AI rubs up against major issues in philosophy, esp philosophy of mind
and philosophy of language, but also to some extent ethics, esthetics,
logic, epistemology, metaphysics, etc.  Here we will focus only on
mind and language.


                    PHILOSOPHY OF LANGUAGE

Philosophy of language asks questions such as: how do words get their
meanings? What is meant by "meaning"? And so on. We saw a little of
this in an earlier lecture, in the work of Paul Grice. Part of what is
at issue here is called the problem of external reference: how do
words refer to things in the world?  This is a different question
from, say, what determines whether "he" refers to John or to Bill in

     John gave Bill a present; he smiled and said "thank you".

The referent of "he" is presumably Bill. But external reference
instead deals with how "John" comes to mean a particular person, or
how "present" comes to mean certain sorts of things, rather than with
what a particular word means in a particular sentence.

External reference has a long history, from John Stuart Mill to
Gottlob Frege to Bertrand Russell, Hilary Putnam, Saul Kripke, etc.

Frege (1848-1925) famously pointed out that a word or phrase often
conveys more than one type of meaning, and that this is essential. For
instance,

     The morning star is the evening star

is true because the planet Venus when seen in the morning was long ago
thought to be a star (and was given the name "morning star") and
similarly for Venus seen in the evening. People many centuries later
discovered that both "stars" are really the same thing as the planet
Venus. But the sentence above does not then simply say

    Venus is Venus

which is a logical tautology and totally uninformative, whereas the
original sentence actually tells us something new.

To help explain this, Frege spoke of the "sense" and the "referent" of
a word or phrase. The referent is the actual external object it stands
for (so both "morning star" and "evening star" have the same referent,
namely the planet Venus). The sense is the thought or idea that the
phrase expresses (e.g., "that thing we see in the morning"). Referents
are external entities, while senses are in the mind.

Similarly, 2+2 and 4 co-refer (both have referent 4); but 2+2 has the
sense of a sum, and 4 simply of itself.  That is why 2+2=4 is
nontrivial.

This was a big step forward. But it still leaves much unsettled. For
instance, how is it that we can use the name "Aristotle" to refer to
someone who lived thousands of years ago? We have no obvious "sense"
of Aristotle, unless via ideas such as "the man who taught Alexander
the Great; and much or even all of any such ideas could actually be
mistaken and not pick out anyone at all.  Kripke and Putnam (roughly
1970) both advocated what is now called the causal theory of
reference, in which names (and more general terms like gold and elm
tree) derive their meaning from those who first employed them in
spcific contexts, and later users simply borrow that usage even if
they don't know the details.

Russell (1872-1970) gave an account of the meaning of definite
descriptions: the such-and-such. One example he gave is "The present
King of France is bald." But France has no king at all. So the
definite description "the present king of France" refers to what? It
does not seem to connect to anyone at all, so what does is mean to ask
about a characteristic of someone who does not exist? His answer was
that the entire sentence has to be unpacked as a more complex entity
such as

1. there is a unique present king of France, K
 and
2. K is bald

Then the original sentence can be said to be false, since 1 is false.

While this has many advantages, it does not work for a sentence such
as

    Dr Watson wrote about the cases Sherlock Holmes solved.

since Watson and Holmes and the cases are all fictional; they do not
exist. Nor does it work to explain how we manage use "Aristotle"
effectively. 

Meinong (1853-1920) introduced the idea of nonexistent objects to try
to overcome the fictional cases. His work is sometimes made fun of,
but it has substantial aspects that relate to the idea of
intentionality, to which we now turn (in stages), since that is a
central theme in the philosophy of mind.


                    PHILOSOPHY OF MIND

What is the mind, and what is its relation to the body that it seems
to inhabit?

This question is at least as old as Aristotle (384-322BCE), who seemed
to regard mind as analogous to a pot made of clay -- the mind is the
shape or pattern or structure, and it is built of some material (e.g.,
clay).  This seems a surprisingly modern view, close to what would now
be called the computational or information-processing paradigm.

