Skip to content

Why the Mary Argument Fails

WHY THE MARY ARGUMENT FAILS                Colin Mitchell     June 2009


Abstract:  I argue that the Mary argument (and Knowledge arguments in general) do not succeed in demonstrating that qualia (and conscious experiences in general) cannot be physical facts.  Then I investigate the source of our intuition that qualia cannot be physical facts.  By means of a thought experiment I show that the hypothesis that qualia are identical with brain processes is a reasonable one.  


The Mary argument is the best known form of what is generally known as the Knowledge argument.  The Knowledge argument asserts that full knowledge of the physical facts of brain processes is insufficient to enable us to know what it is like to experience qualia (or conscious experiences in general) and that therefore qualia (and conscious experiences in general) cannot be physical facts (or at least not the physical facts of brain processes).  Qualia are the ‘quality’ of mental states or perceptions as they are immediately perceived by the individual having them – the ‘what it is like’ of experiences – particularly applied to sensory experiences.


The Mary argument was proposed by Frank Jackson in an article in 1982 under the heading “The Knowledge Argument for Qualia” (Jackson 1982).    Mary is a neuroscientist specialising in the neurophysiology of vision and knows all there is to know about the physical facts involved in the perception of colour.  Yet she herself has lived all her life in a black and white room and has never seen colour.  When she is released from the room she learns something new – ‘what it is like’ to perceive colour.  Therefore this extra knowledge – the qualia of colour – cannot be physical facts.


Here is the Mary argument in the form of two premises and a conclusion:


Premise 1:  Mary has full knowledge of all the facts about the physical world, including full knowledge of the workings of her own brain.


Premise 2:  When Mary experiences the colour red for the first time she learns a new fact.


Conclusion:  The experience of the colour red is not a physical fact.


A fully explanatory account can be given of what happens to Mary assuming identity between qualia and neural processes in Mary’s brain.  This in itself does not do the work of refuting the Mary argument because it assumes that which the Mary argument claims to show is impossible (identity between qualia and physical brain processes).  However it is useful as a guide to point towards possible flaws in the Mary argument.  Under this identity account we see that Mary indeed has a new experience when she sees the colour red for the first time but that this is entirely compatible with a purely physicalist account of Mary’s mind. (I use the term physicalism instead of materialism but materialism can be exchanged with it).


Here is an account of what happens to Mary using the identity theory of mind:




Mary’s brain holds conceptual knowledge of ‘facts’ via particular type/s of specific neural interactions that are probably distributed through different areas of the brain but may be more or less localised in an area of the brain.  Whether or not they are localised the neural interactions are of a specific type or types involved in concept formation and storage.     

Other type /s of neural interactions are concerned with registering and interpreting visual sensory information, in particular the colour red.  Probably different nerve fibres are involved altogether and there may be a different location in the brain for this activity (but if not the type and manner of interaction is different to the concept-forming activity.) 


Mary’s conceptual knowledge of ‘facts’ all involves activity in the conceptual knowledge complex of the brain.  This knowledge can include full details of the facts about nerve fibres and their interactions in her own brain (there may be an issue about the physical capacity of Mary’s brain to hold all these facts but for the sake of the argument assume that she can).  The conceptual knowledge can include the facts about all the neural events that happen in the visual information complex of the brain when appropriate nerve signals come in from the eyes.  Mary also knows all the facts about how objects reflect different wavelengths of light and that these result in different neural events in the visual reception area of the brain.  She also knows that these different wavelengths of light are ascribed their various colour names.  She is able to predict exactly the physical events and changes that will occur to the appropriate neurons in the future if light of a particular frequency (red) should stimulate her retina.  Mary knows all this via activity in the conceptual knowledge complex of her brain. 


However the actual neural activity that results when red light strikes the eyes causing nerve signals to travel to the visual information area of the brain does not occur until those nerve signals actually do come in from the eyes.  Until then the nerve fibres concerned do not fire (or at least the appropriate pattern of firing does not occur).  Until they do fire Mary does not have the qualia experience of the colour red.  Now when they do fire Mary has a new experience.  If the word Knowledge is taken to embrace both conceptual and experiential knowledge then Mary now gains new knowledge.  She gains knowledge both in terms of the direct experience itself (the qualia) and in terms of any extra concept she now forms as a result of the experience.  If the word Fact is taken to embrace experiential facts then Mary now learns a new experiential ‘fact’ (as well as any additional concepts formed about that fact).


Regardless of word usage, Mary certainly does learn “what it is like” to have an experience of the colour red when the appropriate nerves fire in the appropriate way.  Until they do so she cannot.  Of course, if Mary manages somehow to get those nerves to fire in the appropriate way without impulses coming in from the eyes, then yes, she would have the experience of the colour red in advance of seeing with her eyes.  Likewise perhaps she could simulate the appropriate type and pattern of firing in another part of the brain, using other nerves, and get an experience of the colour red that way via some sort of creative imagination. 


Yet all of the above takes place in a physical brain in which the experience of the colour red simply is the appropriate pattern of firing of the appropriate nerves.  The above description gives a full account of what happens to Mary in purely physical terms, assuming identity between brain processes and mental activity including the experience of qualia.




Now I turn to the Mary argument itself.


Here is the Mary argument again in the form of 2 premises and a conclusion:


Premise 1:  Mary has full knowledge of all the facts about the physical world, including full knowledge of the workings of her own brain.


Premise 2:  When Mary experiences the colour red for the first time she learns a new fact.


Conclusion:  The experience of the colour red is not a physical fact.


