  group  to  which  S  makes  her  claim.  

To  put  it  another  way,  the  knowledge  of  everyone  in  the  group  is  within  the  epistemic  grasp  of  any  member:  she  need  

merely  ask.  

(c)  Φ  is  epistemically  possible  for  S  if  S  does  not  know  that  –Φ,  nor  does  any  member  
of  C,  where  C  is  S’s  epistemic  community.  



Isaac  Record   8  

The  advantage  to  this  definition  is  that  it  smoothes  out  some  of  the  peculiarities  of  S’s  particular  thought  processes,  

while  remaining  true  to  human  limitations.  Even  if  S  hasn’t  considered  whether  Φ,  perhaps  someone  else  in  C  (however  

we  wish  to  define  the  community)  has  ruled  it  out.  The  aim  is  not  to  require  that  S  know  everything  known  to  everyone  

else  in  S’s  community,  but  rather  to  hold  S  responsible  for  judgments  that  clash  with  what  is  known  to  someone  else  in  

the  community.3  Variations  on  (c)  abound.  Indeed,  von  Fintel  and  Gillies  identify  as  “canon”  the  view  that  “epistemic  

modals  quantify  over  the  information  available  to  a  contextually  relevant  group.  The  context  decides  the  group  (and  

perhaps  the  standards  by  which  they  know)”  (2011,  108).  In  a  scientific  community,  this  definition  works  rather  well  in  

principle,  because  knowledge  is  (ideally)  made  available  to  the  entire  community  by  mechanisms  such  as  conferences  

and  publication.  On  definition  (c),  S  can  be  deemed  wrong  on  the  basis  of  failing  to  take  into  account  results  published  by  

other  scientists.  Unfortunately,  this  elaborated  version  of  epistemic  possibility  has  difficulties  of  its  own,  as  Ian  Hacking  

shows.  

Imagine  a  salvage  crew  searching  for  a  ship  that  sank  a  long  time  ago.  The  mate  of  
the  salvage  ship  works  from  an  old  log,  makes  some  mistakes  in  his  calculations,  and  
concludes  that  the  wreck  may  be  in  a  certain  bay.  It  is  possible,  he  says,  that  the  hulk  
is  in  these  waters.  No  one  knows  anything  to  the  contrary.  But  in  fact,  as  it  turns  out  
later,  it  simply  was  not  possible  for  the  vessel  to  be  in  that  bay;  more  careful  
examination  of  the  log  shows  that  the  boat  must  have  gone  down  at  least  thirty  
miles  further  south.  (1967,  148)  

No  doubt  it  seemed  possible  that  the  vessel  was  in  the  bay  until  the  ship’s  mate  rechecked  his  calculations.  But  was  it  

really  epistemically  possible  for  him?  To  Hacking,  it  seems  not,  for  the  evidence  the  mate  used  to  justify  his  belief  that  it  

is  possible  that  the  vessel  is  in  the  bay  does  not,  in  fact,  support  that  claim.  It  supports  the  contrary  claim  that  it  is  

impossible  that  the  vessel  is  in  the  bay.  Hacking  concludes  that  “the  mate  said  something  false  when  he  said,  ‘It  is  

possible  that  we  shall  find  the  treasure  here,’  but  the  falsehood  did  not  arise  from  what  anyone  actually  knew  at  the  

time”  (1967,  148).    

Hacking  is  pointing  out  that  in  many  cases  there  is  an  expectation  that  S  has  checked—and  has  done  a  good  job—before  

making  a  claim  about  Φ.  For  Hacking,    

(d)  Φ  is  epistemically  possible  for  S  if  S  doesn’t  know  –Φ  nor  would  any  practicable  
investigations  by  S  establish  that  –Φ.  

Here,  the  idea  is  that  we  expected  the  mate  to  successfully  complete  certain  reasonable  actions  before  coming  to  his  

conclusion.  Since  he  didn’t  complete  them  successfully,  we  have  grounds  to  blame  him.  Hacking’s  definition  allows  us  to  

                                                
3  Exactly  how  to  define  the  relevant  community  is  a  matter  of  earnest  debate.  The  basic  premise  of  relativist  accounts  of  
epistemic  possibility  is  that  contextualist  accounts  cannot  readily  handle  cases  of  agreement,  disagreement,  
eavesdropping,  or  temporally  dislocated  responses  to  purported  possibility  claims.  See  Egan  (2007)  and  MacFarlane  
(2011).  See  also  von  Fintel  and  Gillies  (2011)  for  a  review  of  some  responses.  



Isaac  Record   9  

adjudicate  the  Goldbach  case  satisfactorily:  S  is  now  only  responsible  to  complete  investigations  that  fall  within  

practicable  limits.  Exactly  where  we  draw  that  line  is  still  vague,  but  at  least  we  now  have  a  principle  for  drawing  one.  I  

will  refer  to  this  as  the  “practicability  criterion”  for  epistemic  possibility.  

As  it  turns  out,  “practicability”  alone  doesn’t  always  line  up  with  the  sort  of  epistemic  duties  we  impose  on  S.  Paul  Teller  

poses  this  rebuttal  to  Hacking’s  practicability  criterion:  Teller’s  wife  is  pregnant,  but  he  doesn’t  yet  know  the  sex  of  his  

child.  For  Teller,  it  is  epistemically  possible  that  his  child  will  be  a  boy,  and  at  the  same  time  epistemically  possible  that  

his  child  will  be  a  girl,  and  this  is  despite  the  fact  that  there  is  a  “practicable,  in  fact  quite  easy”  test  to  establish  the  sex  of  

Teller’s  child  (1972,  307).  (Incidentally,  according  to  Teller’s  account,  the  sex  test  was  newly  available  in  1972.  A  few  

years  earlier,  it  would  not  have  been  practicable.)  Teller  is  claiming  that  we  can’t  demand  that  he  have  this  test  

performed  before  he  answers  whether  it  is  possible  that  his  child  will  be  a  boy.  Put  another  way,  practicability  may  be  a  

necessary  condition  for  a  duty  to  be  imposed  on  S,  but  it  is  not  a  sufficient  one.    