This was a departure from the view of Aristotle's teacher Plato
(428-348BCE) who held that all that we know of are faulty pale fuzzy
shadows of what is real: the perfect "ideal forms" that at best we can
only get glimpses of when we think at our best (such as in mathematics). 
A square is never something we actually see; all our "squares" are
poor imitations of a real square, the latter being ideal squares in
some unavailable realm of ideal things. This view is called idealism,
but that does not mean the same thing that we would intend today in
saying "you are such an idealist!  That can never work in the real
world!" Ideals are the actual things themselves, not our poor
conceptions of them; so one must be careful not to confuse Plato's
term "ideal" with our modern usages.

Plato's idealism is also not about mind vs matter so much as a view
about what is what, and what is our situation in what is what. He gave
a famous example of the cave: most of the time we are like people
living in a dark cave staring at shadows on the back wall, formed from
whatever light falls on it from the entrance which is to our
backs. The shadows include our own but also shadows of "real/ideal"
things outside the cave, and we have no conception of those things
except whatever very dim ideas we get from their shadows -- and we
don't even realize they are shadows.  We take them for the real thing.

Idealism thus is not so far from how modern science works: we have
learned that reality might be quite different from what we see: a
table seems quite solid, yet we now believe -- after much careful
study and use of equipment, etc (microscopes, chemical data, and so
on) -- that tables are made of vast numbers of particles far too small
to see by eye alone, and that in fact a table is ostly empty space,
those particles being concentrated in little bundles (molecules or
atoms). And even that is surely not the full story; we learn more and
more as we peer closer and closer and think harder and harder, and at
times find we were far off and have to rethink things.

It is also not so far from the views of Kant (1724-1804) -- who argued
that everything we know is filtered through our minds, and that puts a
barrier between us and reality; he called what we know/perceive/think
"phenomena" and what is actual reality "noumena". He insisted that we
can never find out anything about the latter since we can never get
outside our minds to the world as it is.  Knowing is itself a
view-from-within, it is an act done by and within a mind and so is one
step removed from anything else. Whatever we imagine we are thinking
about is in the world of phenonmena, not that of noumena.  Of course,
the mind itself is, presumably, a noumenal thing, but we cannot see it
as it really is either, for we only see what our minds filter to us,
and when the mind thinks about the mind, it can only do so with the
means available to it.  This is perhaps a bit stricter than Plato's
idealism, placing a complete impenetrable barrier between our thoughts
and reality. This is sometimes called German Idealism.

It is also reminiscent of the ARD -- seeing a red ball as black, and
not realizing it is red but behind a blue sheet of glass.  To Kant,
perhaps, we are all always like three-year-olds in a sense. But he was
not being pessimistic or critical of human reasoning -- he was stating
what he thought was a bare fact of what it is to think, whether one is
a brilliant genius or an ordinary person: to think is to exercise a
mind, which is to percieve things through whatever processes minds
take part in. And those processes "process" whatever data they get,
the outer data itself never comes through unprocessed and we cannot
know what it was before processing. Even if we suspect this -- even if
we "get" the ARD -- we still are locked there in a mind and with
whatever filters that comes with. He referred to "categories" that the
mind sets up, ways it interprets reality, as if these were impossible
to alter. He gave the example of geometry -- lines and points and so
on, with the axioms passed on from Euclid -- as an example of
something that has to be true in our inner world: our minds present
reality to us in a way that makes anything other than the laws of
Euclidean geometry impossible to imagine. Yet not long after that,
mathematicians began to imagine (and put to good use) non-Euclidean
geometries.

[Note that we have drifted a bit: from asking what is mind and how does
it relate to matter or body, we have moved to questions about our
knowledge of things, and about what is "real". We will return to the
original issues, soon.]

One might try to paraphase Kant as saying all is uncertain, we can
know nothing for sure, but even so we can make enormous progress
within the phenomenal realm, which for Kant included science, while
some greater reality remains out of reach -- even out of thought.
[This was -- perhaps unintentionally -- echoed in a famous statement
of Wittgenstein (1889-1951): "Whereof one cannot speak, thereof one
must remain silent" which someone paraphrased as "What we can't say
we can't say, and we can't whistle it either." Wittgenstein was saying
that some things are vague and poorly grasped, and it is a waste of
time to try to draw conclusions in such cases; but this may be more
along the lines of Plato than Kant.]