Looking at premise 1 we see that it is fundamentally an epistemological statement.  It asserts that Mary has knowledge.  It asserts nothing directly about ontology.  Knowledge (as it is usually understood) about things is not the things in themselves.  Let’s take “facts about the physical world” as the things that exist (or events that actually occur).  The first premise asserts only that Mary has knowledge about those things.  Knowledge is not identical with the objects of knowledge.


Looking at premise 2 we see that this is also an epistemological statement.  It asserts that Mary experiences something and she learns something ie she gains new knowledge.  In the usual understanding of knowledge the new knowledge she has gained would be about a fact.  It would not be the fact in itself.  In this case, for the purposes of the Mary argument, we can understand “fact” as something (or events) that exist, whether physical or non-physical.  The Mary argument in total wants to demonstrate that the new fact is non-physical, but premise 2 itself is an epistemological statement referring to Mary’s knowledge about something.  It is not an ontological statement.


The conclusion is an ontological statement.  It asserts that the new knowledge Mary has gained is not itself a physical fact.  We have jumped from statements about epistemology to a statement about ontology.  But there is a distinction usually made between epistemology and ontology.  Knowledge about things is not the things in themselves.  If that is the case then we would think that we cannot draw conclusions about ontology from merely epistemological statements. 


However we also notice that for the first time in the argument the conclusion makes self-reference to what could be considered a type of knowledge; “experience”.  The ontology in the conclusion refers to the experience itself.  If experience is to be counted as a type of knowledge then we do not necessarily have the separation between knowledge and ontology that is usually assumed.  For example, I said in commenting on premise 2 above, that under this assumption the new knowledge Mary gained was about a new fact, it was not the fact in itself.  But if the sensory experience itself is to be included in the category knowledge then the “fact” she ‘learns’ can be considered to be the experiential knowledge itself. 

The whole point of the Mary argument is to comment on the ontology of experience (qualia) so if experience is to be included in the term knowledge we have to admit an ontology of knowledge.


Certainly in the Identity theory of mind ontology and knowledge fuse because knowledge, whether conceptual or experiential, simply is identical with physical neural processes.  And here in the Mary argument a qualia experience of the colour red is being included within the category ‘knowledge’, and it is also being called a ‘fact’ (although we don’t know whether it is a physical or non-physical fact).  If ‘fact’ is confined to the ontological category then here the Mary argument is saying that in this instance (concerning a qualia of the colour red, but presumably all sensory experience) knowledge is identical with ontology (however the Mary argument concludes it is non-physical ontology).


We need to know what is to be included in the category ‘knowledge’ and what is to be included in the category ‘facts’ in order to evaluate the argument.  What is included in these categories should remain consistent throughout the argument in order for the argument to work.


The Identity account I gave of what happens to Mary suggests that we should divide the term knowledge into ‘conceptual knowledge’ and ‘experiential knowledge’ (where in this case experiential knowledge is sensory experience).  Conceptual knowledge is the type of knowledge where there is a distance between the knowledge itself and the facts it relates to.  In Mary’s case she has conceptual knowledge of physical facts about the world including her own physical brain.  Later on she gains some experiential knowledge as well, and in this case the knowledge is itself the fact (whether physical or non-physical).

In addition Mary may now form new concepts about the experience.


In the case of experiential knowledge, items of knowledge become identical with items of ‘fact’.  Conceptual knowledge is separate from the facts it relates to and experiential knowledge is identical with the ‘facts’ concerned.


Likewise we have to distinguish 2 types of learning: conceptual learning and experiential learning.


We can now paraphrase the Mary argument this way:


Premise 1:  Mary has conceptual knowledge of (all) physical facts.


Premise 2:  Mary gains experiential knowledge of a new experiential fact.


Conclusion:  The experiential fact is not a physical fact.


In premise 1 Mary holds concepts which represent in some way physical facts.  These physical facts are not the concepts themselves – the concepts represent the facts.


In premise 2 Mary has a direct experience which is not a concept.  The direct experience is itself the experiential ‘fact’.  We must add the rider that as a result of this experience Mary could now form new concepts about the experience.  These new concepts would be representations about the new experiential ‘fact’.  They would not be the experiential fact itself.


Does the conclusion follow from the premises? 


The type of knowledge and type of fact has changed from Premise 1 to Premise 2.

The new experiential fact is known by experiential knowledge – a direct experience which is not the same way of knowing as knowing by concepts.  Therefore the new fact could be one of the physical facts from Premise 1 known in a new way. 


Premise 1 makes the claim that Mary’s (conceptual) knowledge covers the entirety of physical facts.  There are no physical facts left not covered by Mary’s knowledge.  In principle these could include knowledge of all past events and predictions of all future events, or at least knowledge of exactly what would happen physically in Mary’s brain if she should see something with the colour red.  However when these physical facts do actually happen Mary knows them in a new way – as a direct experience.  Thus the experiential fact (qualia of the colour red) could be those very physical facts.  The conclusion of the Mary argument does not follow from the premises.


One possible response could be to deny that knowledge can be divided into two types.  Perhaps there is only one type of knowledge.  If so, it would have to correspond with what we have been calling conceptual knowledge.  A property of conceptual knowledge is that it is not the things that it represents.  Premise 2 claims that Mary “learns a new fact”.  Perhaps this is the same as gaining conceptual knowledge of a new fact (and there is only one type of learning – conceptual learning).  We can still call the new fact an “experiential fact” (which has an unknown ontology) – this is simply the qualia – the “what it is like” to experience the colour red.  But assume that this is known conceptually.  If desired we can even drop the label conceptual and just use the term knowledge, since there is now only one type of knowledge.


The argument is now:


Premise 1:  Mary has (conceptual) knowledge of all physical facts.


