This file is
    http://www.cs.bham.ac.uk/~axs/misc/my-qualia.text

It is part of a letter sent to Pat Hayes on Thu Dec 16 03:03:23 GMT 1999
as part of an on-going email discussion with Pat, Henry Stapp, Stan
Klein, and others.

To: pat hayes
Subject: My kind of qualia

When I told Dennett a few years ago not only that I had qualia, but
also the germ of a theory to explain how they would inevitably occur
within the architecture of a certain sort of information processing
virtual machine, he was a bit upset and responded by saying that he
wished I would not call them "qualia".

Pat may react the same way.

Let's see.

I think that what I call qualia are the very things that first got
philosophers referring to these internal, non-physical, components
of our experience (though some referred to them as "sense data",
"raw feels" etc.).

(Note: When I say "components" I don't mean they are separately
identifiable enduring entities assembled to form experiences, like
bricks forming a house. Experiences are not like houses and cannot
be built like houses. Likewise, qualia are not like bricks.
More of that anon.)

But just because qualia are not like bricks, and cannot be glued or
jammed together to form larger structures, that doesn't mean they
don't exist.

Unfortunately some philosophers (and their faithful followers among
the scientific elite), started adding extra spurious conceptual
baggage to the little fellars, e.g.

    various kinds of logical privacy,
    causal disconnection from physical mechanisms,
    infallible kinds of knowings,
    intrinsic (non-relational) qualities,
    zombie-phobic characteristics,
    links to quantum collapses,
    unsatisfiable demands for trans-personal or trans-temporal
        identity criteria (see below on identity)
    a requirement for qualia (or sense-data) to provide the
        "foundation" of all empirical knowledge,

    ...and other fascinating stuff,

all of which had the net effect of replacing an admittedly vague and
potentially confusing but still acceptable philosophical technical
term (relevant even to robot designers, as I'll try to show) with a
new concept that was pretty incoherent.

A case of death by accretion.

So when Dennett says qualia don't exist he means that the inflated
and incoherent concept has no instances (not surprisingly).

On the other hand people like Henry then insist Dennett must be
wrong because there ARE instances, namely various components of
experiences, etc. i.e. the very things which were originally
referred to as "qualia", "sense-data", etc.

Well, maybe so. But don't let anyone assume that all those who say
"Qualia exist" mean the same thing by "qualia".

Even worse: don't assume that all those who say "qualia exist" and those
who say "qualia don't exist" mean the same thing by "qualia".

It's not like the pot of gold at the end of the rainbow, or a
unicorn: believers and unbelievers alike would know one if
it turned up.

So what am I referring to when I say that qualia exist?

It's not easy to define in any precise fashion, but the key idea is
to notice that before we become too corrupted by philosophical or
scientific theorising there are different sorts of things we (well,
err... some of us) naturally say we see: some of them obviously "out
there", and some of them obviously not out there, so maybe they are
"in here".

One of the oldest sorts of example is the elliptic entity which it's
useful to learn to see if you want to be able to draw realistic
pictures of circular entities like pennies, viewed obliquely.
(Ah, that brings back memories of many old philosophy tutorials).

Likewise the obtuse-angled and acute-angled corners you can learn to
pay attention to when looking at rectangular table tops.

It's interesting that people don't all naturally notice these odd
entities whose nature is so dependent on viewer and viewpoint. But
not too surprising: the main function of perception, i.e. what it
primarily evolved for, is to provide information about what is in
the environment, e.g. the circularity of the penny the
rectangularity of the table top.

And what's in the environment is that circular or rectangular thing
whose properties don't change just because some viewer bends down to
look at it, or squints, or puts on coloured specs, or grows old and
short-sighted, or has a stroke and can no longer organise sensory
information into the same percepts as before.

My hunch is that very young children and most other animals NEVER
notice the qualia: they only see (hear, smell, feel) actual physical
things and their physical properties. They may HAVE qualia (as
defined below) but either lack the apparatus required to attend to
them, or totally lack the inclination to attend to them (like some
of my old recalcitrant philosophy students.)

(Of course, if some philosophers have *defined* "qualia" in such a
way that having qualia logically entails knowing that you have them
then my kind of qualia don't fit their definition. It's not clear
that anything else fits their definition either.)