Recall  that  Hacking  introduced  practicability  to  indicate  what  S  should  be  expected  to  know,  given  S’s  situation.  He  wants  

us  to  conclude  that  the  ship’s  mate  has  said  something  false  in  claiming  the  wreck  may  be  in  this  very  bay  because  the  

evidence  he  has  examined  should  have  told  him  otherwise.  The  mate  made  a  mistake  in  his  calculation,  and  it  is  easy  to  

seize  upon  this  and  say  that  the  mate  should  have  known  better.  But  the  relevant  contrast  isn’t  between  what  the  mate  

should  have  known  after  he  examined  the  log  and  what  he  actually  knew.  It’s  between  what  he  should  have  known  

before  checking  and  afterward.  What  do  we  want  to  demand  of  the  ship’s  mate  before  he  has  looked  inside  the  logbook  

for  the  first  time?  At  that  time,  the  mate’s  position  is  similar  to  that  of  our  expectant  father  before  a  sex  test  has  been  

performed  on  the  fetus.  The  mate  need  only  make  calculations  from  the  log  and  the  father  need  only  order  the  test.  In  

each  case,  should  S  successfully  complete  some  activity,  knowledge  of  Φ  can  be  had.  The  difference  is  not  in  the  

practicability  of  the  task:  the  father’s  task  is  easier,  if  anything.  The  difference,  Teller  surmises,  is  in  the  expectations  of  

the  community  C  of  which  S  is  a  member.  

Teller  therefore  proposes  the  following  emendation  of  the  “community  C”  version  of  epistemic  possibility  (definition  (c)):    

(e)  Φ  is  epistemically  possible  for  S  if  it  is  not  the  case  that:  
(1)  Φ  is  known  to  be  false  by  any  member  of  community  C,  
(2)  nor  is  there  a  member,  T,  of  community  C,  such  that  if  T  were  to  know  all  the  
propositions  known  to  community  C,  then  he/she  could,  on  the  strength  of  his/her  
knowledge  of  these  propositions  as  basis,  data,  or  evidence,  come  to  know  that  Φ  is  
false.  (Teller  1972,  310-­‐11)  

The  idea  is  to  restrict  epistemic  possibility  to  what  some  member,  T,  of  the  community  would  be  in  a  position  to  know  if  

T  had  all  of  the  relevant  communal  knowledge  at  hand.  For  my  purposes,  Teller’s  formulation  has  a  significant  problem:  

it  doesn’t  accommodate  responsibilities  that  would  have  a  subject  look  beyond  existing  knowledge.  Like  the  original  

“community”  variation,  it  addresses  only  what  is  already  known  by  the  community  (von  Fintel  and  Gillies  arrived  at  a  

similar  point  quite  independently;  see  their  [2011,  112-­‐13  fn.  9]).  



Isaac  Record   10  

Consider  a  slight  variation  on  Hacking’s  salvage  ship  problem.  Suppose  the  mate’s  mistaken  calculation  is  the  result  of  his  

having  skipped  a  line  in  the  log.  This  means  that  neither  the  mate,  nor  any  other  member  of  the  salvage  crew  knows  the  

relevant  propositions  about  the  location  of  the  treasure.  Yet  we  would  still  blame  the  mate  for  this  mistake.  Teller  

acknowledges  this  gap  in  his  account,  and  fills  it  in  by  counting  as  “known  to  community  C”  facts  written  down  in  books  

available  to  the  community  (1972,  312).  But  responsibility  to  access  extant  knowledge  isn’t  quite  what  we’re  after  for  

understanding  responsible  knowledge  in  scientific  contexts—we  usually  want  scientists  to  go  out  into  the  world  and  

check.  

Hacking’s  point  was  that  we  need  to  establish  some  reasonable  grounds  for  saying  the  mate  is  wrong.  His  answer  was  

practicability;  Teller’s  is,  essentially,  a  slightly  more  detailed  version  of  the  responsibility  account  we  saw  earlier  in  

definition  (c).  My  diagnosis  is  that  both  Teller  and  Hacking  have  part  of  the  story  right.4  The  difficulty  in  defining  

epistemic  possibility  is  in  correctly  balancing  the  practicability  and  responsibility  criteria.  This  is  difficult  to  do  outside  of  

specific  contexts,  and  the  solution  is  to  avoid  removing  context  from  the  analysis.  That  is,  rather  than  try  to  define  

practicability  or  responsibility  separately  in  some  objective  manner,  the  solution  is  to  observe  that  communities  

negotiate  and  define  practicable  responsibilities  for  themselves  based  on  their  interests,  including  assessments  of  

epistemic  risk.  In  the  case  of  scientific  communities,  practicable  responsibilities  are  (partially)  explicit:  scientists  must  

meet  specific  standards  of  evidence  and  justification  or  else  withhold  judgment  or  use  qualified  language.  Within  a  given  

community,  C,  if  an  individual,  S,  makes  a  knowledge  claim  and  meets  C’s  epistemic  standards,  E,  then  C  will  accept  it.  

That  is,  

(f)  Φ  is  epistemically  possible  for  S  if  S  does  not  know  that  –Φ,  nor  do  the  epistemic  
standards  E  of  community  C  demand  that  S  carry  out  any  practicable  investigation  
that  would  establish  that  –Φ.  

Let’s  see  how  my  proposal  handles  the  examples  we’ve  just  seen.  On  my  account,  an  expectant  father  can  rightly  say  

that  it  is  possible  his  child  will  be  a  boy  even  if  a  definitive  test  is  available,  because  his  community  (his  family  and  

friends)  does  not  demand  that  he  order  the  test.  By  contrast,  the  mate  on  the  salvage  crew  is  expected  to  eliminate  the  

present  bay  from  the  list  of  possible  locations  for  the  wreck,  since  his  community  (his  shipmates)  demands  that  he  glean  

this  information  from  the  log.  In  each  case,  the  relevant  community  (the  community  to  which  S  is  presenting  a  claim)  

decides  which  practicable  investigations  S  is  obliged  to  undertake.  It  is  the  epistemic  standard  of  the  salvage  crew  that  

lets  Hacking  deem  the  mate  wrong  when  he  claims  the  wreck  may  be  in  this  harbor.  And  it  is  the  epistemic  standard  of  

family  and  friends  that  let  Teller  deem  himself  correct  when  he  claims  that  his  child  may  be  a  boy  (even  if  his  child  were  

                                                
4  Others  have  attempted  to  amalgamate  parts  of  Hacking’s  and  Teller’s  accounts.  DeRose’s  (1991)  “relevant  way”  and  
Egan’s  (2007)  “epistemic  reach”  are  two  of  the  better  examples.  Both  remain  quite  vague  about  how  to  find  out  what  is  
a  relevant  way  within  epistemic  reach.  



Isaac  Record   11  

a  girl).  In  sum,  epistemic  possibility  lies  at  the  intersection  of  epistemic  responsbilities  and  practicable  actions.  There  may  

be  responsibilities  that  are  not  practicable  and  practicable  actions  that  are  not  responsibilities.    