Premise 2:  Mary gains (conceptual) knowledge of a new experiential fact.


Conclusion:  The experiential fact is not a physical fact.


As before premise 1 states that Mary knows the entirety of physical facts.  She can make predictions about future physical facts and knows exactly what will happen to her neurons when she sees the colour red.  Premise 1 appears to leave no room for any additional physical facts.


In premise 2 Mary gains knowledge of a new fact.  But there would appear to be no room for any additional physical facts, and the physical facts from premise 1 cannot now be known in a new way.  So it would appear the new experiential fact cannot be a physical fact.


However we have left out one vital ingredient – time.  What happens to Mary takes place over time, as is made explicit at the beginning of premise 2 (in its original form) with the word “when”.  We saw from the Identity account, that until the neurons concerned with colour vision fired Mary did not have a direct experience of the colour red.  This was regardless of the amount of concepts Mary held in the conceptualising part of her brain.  The event of the firing of those neurons does not take place until it takes place.  The experiential ‘fact’ that takes place does not occur until the event of the firing of those neurons.  Mary could in principle predict that it will take place but as we have already seen – conceptual knowledge about something is not the thing in itself – in this case it is not the actual event, it is a representation of that possible future event.


Using time and facts alone I can show that there is room for more physical facts when Mary experiences the colour red.  This is because physical facts include physical events.


When the experiential event takes place the store of facts (of whatever kind) in the world increases.  It increases with an experiential fact, an event.  The set of facts has increased.

Included in these new facts are physical facts, physical events in neurons.  Mary had predicted these would occur and now has additional knowledge – they have occurred. Even though Mary could predict that the event would occur, and therefore had a representation of that event as a concept, this was not the actual event.  When the event occurs the set of event facts has increased by one (or many if we consider all the events involved in colour perception).  Since the set of facts has increased there is room for that many more physical facts.  When Premise 1 says that Mary has knowledge of all physical facts it can only mean all physical facts as at the time concerned (plus predictions of future events).


Here is the Mary argument paraphrased with time included:


Premise 1:  At time T1 :  Mary has (conceptual) knowledge of all physical facts as at T1 (plus predictions of future events).


Premise 2:  At time T2 :  Mary has (conceptual) knowledge of a new event (the experiential event, qualia of the colour red).  The number of physical facts in the world has increased because there have been new firings of neurons.  These facts are not identical with Mary’s concepts of these new firings at time T1. (Mary’s concepts are about the events but are not the events themselves). 


The conclusion “experiential facts are not physical facts” (or in the original form “the experience of the colour red is not a physical fact”) is no longer tenable.  It is entirely possible that the new experience is new physical facts – ie the new neuronal events themselves.  It is also possible that it is not.  No definite conclusion can be drawn about the ontology of the new experience.


Combining both the idea of two types of knowledge and the time factor: there is a disconnection between conceptual knowledge of events and those events themselves.  On the other hand the experiential knowledge Mary gains in premise 2 is the event itself (plus Mary may form extra concepts resulting from the event).  Mary’s experiential (and possibly conceptual) knowledge has increased but so has the set of physical facts increased, in the form of new neuronal events, so it is quite possible that Mary’s new experience is those extra physical facts themselves.


I suggest a very reasonable hypothesis is that the new experience is identical with the new physical facts.


David Chalmers believes that the strategy a physicalist must take to the Mary argument is to deny that Mary gains any new knowledge about the world when she sees the colour red (Chalmers 1996).  He rejects the idea that knowing an old fact in a new way (eg through experiential knowledge rather than conceptual knowledge) is not learning a new fact:


“It follows that if Mary gains any factual knowledge that she previously lacked – even if it is only knowledge of an old fact under a different mode of presentation – then there must be some truly novel fact that she gains knowledge of.  In particular, she must come to know a new fact involving that mode of presentation.  Given that she already knew all the physical facts, it follows that materialism is false.  The physical facts are in no sense exhaustive.” (Chalmers 1996). 


In my argument I do not deny that Mary learns a new experiential fact plus any new conceptual facts that go along with it.  I accept that Mary gains new knowledge.  What I contest is that Mary’s conceptual knowledge of all the physical facts at a certain time T1 precludes the coming into being of new physical facts (physical events) at a later time T2.

Even if Mary has predicted these events at T1, when the events occur at T2 she gains new knowledge – of the occurrence of the events.  Hence the new novel facts she learns, including the qualia of the colour red, can be physical facts.  Mary’s knowledge of all the physical facts is exhaustive at the time she has the knowledge.  At a later time there can be new physical facts in the world.


Chalmers is wrong that a physicalist has to deny that Mary makes any new discovery about the world.  For the very reason that the world moves on and is continually accumulating new physical events Mary should be continually discovering new facts.





The Mary argument is one form of the Knowledge argument.  All the forms purport to show the impossibility of a physical identity for qualia, the “what it is like” of conscious experience.  The Knowledge argument uses knowledge of the physical facts of brain processes to exclude the experience of qualia from the physical fold.  Regardless of the validity of the argument itself, it can carry intuitive force.  As Frank Jackson, the originator of the Mary argument, says in the article in which he first proposed the argument “…the polemical strength of the Knowledge argument is that it is so hard to deny the central claim that one can have all the physical information without having all the information there is to have.” (Jackson 1982).  The extra information there is to have is the experience of qualia. 