Above I paraphrased Kant thus: "To think is to exercise a mind." This
might call to mind an earlier philosopher: Descartes (1596-1650). He
argued that if we want to be really picky, we can doubt almost
anything. But there is one thing we can be absolutely sure of --
something we cannot sincerely doubt at all: that we are in fact
performing mental acts of some sort (cognizing, thinking, imagining).
For to doubt is to perform a mental act, and we cannot doubt and yet
doubt we are doing that doubting. From this he concluded that we exist
-- at least as entities that carry on mental acts; we are thinking
things. He put it this way:

   (Latin version):  Cogito ergo sum 

   (French version):  Je pense, donc je suis

He did not conclude from this that we can be certain we have bodies,
just that we can be sure we have (or are) minds, things that can
doubt. He did speculate on how minds and bodies might be related, but
regarded them as belonging to very different realms of things:
material substances that exist in space (if they exist at all), and
mental substances that do not have spatial extent (and perhaps exist
in something like an ideal Platonic or even theological realm). This
is called (Cartesian or substance) dualism: there are two
fundamentally different kinds of substances that have their own
separate realms.

Hume (1711-1776) gave a different twist to the doubt idea, and
proposed that all scientific or general knowledge is suspect, that we
can always imagine that any generalization we are inclined to believe
might in principle be mistaken, and there is no pure and perfect
method by which we can determine for absolute certain what is
true. This view is called "skepticism" and contains in it the problem
of induction (not to be confused with the proof method known as
mathematical induction): given that every instance of an X that we
have seen so far has (say) mass between one and two kilograms, why
does that justify concluding that all X's have mass in that range (or
even that most do)? 

Russell gives this striking example: a chicken is fed larger and
larger meals by the farmer, and thinks "Wow! This is great! I can't
wait to eat the wonderful meal I'll get tomorrow, and then the day
after that!" -- until the farmer decides the chicken is now fat enough
to slaughter.

And to this day no answer has been given that explains why we should
trust patterns we have seen as guides to ones we have not seen.

It was this Humean skepticism that Kant was trying to answer: yes, we
can accept skeptcism with respect to noumena (for which it is even
worse than Hume imagined) but with respect to phenomena Kant claimed
we can fare much better than Hume thought: the human mind organizes
oue perceptions and ideas in a way that makes induction work.

OK, what about mind then?  What is it, and what is it's relation to
matter, body, brain?

Brentano (1838-1917) proposed a partial answer: mind is characterized
by intentionality -- and here is another technical term that is easy
to misunderstand since it also has an everyday meaning that is not the
same thing. Brentano was getting at something that he claimed was
present in all mental states, and only in mental states: a
directedness toward something else. Love is directed toward the loved
being; fear toward what is feared; curiousity toward the puzzling
situation; awareness to toward whatever one is aware of; etc.
Intentionality means "being directed toward something" (but not
necessarily being so directed on purpose or by intention, to stave off
the all-to-easy misunderstanding -- although my own take on this is
that it is indeed purposeful, if properly understood).

What about the arrow on a street sign "pointed" toward a driving lane?
Doesn't it exhibit intentionality? No, Brentano says: the only way in
which the arrow is connected to the lane is by means of humans who
take it to have that meaning. The intentionality is in human minds,
and we come to regard it as in the world (arrow directed toward lane),
but that is only "derived" intentionality, based on our "original"
intentionality. So, on this view, mind is whatever has intentionality,
the ability to somehow project itself to outer things, but in an inner
way. Thus the lover feels a conection to the loved, even if nothing
actually crosses space from one to the other. Mind imagines (or makes)
connections, even when they don't exist outside the mind. This applies
as well to language: a mind imaginatively connects "my sweetheart" to
a person even if it stayed in the mind and goes no further than
that. And sometimes people share these connections, as when parents
beam down at their newborn and one says "Let's call her 'Sally' ".
(This view of how words acquire meanings came along with the causal
theory of Kripke and Putnam much later, but seems to have an ancestry
of sorts in Brentano.)