*   *   *  

My  practicable  responsibilities  account  of  epistemic  possibility  can  also  illuminate  practical  discussions  of  possibility,  

such  as  the  one  with  which  I  began  the  paper.  Digital  electronic  computers  became  available  at  a  time  physicists  at  Los  

Alamos  were  butting  heads  with  a  difficult  and  dangerous  subject  matter:  nuclear  bombs.  In  order  for  their  claims  to  be  

accepted  within  their  epistemic  community,  Manhattan  Project  scientists  had  to  fulfill  certain  epistemic  responsibilities;  

for  example,  they  had  to  meet  precise  standards  of  evidence  and  justification  in  order  for  their  work  to  move  forward.  

Before  the  advent  of  the  digital  electronic  computer  running  Monte  Carlo  calculations,  scientists  did  not  fulfill  those  

responsibilities,  and  so  were  stuck—they  could  not  move  forward  on  their  bomb  work,  because  they  needed  knowledge  

that  was  unavailable  to  them.  That  is,  they  had  epistemic  responsibilities  they  could  not  discharge  without  specific  

knowledge  about  bombs,  and  that  knowledge  was  unavailable  because  certain  actions  hadn’t  yet  been  performed.  Agar  

and  Galison  agree  about  all  of  this.  But  they  disagree  about  whether  the  requisite  actions  could  have  been  performed  

before  the  advent  of  the  digital  electronic  computer.  That  is,  they  disagree  about  whether  fulfilling  those  responsibilities  

was  practicable  given  the  specific  situation  Manhattan  Project  scientists  were  in.    

If  Monte  Carlo  was  practicable  before  the  advent  of  digital  electronic  computers,  then  the  knowledge  the  scientists  

sought  was  within  their  epistemic  grasp—that  is,  it  was  epistemically  possible.  But  if  Monte  Carlo  was  impracticable  

before  digital  computers,  then  the  knowledge  they  sought  was  not  within  their  grasp.  What  makes  an  activity  

practicable?  According  to  Galison,  a  technology,  the  digital  electronic  computer,  made  practicable  for  Manhattan  Project  

scientists  the  gathering  of  the  required  evidence  to  fulfill  their  epistemic  responsibilities.  By  contrast,  Agar  thinks  that  

the  Monte  Carlo  method  could  have  been  implemented  using  older  equipment;  that  is,  that  Monte  Carlo  calculations  

were  practicable  before  they  were  actually  put  into  practice.  The  limiting  factor  was  a  lack  of  theoretical  guidance  

without  which  nuclear  experiments  were  too  dangerous  and  expensive  to  be  performed.  The  scientists’  theoretical  

efforts  were  stymied  by  intractable  analytic  equations.  Progress  slowed.  Then  available  technology  changed  and  

progress  resumed.  But  was  the  technological  change  decisive  or  merely  coincidental?  That  is,  did  the  digital  electronic  

computer  offer  new  technological  possibilities  that  made  practicable  a  method  that,  prior  to  the  digital  electronic  

computer,  had  been  impracticable?  

Let  me  be  clear:  the  relevant  constraint  on  epistemic  possibility  in  a  case  like  this  is  whether  (and  when)  the  scientists  

could  meet  their  epistemic  responsibilities  with  practicable  investigations.  If  they  could  do  so  both  before  and  after  the  

advent  of  digital  electronic  computers,  then  Agar  is  right,  and  computers  should  not  be  credited  with  making  the  

investigations  possible.  But  if  the  availability  of  the  digital  electronic  computer  is  what  made  particular  fission  

investigations  practicable,  then  Galison  is  right,  and  computers  can  be  credited  with  making  the  bomb  work  possible.  



Isaac  Record   12  

Either  way,  it  was  actually  carrying  out  these  investigations  that  made  crucial  knowledge  epistemically  possible.  The  

question  is  whether  a  change  in  technology  played  the  deciding  role.  

I  turn  to  technological  possibility  and  the  relation  between  technology  and  action  in  Part  2.  

Part  2.  Technological  Possibility  

In  the  first  part,  I  argued  for  an  account  of  epistemic  possibility  that  can  accommodate  the  epistemic  responsibilities  that  

can  require  a  subject  to  take  action.  At  the  same  time,  my  account  is  sensitive  to  practical  limits  to  fulfilling  those  

responsibilities.  A  given  subject  cannot  undertake  just  any  investigation.  She  will  be  competent  to  perform  only  some  

investigations,  and  her  technological,  economic,  and  ethical  circumstances  will  allow  still  fewer.  Impracticable  

responsibilities  put  limits  on  what  knowledge  is  within  a  subject’s  epistemic  grasp.  Changes  to  situational  constraints  on  

practicability  can  change  what  is  epistemically  possible  for  a  subject.  In  the  present  part,  I  focus  on  technological  

possibilities  as  a  hard  constraint  on  practicability  and  therefore  epistemic  possibility.  Technological  possibility  depends  

on  availability  of  the  material  and  conceptual  resources  required  to  complete  some  action  or  produce  a  desired  state  of  

affairs.  Thus:  

(g)  A  course  of  action  is  technologically  possible  for  a  subject  S  if  S  has  access  to  both  
the  material  and  conceptual  means  to  accomplish  it.5  

The  possibility  of  my  spanning  a  river  with  an  iron  bridge  turns  on  both  what  the  world  is  like  (i.e.,  that  iron  is  available  to  

me  and  has  certain  properties,  and  that  I  have  certain  capabilities  with  respect  to  iron)  and  how  my  concepts  fit  together  

(i.e.,  that  I  think  iron  has  certain  properties  that  I  can  put  to  use  in  making  trusses).  Without  the  material  means,  the  

bridge  would  fail.  Without  the  conceptual  means,  I  would  never  attempt  it.  Given  this  definition,  the  connection  

between  technological  possibility  and  practicability  is  clear:  a  subject’s  technological  tools  are  a  determinant  of  what  is  

practicable.  An  action  is  only  practicable  if  it  is  technologically  possible.  

There  are  two  ways  to  rule  something  technologically  impossible  for  a  given  subject:  either  the  subject  doesn’t  have  

access  to  the  material  means  of  accomplishing  it,  or  the  subject  doesn’t  have  access  to  the  conceptual  means  of  

accomplishing  it.  The  burden  in  assessing  whether  a  course  of  action  is  technologically  possible  is  in  making  a  sensible  

determination  as  to  which  conceptual  and  material  means  should  be  considered  ‘accessible’  to  a  subject  given  the  

peculiarities  of  her  situation.  A  course  of  action  that  would  exhaust  a  subject’s  material  resources,  tax  her  creative  

                                                
5  The  phrase  “possible  for”  makes  my  notion  of  technological  possiblity  subject-­‐relative  (see  Gibbs  [1970]  for  an  analysis  
of  “possible  for”).  We  sometimes  speak  of  technological  possibilities  in  a  more  abstract  sense.  For  example,  there  is  a  
sense  in  which  it  is  technologically  possible  (for  some  unspecified  someone)  to  travel  to  the  moon.  I  have  no  principled  
objection  to  this  usage,  but  in  the  present  paper  I  am  concerned  with  cases  in  which  S  is  specified—and  locked  to  a  
particular  time,  space,  and  circumstance.    