Jackson’s statement is a little misleading.  As I argued above, it is not a matter of just having information in the form of conceptual knowledge, qualia experience is a different kind of ‘direct’ information (or, equivalently, a different way of knowing information)  and it is not possible to have all the information until the event/s of qualia occur.  However there is certainly a strong intuition that qualia cannot be identified with physical events.  I think the source of that intuition is a disconnect between 2 mental pictures, usually visual.  On the one hand we introspect and experience what seeing the colour red is like.  On the other we have a visual mental picture of a tangle of neurons in our brains and wonder how any activity of that could possibly be identical with the beauty and subtlety of our experience of the red in the sunset.  Of course, we don’t only have to imagine the neurons, we can actually look at pictures of them or see the real things through a microscope and measure their activity.  They don’t look anything like our experience of the colour red.


In this section I investigate the source of our intuitive mistake.  The basic error is in comparing qualia from 2 different objects and expecting them to look the same if the Identity hypothesis is correct.  But they shouldn’t look the same because the qualia are caused by different objects.  Instead what happens according to the Identity hypothesis is that the qualia caused by one object are identical with activity in a second object and this activity in the second object is in turn the object for a second set of qualia.


Here is a thought experiment which simply extrapolates into the future the capabilities of neuroscience.  We start our investigation with qualia itself:


Janet is a scientist who experiences qualia of the colour red from a red object in her laboratory.  Call this qualia Q1.  Call the red object RO.  She has already hypothesised, from her experience, that the qualia of red she experiences is caused by the presence of the red object in her laboratory.  She hypothesises this from the correlations between the presence of the red object in front of her eyes and her experience of the colour red.  This is simply the inference we all build up from when we are babies: that there is an external world causing our experiences.  Everyone regards this as a reasonable hypothesis under the circumstances.


Now Janet wants to investigate what is happening in her brain when she sees the red object.  Fortunately technology has advanced to the level where there is a device which can peer into Janet’s brain and display, visually in real time, the electrical activity in the neurons concerned with visual processing.  Janet discovers that every time she experiences qualia of the colour red there is particular activity in particular neurons in her brain, and this occurs concurrently with her experience of the colour red.  She has to look at a screen to see the activity of these neurons (which let us say displays a picture of the neurons as well as showing their activity and giving a time display), or alternatively imagine that she can see the neurons by direct light as under a microscope.  Of course there is a problem with experiencing the colour red and at the same time observing the neurons, but imagine that this can be overcome by a measured time delay on the screen from the recording device, so that Janet knows there is a correlation between her experience of the colour red and the neuronal activity. 


Call the qualia Janet has from observing the activity of her neurons Q2.  This is a different experience from Q1 (which comes from the red object).  Janet hypothesises that Q2 is caused by activity in the neurons, just as Q1 is caused by the red object.  Call the activity in the neurons NA.  Now Janet makes a further reasonable hypothesis: that Q1 is identical with the activity in the neurons: Q1 = NA.  Note that Q1 is not identical with Q2, they are different qualia caused by different objects.


Why is Janet’s hypothesis reasonable?  Because the red object (RO) causes Q1.  And particular neural activity (NA) causes Q2.  But Q2 is correlated with Q1.  So the particular neural activity NA is correlated with Q1.  Hence the hypothesis that Q1 is the particular neural activity.


It is not the only hypothesis that could be made.  Others are

1. The red object causes both neural activity and Q1 separately    (RO causes NA and Q1)

2. The red object causes neural activity which causes Q1         (RO causes NA causes Q1)

3. The red object causes Q1 which causes neural activity    (RO causes Q1 causes NA)

However Ockham’s Razor may reasonably be used to cut out the extra lengths in the causal chain to produce the simpler Identity hypothesis.


Hypothesis 2 could correspond with epiphenomenalism if rephrased as ‘the red object causes neural activity which is accompanied by Q1’ to avoid the word ‘cause’ for the second relationship.  Likewise Ockham’s Razor may reasonably be used to reduce the 2 entities of epiphenomenalism, Q1 and NA, to one.  This single entity could now be described in two ways, as Q1 or NA, because Q1=NA.


Janet’s reasoning can be summarised as:

Q1 is caused by RO

Q2 is caused by NA

Q1 is correlated with Q2

Therefore NA is correlated with Q1 (and RO)

Reasonable hypothesis:  NA = Q1


We do not expect that our observations of the neural activity should look like the qualia of the colour red.  The qualia of the colour red is caused by a red object in the laboratory, whereas the qualia of the screen display of neural activity is caused by the nerves in Janet’s brain. 


Qualia are not identical with the objects of perception (they are caused by them).  They are not even good representations of the objects of perception.  Unfortunately we think that our sensory qualia are accurate ‘pictures’ of what is really out there.  They are not.  They are imperfect representations caused by objects in our environment.  They are accurate to only a very limited extent but do not reproduce the true nature of objects.  For example the qualia of the colour red is caused by light of a particular frequency range from an object.  The red colour we experience is only a representation of that frequency range and tells us nothing directly about the wave nature of light or its electromagnetic nature.  We cannot discern the atoms that compose the red object either.  But we do know that the experience of the colour red is associated with whether we are seeing that particular object, so we can say it is caused by the presence of that object in front of our eyes. 


Likewise the qualia we have when we look at nerves (or as we imagine Janet is able to do – the activity in the nerves) is not the objects themselves or the activity itself.  It is a representation caused by that activity, and a very imperfect representation of what is actually going on at the molecular or atomic level in those structures.  What we have with qualia is a causal story.  We should not expect that we automatically know the true nature of the objects of our perception, through perception itself.  


Now if we turn our attention to the qualia experience itself should we expect that we know its true nature?  If qualia are at bottom representations caused by objects stimulating our senses, then how are we to discern the true nature of those representations?  We would need further qualia caused by those qualia, to peer into the nature of the first.  These further qualia would also be representations – representations of the representations making up the first qualia.  We should not expect that we automatically know the true nature of our representations just from the experience of the first qualia. 