OK, maybe mind is simply whatever has inner directedness, the ability
to aim one part of itself at something else (at least in imagination).
But then, what are these imaginative mental actions, and what does it
take to be able to do them? Can material things, suitably organized
like brains, carry out such actions? This is the mind/body problem,
all over again, but with a Brentano-esque add-on. 
a tentative equation:

We can present this as a tentative equation:

     mind (aka intentionality) 
           =?= 
             suitably organized matter (e.g., as in a brain) 

Another philosopher -- Thomas Nagel (1937 - ) wrote an influential
essay ("What is it like to be a bat?") in which he provided a different
characterization of mind: a mind is something it is like something to
be. It would not be like anything at all to be a rock; nothing;
zip. But being a bat -- while perhaps extremely different from being a
human -- would be like *something*. It would have a feel, an
experiential quality, so-called "qualia". That is because bats are
conscious; or so Nagel supposes. And if they are not, then it would
not be like anything at all to be one. This provides a neat divide to
go alongside Brentano's: if it would be like something to be an X,
then X is the kind of thing that has consciousness (or has a mind).

All this is well and good, but where is the AI?

Hilary Putnam (1926- ) may have been the first to draw the
connection in a sharp way -- beyond the basic question "Can a machine
have a mind?" -- in his notion of "functionalism". This issue arose
out of the so-called "mind/brain identity thesis" (due to psychologist
Boring and philosphers Place, Feigl, and Smart).

The identity thesis says that not only is mind a physical phenomenon
in the brain, but that each and every mental state is a particular
kind of brain state: get these particular neurons firing at those
particular rates and bingo! -- you have a state of happiness, or a
state of puzzling about a calculus problem, etc. But this was
criticzed as follows: what if Martians have evolved differently from
us, maybe their brains don't have neurons but some other signalling
mechanisms -- does that mean they can't puzzle about calculus, or feel
happy? Or what if some of the neurons that are involved in your
happiness are ones that are involved in my sadness (but not my happy
states)? Why should the identical *hardware* have to be in place in
the same way, for the a particular mental state to occur? After all, a
Volkswagen and a Ferrari have very different engines, but both serve
to make the wheels turn and the car move.

Putnam proposed that it is not the precise hardware that matters, but
what it does, in analogy with an algorithm and its implementation as a
program. Many implementations exist that carry out the same
function (doing merge-sort on an array of numbers, say). So maybe many
different neural (or other) details of brain-architecture can carry
out the same function of imagining a connection (Brentano) or it's being
like something to be a whatever (Nagel), or just plain having a particular
kind of conscious experience. Then Martians can be conscious even with
brains very unlike ours -- and computers too (as long as they are
running an implementation of the "right" algorithm).

Well, this caused a storm. In particular, John Searle (1932 - ) in his
famous Chinese Room thought-experiment, tries to show that
computational states -- information processing -- on its own, cannot
constitute consciousness (or understanding). In it's barebones logical
form, his argument goes like this: programs are formal, mere
symbol-manipulation specifications with no meaning at all on their
own. And understanding (and consciousness) are full of meaning (e.g.,
Brentano's intentionality of directedness). So programs alone -- even
when being executed on a machine -- cannot in and of themselves create
meaning, hence not mind either. [Note that Searle has no beef with
minds being material in nature, as in brains. It is the formal
symbolic nature of programs that he says is not enough. Brains must
contribute something additional, perhaps something biological, that is
lacking in programs.]

Here is his argument in more colorful form: imagine a room in which a
person sits (Searle, for instance). There is a slot in one wall,
through which a Chinese speaker passes a sheet of paper with sentences
in Chinese. Searle knows no Chinese at all, so to him it is just
"squiggles and squoggles" on the paper. But he has a book with
instructions: it says, for instance:

"If you are given a piece of paper with the marks <#%$> then take a new
sheet and write the marks <*@%> and pass it back through the slot."