  



Isaac  Record   13  

faculties,  and  take  a  long  time  to  construct  would  be  difficult,  but  nevertheless  accessible.  Note  that  in  the  case  of  

complex  investigations  like  scientific  experiments,  the  most  challenging  aspect  of  completing  an  inquiry  is  often  in  

determining  whether  the  equipment  has  functioned  properly,  whether  the  desired  intervention  actually  occurred,  or  

whether  a  particular  inference  is  actually  warranted  by  the  data.  All  of  these  tasks  should  be  included  in  the  calculus  of  

the  technological  possibility  of  the  inquiry  as  a  whole.  In  the  following,  I  will  consider  how  to  draw  the  line  between  

accessible  and  inaccessible  material  and  conceptual  means.  

Let  me  begin  with  material  means.  I  said  above  that  the  possibility  of  my  spanning  a  river  with  an  iron  bridge  depends  on  

my  having  access  to  iron,  iron  having  certain  properties,  and  my  having  certain  capabilities.  This  suggests  a  way  to  divide  

material  means  into  three  further  considerations.  The  first,  access  to  the  material  itself,  is  simply  a  logistical  

consideration  that  depends  on  a  subject’s  situation—roughly,  what  a  subject  has,  or  can  beg,  borrow,  or  steal.  The  other  

two,  the  properties  of  that  material  and  the  subject’s  capabilities,  ultimately  rest  on  physical  possibility.  

(h)  A  state  of  affairs  is  physically  possible  if  it  is  not  precluded  by  the  laws  of  nature.  

Physical  possibility  is  about  the  world,  not  our  ideas  of  it;  that  is,  physical  possibility  is  not  subject  relative.  Physical  

possibility  is  about  what  is  possible  given  the  actual  laws  of  nature,  not  our  account  of  those  laws.  This  means,  for  

example,  that  fusion  experiments  have  been  physically  possible  for  billions  of  years  (light  from  stars  billions  of  light  years  

away  substantiates  this).  By  contrast,  technological  possibility  is  about  what  is  possible  for  a  particular  person  (or  group  

of  persons)  in  a  particular  context:  it  takes  a  technological  advance  like  the  construction  of  the  Tokomak  fusion  reactors  

in  the  1950s  and  60s  for  scientists  to  perform  fusion  experiments.  Physical  possibility  is  a  hard  constraint  on  

technological  possibility  because  technologies  cannot  subvert  the  laws  of  nature,  and  neither  can  users  of  technology.  

That  is,  my  ability  to  build  an  iron  bridge  depends  on  the  physical  possibility  of  iron  taking  the  form  of  a  bridge  and  on  

the  physical  possibility  of  my  body  (and  other  available  technology)  giving  it  that  form.  Physical  possibility  is  a  necessary  

condition  for  technological  possibility,  but  physical  possibility  is  not  sufficient  for  technological  possibility.  There  are  

many  actions  that  are  physically  possible  and  yet  beyond  the  means  of  a  particular  subject  to  bring  about,  even  if  the  

subject  has  access  to  the  requisite  material.  

Let  me  be  clear  about  the  relationship  between  physical  possibility  and  a  subject’s  capabilities.  I  have  just  claimed  that  a  

subject’s  capabilities  are  ‘ultimately  limited’  by  physical  possibility,  where  physical  possibility  is  to  be  understood  in  

terms  of  the  laws  of  nature.  This  means  that  physical  possibility  is  sensitive  to  physiological  limits  in  addition  to  

mechanical  ones.  To  give  a  concrete  example,  suppose  that  my  physiology  is  such  that  I  cannot  leap  tall  buildings  in  a  

single  bound  (unaided),  but  I  can  easily  jump  a  hurdle.  But  what  about  edge  cases?  Suppose  that  I  cannot  presently  dunk  

a  basketball,  but  that  with  enough  conditioning,  I  could.  Then,  strictly  speaking,  dunking  is  an  accessible  course  of  action  

for  me.  However,  if  I  am  presently  considering  whether  to  dunk  this  basketball  or  just  lay  it  in,  then  I  had  better  lay  it  in,  

because  my  time-­‐sensitive  context  does  not  afford  the  months  of  training  and  conditioning  necessary  for  me  to  dunk  

today.  Evaluating  technological  possibility  requires  specifying  a  context  more  or  less  precisely,  and  this  situated  analysis  

is  often  helpful  in  identifying  the  relevant  limiting  factors,  which  may  include  time,  money,  material,  equipment,  or  skills.  



Isaac  Record   14  

The  second  basic  component  of  technological  possibility  is  a  subject’s  access  to  conceptual  means.  Here,  the  relevant  

hard  limit  is  epistemic  possibility.6  Anything  that  is  epistemically  impossible  for  subject  S  is  also,  necessarily,  

technologically  impossible  for  her.  In  other  words,  considerations  of  epistemic  possibility  allow  us  to  deem  

technologically  impossible  those  courses  of  action  that  a  subject  knows  she  cannot  accomplish.  In  cases  in  which  a  

subject’s  reasons  for  ruling  out  a  course  of  action  have  to  do  with  physical  considerations,  as  in  the  basketball  example  

above,  this  seems  redundant.  But  in  other  cases,  a  subject  may  know  that  a  course  of  action  costs  too  much  or  is  

unethical  or  is  simply  beyond  her  cognitive  means.  In  such  cases,  a  subject  can  rule  out  the  course  of  action  on  the  basis  

of  its  epistemic  impossibility.  We  can  also  rule  out  those  courses  a  subject  has  a  responsibility  to  determine  that  she  

cannot  accomplish.  These  courses  of  action  are  precisely  the  ones  that  are  conceptually  unavailable.    