But the former situation is exactly Janet’s experiment.  She used qualia of nerve activity to peer into the nature of her qualia of the colour red.  She observed a correlation.  This did not give her a direct perception of the nature of her qualia of the colour red.  This gave her a representation of the elements and structure (actually events) of her representation of the colour red.  The one is not more real than the other.  They are both different representations of an underlying reality.  But they enable Janet to make a causal correlation between Q1 and Q2.  Not posit identity between Q1 and Q2, far from it – they are different representations and under the Identity theory of mind they are identical with very different neural events.  Instead Janet hypothesises identity between the object of Q2 and Q1 itself, identity between whatever is being imperfectly represented by Q2 and Q1 itself.


We cannot expect that we, as representing machines, be able to directly discern the elements of our own representations.  Instead we need further representations to represent our representations.  This does enable us to make correlations however.  And from those correlations we can make the very reasonable hypothesis that our experiences, our representations, are identical with what we observe when we observe particular brain processes.  This was Janet’s reasonable hypothesis. 




Chalmers D. J.  ‘The Conscious Mind’  (1996)  New York, Oxford  Oxford University Press  p145 & p144


Ibid  p142


Jackson, F. ‘Epiphenomenal Qualia’ in Philosophical Quarterly. V 32 (1982) pp.127-36


Ibid  p130



CV:   Colin Mitchell is a BA graduate in English Literature who has studied Philosophy and Physics at AdelaideUniversity, FlindersUniversity and Melbourne University, Australia.





Could a Machine have a Mind?

COULD A MACHINE HAVE A MIND?                Colin Mitchell


Could a human-made machine constructed with non-biological material have a mind? (I define mind to include all capabilities of the human brain including qualia and consciousness).  Many would agree that if the functional operations of the human brain could be duplicated with non-biological material then this creation would have a mind.  Such a project may use connectionist computational architecture in the form of artificial-neural-networks.  This would be multiple formal systems in parallel and distributed operation.  Since these operations are effective, by the Church-Turing Thesis, they are register machine computable. (1)


Yet there are those like John Searle who deny that the operations of a formal system alone, could have understanding or consciousness.  These people believe that there must be something more happening in the brain to produce understanding/meaning/qualia/consciousness. 


Searle characterises this ‘something’ as “causal-powers”, by which he means the ability of the brain to cause “intentionality” in mental states. (2)  For Searle intentionality is content which is directed at the external world. (3)  For Searle also “understanding” implies intentionality. (4)  So he is simply saying that ‘something’ adds meaning and understanding to certain mental states but we don’t know what that is. 


Others like Frank Jackson deny that qualia/consciousness could be purely physical.  David Chalmers also rejects a reductive physical account of consciousness in terms of already known physical entities.  In this article I argue that the basis for these denials is inappropriately switching between an objective and a subjective viewpoint, and that a machine could have a mind solely on the basis of the operations of a formal system, as posited by Computationalism (Computationalism is the view that the mind is a kind of formal system in operation). (5)


John Searle’s Chinese Room (CR) thought experiment is designed to show that a mind which “understands” could not consist only in the operation of purely formal systems. (6)  He shows that the operations of his CR as he describes it produce no “understanding” of the symbols manipulated.  His test for “understanding” is whether a human either sitting in the room performing the operations or internalising the same operations in her mind, understands the symbols being manipulated.  These ‘symbols’ happen to be Chinese language characters.  The point of making the language Chinese is that the human understands only English.    I argue that the CR needs extra formal operations using another set of data to produce “understanding”. 


The human uses a subjective viewpoint to determine her understanding.  But she is a system that has undergone a process of learning the meaning of English through interaction with the external world.  Let us radically simplify things and hypothesise that the basis for understanding the meaning of a language-item is association between the language-item and other sense-data from the world.  Let us make the further hypothesis that the “understanding” simply is this association between 2 sets of data.  This is a radical simplification since actual understanding no doubt involves very many associations between very many items of data.  Nevertheless the point is that in order to produce understanding of Chinese in the CR (by this hypothesis of what understanding language is) another set of symbols is needed in the room – symbols representing data about the world that can be associated with the Chinese symbols – just as the human in the room has data from the world associated with English symbols. 


For the human this extra data has been acquired from the world through sense-organs.  In the case of the CR this data could be input into the room from outside and a formal learning program included so that this data could over time be matched to Chinese language-characters (extra books written), or it could be included ready-made into the room (as another ready-made set of books).  Likewise, I argue, a computer could either receive data from the world through artificial sense-organs and ‘learn’ language or have this data already built-in.  Either way understanding occurs, by my hypothesis of what understanding is.  The learning process itself is not a necessary condition for understanding, what is necessary is the ability to associate the 2 sets of data.


The human in the CR could in fact learn the meaning of Chinese with this extra set of data – in terms of the symbols of the extra data themselves.  Since this data is analogous to her own sense data, this is analogous to her own knowledge of the English language.  But more importantly such an enhanced CR can be regarded objectively by an outside observer as fulfilling the requirements for understanding language since, by this hypothesis, understanding language is the association between the 2 sets of data (which then, after further formal symbolic operations, results in appropriate output).  It would not be valid, I argue, to say that the human in the Chinese Room must understand Chinese in the same way that she understands the English language to demonstrate that the CR produces understanding, because the sense-data by which she understands the English language is a different set of data from that which is representing the sense-data for Chinese in the CR.  In addition Searle’s test demands that we switch from an objective perspective in which we look at what is going on in the Room, to a subjective perspective of the human in the room.  The two views are different and don’t have to coincide. 