Searle does this, over and over, and the person outside the room sees
the room as able to engage in maningful dialog in excellent
Chinese. The book, of course, consists of instructions for how to
respond in Chinese to Chinese input. It does not translate the input
into English at all. It may be that the Chinese input means "I love
hamburgers with lots of ketchup" and the Chinese response means "Yes,
me too", even if perhaps Searle hates ketchup. Searle's claim then is
that

 (i)  neither the room itself nor anything or anyone in it understands
      Chinese 

 (ii) the room arrangement as a whole amounts to execution of a program
      of purely symbolic (read-write) instructions without meanings of
      their own  

(iii) the room arrangement simulates understanding, and very well, even
      though it has no real understanding

Here is one among the many (vociferous!) replies to Searle: the "room
system" as a whole may understand Chinese, even if Searle doesn't.
Searle is playing the role of CPU, at the machine code level, whereas
it is in the much more abstract activity of (say) high-level interpreter
that mental activity occurs, much as in the brain. No one supposes
that individual neural spikes constitute mind, but maybe large-scale
coordinated activity across the brain as a whole does.

This has been linked to the windmill parable of Leibniz (1646-1716):
suppose the mind were a machine. Imagine it very large, like a
windmill, so one can enter inside and inspect its parts. The parts,
being mechanical, are each in themselves incapable of anything other
than simple mechanical action: motion. No consciousness there, in any
of them, no "I". But then where? The "I" is a unitary thing, it cannot
be partly here, partly there. One's full experience of a given
moment -- the emotions, thoughts, perceptions -- all come together as
one total experience held by "me", not scattered about. But nothing
made of simple parts can bring everything together as a unitary thing.

Is this related to Brentano's directedness, perhaps, coming from a
single "I" who does the directing?  Or shades of Perry's supermarket
experience?  Neuroscientists have coined the phrase "binding problem"
for a related issue: how does the brain link the color, motion, shape,
sound, etc of a bluebird flying past, as all belonging to the same
external entity?  Where do all these bits of information (that are
known to be processed in different areas of the brain) come together
so we can say "Ah, see the bluebird flying by!"?

Frank Jackson (1943-) discusses another thought-experiment: Mary is a
brilliant scientist who has spent her entire life in a laboratory, in
which all objects are black-and-white (and shades of gray). Her eyes
have never received light except in the full visual spectrum ("white
light"). So in growing up she never experiences what it is like (to use
Nagel's idea) to see a red apple. But she reads everything (there is a
vast library available to her), and knows all about color vision (how
it works) and knows that when people are confronted with something
that reflects mostly blue light, their brains do this and that and
they experience that special thing they call "seeing blue", and so
on. Mary even knows how brains manage to make that experience happen,
and she understands precisely how -- in neural terms -- it differs
from the experience people call "seeing green" etc. And so she already
knows apples (not Granny Smith apples, but say Red Delicious ones) are
red. But she has never seen one, or anything red at all.

Then one day someone brings a nice red apple into the lab. Mary sees
it. This is the first time mostly red light has ever fallen on her
retinas. She knows exactly what is happening in her brain as a
result. But does she now also knows something more, that all her
understanding of the material aspects of brain function cannot
capture? Does she now know -- and did not know before -- what it is
*like* to see red?

If so, then this suggests again that perhaps one really has to be a
bat to know what it is like to be a bat. And that no amount of
neuroscience -- even an "all details solved" neuroscience of the
future -- can ever capture the what-it's-like of experiences. This is
called the problem of the "explanatory gap": even if all the data are
in and thoroughly understood in all details large and small, the
experience itself (which those details amount to, assuming the mind is
in fact nothing but a form of brain activity) is not revealed except
to that one brain that actually is that activity itself.

But then what is it exactly that Mary comes to know when she first
sees the apple? What is she suddenly and newly conscious of? What is
it that happens when we awake in the morning? What happens when we
close our eyes and then open them again?  The world out there does not
change (much); but our inner world does a lot: something goes away,
something comes back. Is it knowledge? If so, it is not the usual kind
-- it is knowledge of visual our perception going on.  

Or rather, is it instead a kind of knowledge that simply is that very
occurence of perception (not a separate side observation "oh, I have
perception?")?  Does knowledge have to be propositional -- have a
truth value, as in knowing that X is the case? Or can it be a kind of
knowing that is, simply, the act of knowing? Not directed toward
something else, but rather self-directed, the experienced "I'm back"
on first coming awake after deep dreamless sleep?