There  are  two  ways  for  a  subject  to  be  mistaken  about  availability.  Either  she  thinks  a  course  of  action  is  not  available  

when  it  is  or  she  thinks  it  is  available  when  it  is  not.  For  example,  there  may  well  be  courses  of  action  that,  because  of  

mistaken  beliefs,  seem  conceptually  unavailable.  While  a  subject  would  probably  not  attempt  a  course  of  action  that  she  

thinks  is  unavailable,  it  is  nonetheless  the  case  that  such  a  task  is  technologically  possible  for  her  if  what  ruled  it  out  was  

a  mistaken  belief.  Indeed,  only  those  tasks  that  she  knows  to  be  unavailable  (or  is  required  to  find  out  are  unavailable)  

are  genuine  technological  impossibilities.  The  converse  mistake  is  for  a  subject  to  fail  to  fulfill  a  responsibility  to  find  out  

that  a  course  of  action  is  impossible.  In  such  a  case,  a  subject  may  believe  a  course  of  action  is  available  to  her,  when  in  

fact  she  should  have  (for  example)  made  some  investigation  that  would  have  eliminated  that  candidate  action.  For  

example,  before  embarking  on  a  laboratory  experiment,  a  scientist  might  be  expected  to  perform  a  back-­‐of-­‐the-­‐

envelope  calculation  to  ensure  that  the  experiment  could  have  the  expected  result.  If  she  fails  to  do  the  calculation  (or  

gets  the  wrong  answer),  she  might  attempt  the  experiment.  But  this  does  not  change  the  fact  that  it  is  technologically  

impossible  for  her,  and  it  does  not  change  the  fact  that  she  should  have  known  it  was  technologically  impossible  

(because  it  was  epistemically  impossible).  

Epistemic  possibility  is  a  necessary  condition  for  technological  possibility,  but  not  a  sufficient  condition  for  it.  That  is,  just  

because  an  action  is  epistemically  possible  doesn’t  make  it  technologically  possible.  Physically  impossible  actions  that  I  

do  not  know  are  impossible  are  epistemically  possible,  but  not  technologically  possible.  In  other  words,  physical  and  

epistemic  possiblity  are  both  necessary  conditions  for  technological  possibility.  

Let  me  now  address  four  potential  concerns  about  the  role  epistemic  possibility  plays  in  determining  technological  

possibility.  First,  I  concluded  the  earlier  part  of  the  paper  with  the  claim  that  technological  possibility  is  an  enabling  

condition  for  epistemic  possibility.  Now  I  have  proposed  that  epistemic  possibility  is  an  enabling  condition  for  

                                                
6  I  owe  a  considerable  debt  to  an  anonymous  reviewer  for  urging  me  to  reconsider  an  earlier,  wrongheaded  approach  to  
analyzing  conceptual  means.  



Isaac  Record   15  

technological  possibility.  This  raises  the  spectre  of  circularity.7  Second,  “available  conceptual  means”  depends  in  large  

part  on  the  conceptual  analysis  a  subject  performs.  It  might  not  seem  like  epistemic  possibility  is  the  concept  to  capture  

this.  Third,  we  might  be  concerned  about  the  role  community  standards  play  in  specifying  responsibilities.  For  example,  if  

a  subject  moves  between  epistemic  communities  with  different  expectations,  a  course  of  action  may  switch,  seemingly  

willy-­‐nilly,  between  being  technologically  possible  and  impossible.  Finally,  it  might  be  objected  that  epistemic  possibility  

doesn’t  correctly  capture  what  it  means  for  the  conceptual  means  for  a  course  of  action  to  be  “available.”  

The  first  potential  concern  results  from  a  conflation  of  the  epistemic  possibilities  regarding  the  completion  of  a  task  and  

those  based  on  the  completion  of  a  task.  For  example,  “it  is  possible  to  reveal  surface  details  of  the  moon  through  

telescopic  investigation”  is  a  quite  different  epistemic  possibility  than  “it  is  possible  that  the  moon’s  surface  is  smooth.”  

The  line  between  epistemically  possible  and  impossible  claims  about  the  geology  of  the  moon  has  the  potential  to  shift  

with  the  advent  of  any  number  of  technological  advances,  including  the  telescope  and  space  travel.  The  claim  that  “the  

moon’s  surface  is  smooth”  was  epistemically  possible  (and  indeed  widely  believed)  before  Galileo’s  telescopic  

investigations  showed  surface  features.  Galileo’s  instrument  provided  new  evidence  that  could  make  a  difference  in  

determining  what  propositions  about  the  moon’s  surface  were  epistemically  possible  for  various  subjects,  and  in  

addition  the  use  of  the  telescope  was  available  as  a  candidate  responsibility  for  some  subjects  who  wanted  to  make  

claims  about  the  surface  of  the  moon.  (Exactly  who  was  responsible  for  what  depended  in  part  upon  the  expectations  of  

the  relevant  community.)  In  short,  however,  the  epistemic  possiblity  that  “telescopes  can  reveal  distant  features”  

enables  the  technological  possibility  of  viewing  lunar  surface  details  through  a  telescope,  which  in  turn  makes  the  claim  

that  “the  surface  of  the  moon  is  smooth”  epistemically  impossible.  To  put  the  point  more  generally,  an  epistemic  

possibility  regarding  the  completion  of  a  task  enables  a  subject  to  pursue  it,  while  the  line  between  epistemic  

possibilities  and  impossibilities  can  shift  based  on  the  completion  of  that  task.  

The  second  worry  inquires  into  the  relationship  between  epistemic  possibility  and  conceptual  analysis.  For  example,  in  

the  telescope  example  just  described,  I  noted  that  the  epistemic  possibility  of  using  a  telescope  to  investigate  the  

surface  features  of  the  moon  has  to  do  with  “supposed  properties”  of  glass  and  brass.  Conceptual  analysis  means  

determining  the  compossibility  of  a  proposed  state  of  affairs  with  a  particular  background  context.  This  may  seem  

orthogonal  to  epistemic  possibility.  But  as  Hacking  says,  we  bring  logic  into  the  fold  when  the  “terms  of  individuation”  

produce  a  contradiction  (1975,  333),  and  this  allows  us  to  rule  out  contradictory  situations  on  the  grounds  that  they  are  

epistemically  impossible.  Let  me  be  clear  about  how  this  works.  Whether  a  conceptual  incompatibility  exists  can  change  

depending  on  the  level  of  detail  we  give  to  the  terms  we  use  to  pick  out  a  situation.  In  considering  whether  I  could  leap  

tall  buildings,  I  might  at  first  neglect  to  take  into  account  some  relevant  details,  such  as  what  I  know  of  the  laws  of  

physics,  and  on  the  basis  of  that  incomplete  picture  judge  the  deed  possible.  But  if  I  carried  on  filling  in  details,  says  

                                                
7  My  thanks  to  an  anonymous  reviewer  for  bringing  this  concern  to  my  attention.  



Isaac  Record   16  

Hartshorne,  I  would  wind  up  in  perfect  agreement  with  physical  possibility—in  the  end,  the  two  are  indistinguishable  

(see  Hartshorne  1963,  595).  This  contention  is  mistaken  for  two  reasons.  First,  it  makes  an  unwarranted  demand  on  

epistemic  responsibilities,  and  second,  it  conflates  physical  possibility  with  scientific  theories.  