Another way of looking at it is the ‘systems reply’ in which it is the whole system that has to be considered when judging understanding, not just part of the system.  Searle’s reply is that the whole system can be internalised in the mind of the human without the human understanding Chinese.  My reply is that it is only the CR system that is internalised that has to be considered when evaluating understanding within that CR system, not systems outside that CR system such as the human’s already established understanding systems in the rest of her mind.  Thus it is invalid, I argue, to use a human’s pre-established understanding systems to evaluate and judge the understanding of a different system. 


Searle and others insist that there must be a connection with the external world for “intentionality” and “reference” and therefore understanding.  This is a consequence of how they define intentionality and reference.  I have shown that it makes no difference to the internal processes where the data that is referred to by language items comes from.  The act of understanding should properly be regarded as a complex of internal associations.


Hilary Putnam’s “brains-in-a-vat” scenario purports to show the necessity of embodiment and connection with the external world for a mind. (7)  But there is no essential difference between our situation and Putnam’s “brains-in-vats.”  Our brains float in fluid in the vats of our skulls.  We receive electronic impulses from our afferent nerves and send them out through our efferent nerves.  The afferent impulses come from an automatic machine – the machinery of the external physical world.  The connecting plugs are our senses. The efferent impulses are sent out to the automatic machine of the external physical world.  Our experience is mediated and determined entirely by the operations of this machine.  


Scientists concur.  Physicists realise that we are modelling the physical world in our brains via our senses and that our physical concepts and descriptions are not in themselves identical with these entities.  But most physicists nevertheless think there is an actual reality.  Our constructed reality reflects the external to some extent but so does the reality of a vat-brain – its reality reflects the changing impulses from the machine.


Putnam’s contention that the hypothesis “we are vat-brains” is a self-refuting hypothesis is flawed.  The flaw is his assertion that part of the hypothesis is “we are not vat-brains in the image” (the “image” derived from the computer).  The complete hypothesis should be “we are vat-brains in the image and we are brains in actual vats.”  The constructed reality consists of electronic impulses and events in the brain which are physically real.


The Mary-argument (one version of the Knowledge-Argument) purports to show the difficulty of explaining qualia in purely physical terms. (8)  Yet a purely physical description can be given of what happens to Mary.  In terms of her brain she has a lot of conceptual knowledge “stored away” in neurons.   And when she gets out of her black-and-white room a different set of neurons is activated for the first time – colour perception.  Even if she has conceptual knowledge about the colour-perceiving neuron circuits the fact remains that they have not been activated till now.  Two sets of neurons – two types of knowledge.


That means that we can hypothesise the identity of the experience of colour-qualia with the activation of those particular neurons.   Even if we want to say that Mary magically has access to all possible conceptual knowledge (far beyond the capacity of any brain) it is still conceptual knowledge.  It is not experiential.


The ‘explanatory gap’ as regards qualia/consciousness, I argue, simply results from the non-identity of the observer and the observed.  We use consciousness/qualia to investigate everything in an objective manner (oberver).  That does not mean that we cannot give an objective physical account of qualia/consciousness (observed).  The fact that it is possible to give a physical account of what happens to Mary demonstrates this.  That means we can say that the difference between “objective” and “subjective” is simply that the physical qualia/consciousness events of different observers are different (they are at least different tokens if not different types).


I think this shows that an objective physical account of consciousness/qualia is possible.  There is no reason in principle that a reductive account of consciousness in terms of the operation of formal systems cannot be given.   David Chalmers characterises what he calls the “hard problem” of consciousness in the following way:

            Even if every behavioural and cognitive function relating to consciousness were explained, there would still remain a further mystery: Why is the performance of these functions accompanied by conscious experience?  (9)

But if we identify certain cognitive processes themselves with consciousness then the hard problem goes away.  The performance of those cognitive functions is conscious experience so it is not at all surprising that they are “accompanied” by conscious experience.


Under a physical account of consciousness the subjective perspective results from the hard fact that cognitive consciousness processes in different observers are not token-token identical.  And the difference between conceptual and experiential knowledge results from the hard fact that conceptual neuron processes are not type-type identical with experiential neuron processes.


The semantic fallacy in the Mary argument rests on switching between an objective perspective (conceptual knowledge) and a subjective perspective (the experience of qualia).  Instead of keeping these perspectives distinct they are both called “knowledge.”  Really they are different.  And an account of this difference can be given in purely physical terms.


We can speculate 3 types of solution to the problem of consciousness/qualia.

Firstly consciousness/qualia is beyond a physical explanation.  Secondly consciousness/qualia can be given a reductive physical explanation in terms of already known physics.  Thirdly, as proposed by Chalmers, consciousness/qualia is a new fundamental type of physical entity, supervening on the rest of physical reality.  But there is no fundamental entity yet known which is not part of the closed causal system of physics.  If Chalmer’s new entity operates outside of this system under new laws then this would make it less credible.


Chalmers hopes that using his theory we might be able to predict “….the precise structure of our conscious experience from physical processes in our brains.” (10)  But we would also be able to do this using a reductive account of consciousness/qualia in terms of already established physics.  Occam’s Razor should make us prefer an explanation in terms of already known entities and laws of physics.


As Chalmers points out, a range of kinds of computation might suffice for conscious experience, not just an identical simulation of brain processes. (11)   It appears possible that a human-made machine could have a mind and that it could do so using purely formal processes.


Colin Mitchell     







A Chinese Room that Learns Meaning



Summary:  I argue that John Searle’s  Chinese Room can learn the meaning of a written language by including an extra ‘language’ input of symbols  paralleling the acquisition of sense data in human beings.  This can be achieved purely by the operation of programs of computational processes operating on formally defined elements. 