Whether true self-reference is possible at all has been questioned,
i.e., reference in which something, S, refers literally to that very
same S at that same moment of referrring. David Rosenthal (City Univ,
NYC) suggests -- in his HOT (higher-order thought) theory -- that for
a mental state S to be conscious, one must have a thought that "mental
state S is going on in me".  But the "me" here seems to be the hard
part, and he says little about that. HOT seems to mostly be about the
case of one part of a system being about another part, what is perhaps
better called meta-cognitive rather than higher-order.

David Chalmers (Australian National University) coined the phrase "the
hard problem of consciousness" to call attention to the actual feels,
or qualia, or qualitative or "phenomenal" or subjective character we
associate with consciousness -- the what-it's-like. While many had
written about it, his phrase has caught on, as a way of distinguishing
the "easy" things we can study about mind (memory, low-level vision,
etc) that in themselves seem -- so far -- to say nothing at all about
why roughly 100,000,000,000 neurons undergoing certain firing patterns
should be accompanied by any sort of feel at all.  Chalmers also
proposes his own ideas about what consciousness is; they are very
complicated. But one part of his argument involves philosophical
"zombies", or p-zombies. These are not at all the usual kind of horror
movie creature. A p- zombie is exactly like any human, molecule for
molecule. Its brain works exactly like anyone else's. And as a result,
a zombie is completely indistinguishable from anyone else in all
respects. The only difference -- totally undetectable -- is that
zombies have no qualia, no consciousness, no actual experiences; there
is nothing it is like to be a zombie, no thoughts or feelings to
accompany the brain activity. Chalmes asserts that while zombies might
not be possible in the actual world we inhabit, we can conceive of a
world where they might exist; and he uses that idea in a very
difficult argument to conclude that mind (consciousness) cannot be
simply a special case of the workings of matter. This is very
controversial, to say the least.

Daniel Dennett published "Consciousness Explained" in 1991,
in which he argues that consciousness is an illusion or story
that brain tells itself, but that there is no such thing other than
the story (or multiple stories, which he sees as competing with each
other in the brain). He advocates an "intentional stance", that we
take ourselves (and others) to have Brentano intentionality since that
makes lots of things easier to understand, but that there is nothing
to it aside from that.

Until about 15 years ago (1994 in fact) it was considered in bad
taste (and risky to one's career) for a scientist to talk about
consciousness as a problem worth investigating. But this all changed
when Francis Crick (1916-2004) published his book "The astonishing
hypothesis". The hypothesis is just this: that the brain, by the
firing pattern of its neurons, does in fact constitute mind itself, --
with all its wonders, ecstasies and agonies, the very essence of being
a thing that it is like something to be, the essence of me, you,
everyone). And he discussed in some detail how one might go about
studying -- scientifically -- how this might be. His prestige (based
on his Nobel-prize-winning work with James Watson that led to their
discovery in 1953 of the DNA doublehelix structure) made the topic
of consciousness scientifically respectable overnight.

My own thoughts: self-reference is one of the keys. If we can figure
out how there can be a process that is "about" itself, literally, then
we will be a lot closer to cracking the hard problem, closing the
explanatory gap. And I think language may be a good way to proceed. If
we can find a particular language task that -- for an NLP system to do
it -- requires the system to use "I" in a way that meets Perry's
challenge, we might be a lot further along.  And Descartes' own "I
think..." seems to be a kind of self-reference of this sort, where its
doing is the same thing as its thinking (about that very
doing/thinking). And that, I suspect, will turn out to have a lot to
do with real-time error-correction. A smart system in a complex
dynamic world has to make adjustments, and sometimes those adjustments
themselves have to monitored and adjusted on the fly.

This seems to require the ability to be reasoning mid-way during the
process of acting on a plan as it is being (-re-)planned in real time,
as in "I am planning/acting these few current seconds of activity and
I am adjusting it as I see it unfold; my plan is about the
past-present-future which is the unfolding of this evolving plan. "