To  better  illustrate  Hartshorne’s  contention,  we  can  draw  on  George  Seddon’s  example  of  how  the  relevant  analysis  

should  work.  An  iron  bar  that  floats  on  water  has  been  supposed  by  some  philosophers  to  be  conceivable,8  but  physically  

impossible.  (“Bar,”  clarifies  Seddon,  is  meant  to  rule  out  needles,  which  float  on  surface  tension,  and  the  Queen  Mary,  

which  floats  on  “Zurich  capital”  [1972,  483].)  Since  it  is  physically  impossible  for  an  iron  bar  to  float  on  water,  filling  in  

our  concepts  with  more  information  about  what  it  is  to  be  water  and  what  it  is  to  be  iron  and  what  it  is  to  float  will  lead  

to  just  the  sort  of  self-­‐contradiction  that  would  allow  it  to  be  ruled  epistemically  impossible  on  conceptual  grounds.  But  

for  someone  ignorant  of  the  latest  scientific  theories  and  without  practical  experience  with  the  relevant  materials,  there  

is  no  such  additional  information  to  fill  in  the  concepts.  It  may  well  be  conceivable  to  her  for  an  iron  bar  to  float  on  water  

because  there  is  nothing  inconsistent  in  the  concepts  she  has.  Assuming  she  has  no  epistemic  duties  requiring  her  to  

investigate  further,  floating  iron  bars  are  epistemically  possible  for  such  a  subject.  On  the  other  hand,  for  anyone  with  

relevant  common  experience  or  a  passing  acquaintance  with  our  best  scientific  theories,  it  is  conceptually  impossible  for  

iron  to  float  on  water.  Furthermore,  iron  floating  on  water  would  be  epistemically  impossible  for  anyone  with  a  

countervailing  epistemic  duty.  

The  third  potential  anxiety  about  the  role  of  epistemic  possibility  in  technological  possibility  is  that  differing  community  

standards  would  seem  to  make  courses  of  action  switch  haphazardly  between  being  epistemically  possible  and  

impossible,  and  therefore  between  being  technologically  possible  and  impossible.  According  to  my  account,  different  

community  expectations  can  result  in  the  same  action  being  technologically  possible  in  by  one  community’s  standard  

and  impossible  by  another’s.  But  the  difference  is  not  haphazard.  One  community  may  require  that  a  subject  take  action  

that  would  rule  a  course  of  action  epistemically  impossible  (and  therefore  technologically  impossible  as  well),  while  

another  community  has  no  such  requirement.  The  concern  is  that  the  specified  course  of  action  would  actually  be  

impossible  for  the  subject  to  carry  out  in  either  case,  and  my  definition  of  technological  possibility  doesn’t  correctly  

reflect  this  because  it  gives  different  answers  for  the  two  communities.  But  remember,  the  argument  isn’t  that  a  subject  

can  actually  accomplish  every  technologically  possible  course  of  action.  It’s  that  a  subject  cannot  accomplish  any  course  

of  action  that  is  technologically  impossible.  

The  fourth  concern  is  related  to  the  third,  but  is  more  general.  It  might  be  objected  that  epistemic  possibility  doesn’t  

correctly  capture  what  it  means  for  the  conceptual  means  for  a  course  of  action  to  be  “available.”  At  the  outset  of  the  

discussion  of  access  to  conceptual  means,  I  stated  that  one  of  the  ways  to  be  mistaken  about  availability  is  to  mistakenly  

consider  a  course  of  action  available  when  it  is  not.  Similarly,  there  may  be  cases  in  which  a  subject  does  not  know  that  a  

                                                
8  Seddon  actually  discusses  logical  possibility,  so  I  have  made  some  minor  adjustments.  



Isaac  Record   17  

course  of  action  is  unavailable,  nor  will  she  have  a  responsibility  to  rule  it  out.  In  such  a  case,  the  course  of  action  will  be  

epistemically  possible  for  S  (even  if  she  doesn’t  explicitly  think  the  course  of  action  is  available).  If  such  a  course  of  action  

is  also  physically  possible,  then  it  will  be  technologically  possible,  even  though  it  could  never  actually  be  accomplished.  

This  seems  to  suggest  that  epistemic  possibility  is  the  wrong  measure  to  determine  whether  a  conceptual  means  is  

available.  If  so,  it  would  appear  that  we  are  left  with  three  alternatives  (besides  starting  over).  First,  we  could  deny  that  

cases  like  the  one  I  just  constructed  actually  exist.  But  I  have  no  sturdy  basis  for  making  such  an  argument.9  Second,  we  

could  try  to  shore  up  technological  possibility  by  adding  some  additional  condition,  but  I  have  no  suggestions  as  to  what  

that  condition  should  look  like.  Third  (and  this  is  the  option  I  prefer),  we  can  accept  that  physical  possibility  and  

epistemic  possibility  are  not  quite  jointly  sufficient  for  technological  possibility  after  all.  Even  so,  technological  possibility  

is  a  useful  and  principled  means  of  drawing  a  hard  line  between  practicable  and  impracticable  actions,  because  it  is  still  a  

necessary  condition  on  practicability.  Put  another  way,  my  account  admits  as  technologically  possible  some  courses  of  

action  that  we  might  wish  it  deemed  impossible.  But  it  deems  no  course  of  action  impossible  that  we  should  wish  it  to  

deem  possible.  

*   *   *  

I  began  this  paper  by  considering  conflicting  claims  about  what  difference  technology  makes  in  the  practice  of  science.  

My  diagnosis  is  that  such  conflicts  can  be  understood  by  analyzing  differences  in  implicit  assumptions  about  possibility.  

That  is,  we  should  understand  the  conflict  between  Galison  and  Agar  as  stemming  from  imprecise  expressions  of  what  

was  possible.  Peter  Galison  observes  that  “some  kind  of  numerical  modelling  was  necessary  [for  completing  fission  bomb  

work],  and  here  nothing  could  replace  the  prototype  computer  just  coming  into  operation  in  late  1945:  the  ENIAC”  

(Galison  1996,  122),  while  John  Agar  argues  that  “computerization  was  usually  first  proposed  when  the  existing  practices  

and  technologies  were  still  capable  of  the  computational  task  at  hand”  (Agar  2006,  873).  