John Searle’s thought experiment of the “Chinese Room” is intended by him to show that “meaning” (or semantic understanding) cannot be created simply by manipulation of symbols such as occurs during the operation of formal programs in a computer.  Note that the word “symbol” is strictly speaking inappropriate because a symbol already has a meaning since it is meant to ‘stand for’ something or represent something.  A better term would  perhaps be ‘character’ but since Searle used the word symbol I do so on the understanding that no intrinsic meaning is meant to be attached to them.


The fatal flaw in his Chinese Room is that the central processing unit (CPU), the human being sitting at the table manipulating the symbols, has not learnt Chinese.  So, of course, whereas the person knows English, he or she does not understand the meaning of Chinese. 


The meaning of a language has to be learnt by reference to the external world through sense data (or by cross-reference with a language already learnt by this process).

A child learns language by associating a symbol or sound with sensory experience of the external world.  For example the meaning of “round” (as a shape) can be learnt by hearing the sound or seeing the symbol in association with seeing round objects.  The symbol becomes associated with the particular experience associated with a class of viewed objects.  From these simple learnt associations can be built up the whole edifice of much more complicated meanings including abstract concepts expressed by language.  These may be constructed not only through sensory experience but through cross-reference with other already learnt meanings as well (already learnt items of language). 


Meaning is thus created by reference to an extra intervening “language” – the language of sense data.  More specifically, what I am saying is that meaning is this association between 2 sets of ‘symbols’ – language symbols and the symbols of sense data.  This extra language is missing in the Chinese Room.  If we put it in then meaning can be created – and yet all that is occurring is the manipulation of these symbols according to formal rules.


So to the Chinese Room we add an extra input – a language of sense-data which we will call the SD language.  This consists of symbols, no more.  It represents the data coming into the brain via nerve impulses from the senses (and processed into whatever other impulses by the brain).  These nerve impulses do not differ in basic nature from the nerve impulses which accompany seeing or hearing the symbols of language, or from the nerve impulses which originate from within the brain (as when we remember an item of language or remember a sensory experience or manipulate a concept).  In the Chinese Room we are replacing nerve impulses with written symbols.


Now when the Central Processing Unit (CPU), the person at the table, receives an input of Chinese characters they are accompanied by another input, the SD language, and the CPU associates these Chinese characters with the SD symbols.  The rules for doing this are related to the timing of reception.  Gradually, over time, the CPU learns which Chinese characters go with which SD symbols.  This process also involves sorting back over records and seeing which SD symbols consistently go with which Chinese characters, eliminating sweeping associations in favour of more specific associations and gradually honing in on the more specific association between the SD symbols and the Chinese words – this more specific association is the more precise meaning.  The whole process is conducted according to a set of rules called the “learning program.”


Now the CPU has a new book of rules it has created – rules which tell which associations are to be made between particular Chinese symbols and SD symbols.  The rules are more complicated than this because they tell of associations also between these primary associations.  For example the association between the concept of round shape and the concept of ‘going around’ something.  The associations between SD symbols standing for round objects and certain Chinese character/s standing for round shape are also linked to another set of associations between SD symbols representing the experience of running around a pole (plus other experiences of seeing merry go rounds etc) and Chinese characters expressing the act of going around something.  The book of meaning rules is very complicated and becomes more complicated and more specific as it grows.


Now when the CPU receives an input of Chinese characters he or she can refer to this new book of associations called the book of meaning.  From there he or she can decide which would be the appropriate meanings to respond with.  This act of decision itself can be made according to a set of rules.  There are certain responses that go with certain inputs.  These are also largely learnt, but we can speculate that there is a certain amount of freedom or creativity involved, perhaps through the introduction of a random event generator into the system which selects between several available options.  In any case the response is largely or completely determined by the meaning.  Certain meanings are appropriate responses to other meanings.


Once the response meaning is selected from the book of meaning then it is a simple matter to look up the appropriate Chinese character to express that response.


Even though human beings have to learn the meaning of language it would be entirely possible to build the meaning into the Chinese Room from the start so that it did not have to learn anything.  Simply put the book of meaning ready made into the room.


I am arguing that meaning and understanding are simply associations and that it is entirely possible for a computer to understand meaning just by the manipulation of symbols.  But it requires an extra language (the SD language) because what the meaning is is association between the language in question and this extra ‘language’ which in humans results from sensory experience.  In computers it could either result from sensory experience or be built in ready-made.


Searle may argue that the person at the desk (the CPU) would still not have the experience of understanding the meaning of Chinese because all he or she is doing is manipulating symbols without knowing what their English meaning is and without knowing what sensory experiences from his or her memory they are associated with.  That would be to put the cart of understanding before the horse of experience.  The CPU has not learnt the Chinese language in terms of the same SD ‘language’ that it has used to learn English.  Instead it is associating a different SD ‘language’ (written symbols on paper) with a different written language – Chinese.  One can hardly expect that it will be able to then associate Chinese characters with its other SD ‘language’ which came from the senses.  They are physically different systems (although running on the same principles).


 The trouble with using a human being in the Chinese Room, a human being that has already learnt English and already had a lifetime of sensory experience (and therefore has already learnt meaning in terms of the English language) is that we have put what we are trying to model inside our model!  This misleads us. 


What we are trying to do with the Chinese Room is to see whether we can duplicate what has already occurred in the human being.  I am arguing that what has already occurred in the human being is equivalent to the process described above which is a purely formal process of manipulation of symbols according to rules.  This process is the experience of understanding meaning which occurs in the human being.  The CPU in the Chinese Room has done this with the English language.  So this CPU itself is a miniature Chinese Room within the Chinese Room!  We cannot use this miniature Chinese Room to judge the effectiveness of our larger Chinese Room.  It is the larger Chinese Room as a whole which has to learn the meaning of Chinese.  (More accurately perhaps the CPU is an English Room in a Chinese Room, but, as I argue, they run along the same principles).


Searle would grant that if the person sitting at the desk had available sensory data to match the input of Chinese characters, such that he or she could learn the meaning of Chinese through experience, then this person would come to understand Chinese and could respond appropriately to Chinese input according to meaning.  I am saying put this data into the Chinese Room (or include it ready made).  This data comes in as a string of symbols.  These ‘symbols’ are information approximating information coming into our brains via electrical impulses from our senses.  We do not have actual objects coming into our brains from the external world, we have a representation of them in the form of electrical impulses.  Likewise in the Chinese Room we have a representation in the form of a string of symbols on paper.  The person at the desk now associates these symbols with the Chinese characters.  In the same way a human being learns the meaning of a language by associating the electrical impulses associated with the language characters (these electrical impulses correspond in the Chinese Room to symbols on paper) with the electrical impulses associated with sensory experience (these electrical impulses correspond in the Chinese Room to a different set of symbols on paper).


Once these new rules are established (associating symbols on paper with other symbols on paper or one lot of electrical impulses with another lot of electrical impulses) then the process of understanding meaning and responding appropriately to language input is a matter of symbol manipulation following the rules.


The fact that we are replacing a structure of nerves and electrical impulses with paper tape, drawn symbols and pencils does not alter the associations.  But it is wrong to judge what is happening in the whole Chinese Room by what is happening in part of the Chinese Room – the person sitting at the desk.  The reason is that what is happening within the brain of the person sitting at the desk is being modelled, or duplicated, by what is happening in the room as a whole, outside of his or her brain.

Therefore one cannot expect that the understanding of meaning taking place in the Chinese Room as a whole will be duplicated in a corresponding understanding within just a part of that room. 


The reason the person at the desk does not have the same comprehension of meaning of Chinese when he or she learns the meaning in terms of paper symbols as opposed to in terms of familiar sensory data is that the systems are different.  The person has had a whole lifetime of electrical impulses behind him or her – this has created a system of associations (meanings) within them.  This system is factually, in hard concrete terms, a different system to the one taking place in the Chinese Room.  It is the Chinese Room itself that has to learn meaning – not the person sitting at the desk who forms only one component of the room.  The person at the desk could and should be replaced by a robot to make the Chinese Room a consistent model.


Searle argues that whatever happens in the Chinese Room can be duplicated and internalised in imagination in the mind of the person at the desk.  He then argues that when this is done the person has no understanding of the meaning of Chinese.  If we do that with the above arrangement for the Chinese Room then this is still true.  However, I argue, this is to be expected and it is not a valid way of judging whether the Chinese Room produces meaning. 


The reason is that we are trying to judge an experience of meaning in one system or language with an experience of meaning taking place in another language entirely.  The languages do not connect.  The person’s brain uses a system of meaning established with totally different symbols (which happen to be made of electrical impulses) which have come from their senses.  When the person internalises the Chinese Room system of paper and pencil by visualisation the associations taking place in the Chinese Room do not match the associations which the person has learnt via a different ‘language’ entirely (the ‘language’ of impulses from the senses).  The person is re-interpreting what is happening in the Chinese Room in terms of their own inner ‘language’ of visualisation.  It is not going to match their own associations of meaning which have been built up using a different inner ‘language’.


The person has built up their understanding of the meaning of English using sense data which has come from the external world and its objects.  Now what they are attempting to do is interpret what is happening in the Chinese  Room in terms of this already established system by visualisation of symbols on paper being shuffled around.


Look at it this way: the person at the desk receives his or her SD language in the form of electrical impulses representing sense-data.  The Chinese Room receives its SD language as written symbols on paper.  But then because the human being operates through his or her own sensory system these symbols on paper then have to be translated into electrical impulse SD so that they can see them and follow the rules etc.  The electrical impulse SD of the written characters is not the same experience as the SD of the objects that he or she sees in the world when he or she is looking at the world outside.  He or she has already learnt to associate this external world SD with the English language and thus learnt the meaning of the English language.  Now instead of that external world SD from external world objects he or she is getting SD from written Chinese symbols.  The person can indeed learn the meaning of the Chinese language in terms of these written SD symbols but to expect him or her to be able to interpret the Chinese characters in terms of the SD electrical impulse language they have from the external world (as they have done in the case of English) is unreasonable.


In fact, I argue, the process by which the human being in the Chinese Room does indeed learn the meaning of the Chinese characters in terms of the written SD symbols parallels the process by which the human being has previously learnt the meaning of the English language before entering the Chinese Room.


The human being cannot be used to judge whether the Chinese Room produces meaning by ascertaining whether they come to understand Chinese in terms of their experience of the world or the English language.  The human being is only useful in the Chinese Room as an agent to follow the rules and execute the program (but then we might as well use a robot).




Colin Mitchell              





Reference:  Searle J. 1980  Minds, Brains and Programs  in The Behavioural and Brain Sciences (Cambridge: Cambridge University Press, 1980) vol 3. pp.417-24

Also in Introduction to Philosophy ed. J. Perry and M. Bratman, Oxford: OUP 1993.

ISBN: 0-19-506936-6

Hello world!

Welcome to! This is your very first post. Click the Edit link to modify or delete it, or start a new post. If you like, use this post to tell readers why you started this blog and what you plan to do with it.

Happy blogging!