Let’s  put  these  claims  into  the  language  of  epistemic  and  technological  possibility.  According  to  Galison  and  Agar,  

Manhattan  Project  scientists  considered  three  approaches  to  the  problem:  they  could  perform  fission  experiments  to  

learn  crucial  facts,  they  could  solve  difficult  analytic  equations,  or  they  could  perform  a  brute-­‐force  numerical  attack  on  

the  bomb  equations.  Any  of  these  three  approaches  could  satisfy  their  epistemic  responsibilities.10  The  question  was  

whether  they  were  practicable,  and  this  was  a  matter  of  some  debate.  Fission  experiments  were  considered  

                                                
9  It  might  be  objected  that  what  makes  a  conceptual  means  “too  complex”  to  be  conceivable  necessarily  has  to  do  with  
brain  structures,  and  therefore  could  be  ruled  out  on  grounds  of  physical  impossibility.  But  this  is  merely  a  plausible  
retort,  not  a  definitive  one.  
10  The  third  approach,  numerical  attack,  particularly  when  it  took  the  form  of  Monte  Carlo  calculation,  required  some  
advocacy  before  it  was  considered  acceptable  throughout  the  scientific  community,  but  it  was  almost  immediately  
accepted  within  the  Manhattan  Project  itself,  perhaps  in  part  because  of  enduring  frustrations  with  the  other  two  
approaches—an  interesting  story  in  its  own  right,  but  beyond  the  scope  of  the  present  paper.  Galison  (1996)  is  a  good  
starting  point  for  that  story.  



Isaac  Record   18  

impracticable  given  the  particular  time  constraints,  economic  pressures,  and  allowable  risks  associated  with  the  

Manhattan  Project.  Dozens  of  the  world’s  best  mathematical  minds  were  basically  stumped  by  the  intractable  analytic  

equations,  so  that  approach  also  appeared  impracticable  (whether  it  was  or  not).  And  finally,  it  was  not  at  all  clear  that  

there  was  time  enough  to  run  a  numerical  brute  force  attack  on  the  bomb  equations.  What  is  now  clear  is  that,  before  

the  digital  electronic  computer,  numerical  analysis  had  not  been  successful,  but  with  the  computer,  such  methods  were  

successful.  The  question  is  whether  it  was  the  computer  that  made  the  difference.  

Could  Monte  Carlo  calculations  have  been  performed  using  existing  computational  methods?  A  human  computer  could  

follow  any  of  the  instructions  ENIAC  performs,  and  the  Manhattan  Project  employed  many  such  computers.  But  a  human  

would  do  the  job  much  more  slowly,  so  there  is  some  question  as  to  whether  the  calculations  could  be  completed  within  

the  required  time  constraints.  Early  computer  literature  is  full  of  direct  comparisons  between  human  and  digital  

electronic  computers.  A  typical  example  is  that  ENIAC  could  perform  a  particular  calculation  in  60  milliseconds  (or  30  

seconds  if  the  result  was  to  be  printed  out).  It  would  take  an  individual  human  computer  7  hours  to  solve  the  same  

problem  (von  Neumann  1961,  9).  ENIAC  did  not  make  the  individual  calculations  physically  possible—they  always  were  

physically  possible.  Nor  did  ENIAC  make  the  calculations  epistemically  possible—scientists  had  quite  specific  calculation  

methods  in  mind  well  before  ENIAC  came  along  (indeed,  the  bulk  of  Agar’s  account  goes  to  substantiate  this  claim:  the  

Monte  Carlo  method  was  known  to  mathematicians  decades  before  the  first  computer  was  constructed).  What  ENIAC  

provided  was  faster  calculation—by  many  orders  of  magnitude  as  compared  to  individual  human  computers,  and  by  

smaller  orders  as  compared  to  other  existing  computational  methods.  

Before  ENIAC,  “a  single  hydrodynamics  problem  in  an  implosion  simulation  required  passing  a  deck  of  punched  cards  

through  a  dozen  machines,”  a  process  requiring  a  full  month,  even  after  Richard  Feynman  and  his  computing  group  cut  

the  process  by  two  thirds  by  devising  a  parallel  computing  method,  and  even  with  the  machines  running  24  hours  a  day  

(Seidel  1998,  34).  By  contrast,  von  Neumann  estimated  that  “one  criticality  problem  requires  following  100  primary  

neutrons  through  100  collisions  (of  the  primary  neutron  or  its  descendants)  per  primary  neutron,”  which,  computed  

using  the  Monte  Carlo  method  on  ENIAC,  “should  take  about  5  hours”  (Richtmyer  and  von  Neumann  1947,  752).  These  

aren’t  identical  calculations,  but  the  examples  give  a  sense  of  the  dramatic  change  in  speed  offered  by  the  Monte  Carlo  

method  on  the  ENIAC  as  compared  with  prior  computational  approaches.  

It  is  on  this  significant  difference  in  speed  that  the  debate  finally  turns.  For  Galison  to  be  right,  we  must  accept  that  

ENIAC’s  faster  computations  moved  this  particular  application  of  the  Monte  Carlo  method  from  the  “too  slow  to  

consider”  category  into  the  “can't  rule  it  out”  column.  For  Agar  to  be  right,  we  must  accept  that  scientists  could  have  

implemented  this  application  of  Monte  Carlo  using  older  methods  of  calculation.  This  remains  a  matter  of  debate,  but  it  

is  a  much  more  precise  debate  than  the  one  we  started  with.  I  tend  to  side  with  Galison  in  this  particular  case,  because  if  

Manhattan  Project  scientists  had  considered  implementing  Monte  Carlo  using  traditional  methods,  their  back-­‐of-­‐

envelope  estimations  would  not  (in  my  estimation)  have  suggested  any  advantage  over  their  current  approaches.  It  is  

only  with  the  considerable  improvement  in  speed  that  came  with  ENIAC  that  the  problem  could  be  solved  in  practical  



Isaac  Record   19  

time.  It  is  worth  noting  that  if  the  computational  problem  were  even  larger—say,  fusion  bombs  rather  than  fission—the  

case  would  be  even  stronger.  

It  is  also  worth  noting  that  once  the  method  was  suggested  to  him  by  Ulam,  von  Neumann  immediately  thought  to  

implement  it  using  ENIAC  (see  Richtmyer  and  von  Neumann  1947,  751-­‐52).  Perhaps  this  is  a  case  of  the  “Birmingham  

screwdriver”—with  a  hammer  in  hand,  everything  looks  like  a  nail.  And  when  faster  computers  are  available,  more  

problems  begin  to  look  susceptible  to  a  numerical  approach.  But  this  is  to  suggest  a  psychological  mechanism  by  which  

the  scientists  became  cognizant  of  the  fact  that  Monte  Carlo  was  a  promising  approach—it  is  not  to  say  that  Monte  Carlo  

was  inconceivable  (or  unconceived  of)  before  the  computer.  It  is  clear  that  scientists  had  the  conceptual  resources  

necessary  before  the  advent  of  the  digital  electronic  computer—no  matter  what  role  the  computer  might  have  had  in  

reminding  them  of  this  possible  solution.  

*   *   *  

I  began  this  paper  by  considering  conflicting  claims  about  what  difference  technological  change  makes  to  the  pursuit  of  

knowledge.  My  diagnosis  was  that  the  conflict  is  due  to  differences  in  implicit  assumptions  about  possibility.  In  the  first  

part,  I  argued  for  the  inclusion  of  practicable  responsibilities  in  the  analysis  of  epistemic  possibility.  In  the  second  part,  I  

introduced  technological  possibility,  which  depends  on  access  to  the  material  and  conceptual  means  of  bringing  about  a  

desired  state  of  affairs,  as  one  constraint  on  practicability,  making  technological  possibility  a  necessary  but  insufficient  

condition  for  epistemic  possibility.  

Acknowledgments  

I  would  like  to  thank  the  audience  at  the  2011  meeting  of  the  Society  for  the  Philosophy  of  Science  in  Practice  for  

comments  on  portions  of  this  paper.  I  also  owe  debts  to  Anjan  Chakravartty,  Boaz  Miller,  Eleanor  Louson,  an  anonymous  

reviewer,  and  especially  Greg  Lusk  for  their  comments  and  suggestions  that  have  substantially  improved  the  paper.  This  

paper  was  completed  while  I  was  a  Postdoctoral  Research  Fellow  at  the  Faculty  of  Information,  University  of  Toronto.  

References  
Agar,  J.  (2006).  What  Difference  Did  Computers  Make?  Social  Studies  of  Science  36(6),  869-­‐907.  

Conee,  E.  and  Feldman,  R.  (2004).  Evidentialism:  Essays  in  Epistemology.  Oxford:  Clarendon  Press.  

DeRose,  K.  (1991).  Epistemic  Possibilities.  The  Philosophical  Review  100(4),  581-­‐605.  

Egan,  A.,  Hawthorne,  J.,  and  Weatherson,  B.  (2005).  Epistemic  Modals  in  Context.  In  G.  Preyer  and  G.  Peter  (Eds.),  
Contextualism  in  Philosophy  (pp.  131-­‐69).  Oxford:  Oxford  University  Press.  

Egan,  A.  (2007).  Epistemic  Modals,  Relativism  and  Assertion.  Philosophical  Studies  133(1),  1-­‐22.  

Foley,  R.  (2003).  Justified  Belief  is  Responsible  Belief.  In  M.  Steup  and  E.  Sosa  (Eds.),  Contemporary  Debates  in  
Epistemology.  Available  online:  philosophy.fas.nyu.edu/docs/IO/1161/justifiedasresponsible.pdf.  

Galison,  P.  (1996).  Computer  Simulations  and  the  Trading  Zone.  In  P.  Galison  and  D.J.  Stump  (Eds.),  The  Disunity  of  
Science:  Boundaries,  Contexts,  and  Power  (pp.  118-­‐57).  Stanford,  CA:  Stanford  University  Press.  



Isaac  Record   20  

Gendler,  T.S.  and  Hawthorne,  J.  (Eds.).  (2002).  Conceivability  and  Possibility.  Oxford:  Clarendon  Press.  

Gibbs,  B.  (1970).  Real  Possibility.  American  Philosophical  Quarterly  7(4),  340-­‐48.  

Hacking,  I.  (1967).  Possibility.  The  Philosophical  Review  76(2),  143-­‐68.  

Hacking,  I.  (1975).  All  Kinds  of  Possibility.  The  Philosophical  Review  84(3),  321-­‐37.  

Hartshorne,  C.  (1963).  Real  Possibility.  The  Journal  of  Philosophy  60(21),  593-­‐605.  

Huemer,  M.  (2007).  Epistemic  Possibility.  Synthese  156(1),  119-­‐142.  

Kornblith,  H.  (1983).  Justified  Belief  and  Epistemically  Responsible  Action.  The  Philosophical  Review  92(1),  33-­‐48.  

Longino,  H.  (1990).  Science  as  Social  Knowledge.  Princeton,  NJ:  Princeton  University  Press.  

MacFarlane,  J.  (2011).  Epistemic  Modals  are  Assessment-­‐Sensitive.  In  A.  Egan  and  B.  Weatherson  (Eds.)  Epistemic  
Modality  (pp.  144-­‐178).  Oxford:  Oxford  University  Press.  

Moore,  G.E.  (1962).  Commonplace  Book,  1919–53.  Allen  and  Unwin,  London.  

Richtmyer,  R.D.  and  von  Neumann,  J.  (1947).  Statistical  Methods  in  Neutron  Diffusion.  In  A.H.  Taub  (Ed.)  John  von  
Neumann:  Collected  Works  (pp.  751-­‐64).  Oxford:  Pergamon  Press.  

Seddon,  G.  (1972).  Logical  Possibility.  Mind  81(324),  481-­‐94.  

Seidel,  R.W.  (1998).  ‘Crunching  Numbers’  Computers  and  Physical  Research  in  the  AEC  Laboratories.  History  and  
Technology  15,  31-­‐68.  

Stanley,  J.  (2005).  Fallibilism  and  concessive  knowledge  attributions.  Analysis  65(2),  126-­‐31.  

Teller,  P.  (1972).  Epistemic  Possibility.  Philosophia  2(4),  303-­‐20.  

von  Fintel,  K.  and  Gillies,  A.S.  (2011).  ‘Might’  Made  Right.  In  A.  Egan  and  B.  Weatherson  (Eds.)  Epistemic  Modality  (pp.  
108-­‐130).  Oxford:  Oxford  University  Press.  

von  Neumann,  J.  (1961).  On  the  Principles  of  Large  Scale  Computing  Machines.  In  A.H.  Taub  (Ed.)  John  von  Neumann:  
Collected  Works  Volume  V  Design  of  Computers,  Theory  of  Automata  and  Numerical  Analysis  (pp.  1-­‐32).  
Pergamon  Press:  Oxford.  

Weatherson,  B.  and  Egan,  A.  (2011).  Epistemic  Modals  and  Epistemic  Modality.  In  A.  Egan  and  B.  Weatherson  (Eds.)  
Epistemic  Modality  (pp.  1-­‐18).  Oxford:  Oxford  University  Press.  

Williams,  M.  (2001).  Problems  of  Knowledge:  a  Critical  Introduction  to  Epistemology.  Oxford:  Oxford  University  Press.  
Yalcin,  S.  (2011).  Nonfactualism  About  Epistemic  Modality.  In  A.  Egan  and  B.  Weatherson  (Eds.)  Epistemic  Modality  (pp.  

295-­‐332).  Oxford:  Oxford  University  Press.