What are my qualia then? To a first approximation we can say that
they (at least sensory qualia) are items of information typically
created in the intermediate stages of a perceptual system, and which
can also be created when perceptual systems instead of being driven
by normal sensory input are driven by other processes (e.g. in
dreaming, having drug-induced hallucinations, etc.)

The elliptical qualia that exist when we look at or hallucinate
obliquely viewed pennies are information structures reflecting some
of the relationships normally projected from the optic array, but
which can also be created by other means.

(Once you've noticed the elliptic qualia you can pay attention
to the circular qualia when you look horizontally at pennies held
vertically, or when you hallucinate such.)

Temperature qualia are the things to which philosophers of old drew
attention in connection with a person who starts off with one hand
in ice cold water, the other in very hot water, and then plunges
both into a lovely bucket of lukewarm water.

Our temperature sensors need to be properly calibrated, and if the
calibration is mucked about with we are more likely to start paying
attention to the qualia which were there all along (like the
circular qualia when you peer straight down at the penny).

We also start paying attention to them when we have flu, and notice
that we are shivering with cold when everyone else seems to find the
ambient temperature quite warm.

There are far more complex and subtle qualia besides colour, shape
and temperature qualia. E.g. when you hear a piece of music and
recognise it as being in the style of Beethoven you may or may not
be able to attend to those aspects of the experience which give it
that style: you are probably too used to attending only to the
global effect, until you take a degree in music.

Likewise hearing spoken accents: only the expert can attend to the
qualia that make up the difference between accents from neighbouring
territories.

Some musicians, on hearing a large bell, can disassemble the
harmonics in their experience and then pick out notes on a piano
which when played together sound remarkably like the bell. I can't:
though I can attend to the global similarity between the two sounds.

What's going on when we start to notice and attend to our qualia?
The basic idea is very simple, the detailed story and its
implementation very complex:

All animals have various kinds of sensory processes which deliver
information about the environment (and sometimes about bodily
states) to a number of internal mechanisms (for learning, for
decision making, for sensory-motor feedback in controlling accurate
movements, for modulating a host of bodily functions, for
triggering blinking and saccadic reflexes, and many more).

SOME animals, including humans (thoug perhaps not immediately after
birth) have additional architectural layers (which I suspect evolvd
much later, and in relatively few animals) which can not only gain
the previously mentioned information about the environment, but can
also access various intermediate stages in the processing of that
information (e.g. some of the databases of intermediate 2-D
information structures created during visual processing) and can
also apply different analysis and recognition procedures to that
information.

Moreover, in truly sophisticated information processing
architectures the process of attending to and using such "qualiar"
information structures can itself be attended to -- up to a point
(see below on architectural resource limits).

When philosophers first started talking about sense-data, qualia,
impressions (Locke? Berkeley? Hume?) intuitions (Kant?) that was
because they had started attending to the qualia of attending to
sensory qualia.

Painters, reporters of dreams, etc. had previously attended only to
sensory qualia.

None of this can happen by magic. It needs a suitable information
processing architectures to support the processes.

Exactly what is required will be different for different sorts of
qualia. Spelling out all those requirements is a very long and
complex and quite difficult task because our introspections on these
matters, far from being infallible are grossly fallible and very
liable to be corrupted by philosophical and other theories we
absorb.

But if you have the appropriate architecture and it does its stuff,
you have the qualia: even if that fact is totally hidden from
everyone else because you are asleep or paralysed, etc. etc.


WARNING: none of this assumes or entails that there are identity
criteria which make it intelligble to ask whether different people
do or do not have the same qualia, or even whether one person has
"really really" the same qualia now as two minutes ago. The identity
of qualia is like the identify of points of absolute space: a
mythical source of incoherent questions.

I can say this portion of space exists here, now, without being
prepared to answer any questiosn about where it was five minutes
ago, or whether a point identified ten minutes later is or is not
the same point.

Similarly with qualia.

If a philosopher *defines* qualia so that such identiy questions
*must* have answers then that poor, sad, philosopher's qualia don't
exist.

But mine do.

So will the qualia of my robots, when I finally get around to
building them and arguing with them about whether humans have
qualia.

Dennett, unlike me, accepts the definition of qualia which renders
the notion incoherent, and then says that although he has
experiences he has no qualia because an incoherent concept cannot
have instances.

I think Pat follows Dennett.

The reason I reserve the word "qualia" for the non-incoherent
concept is that I need *something* fairly familiar to refer to what
I am talking about, and the simplistic coherent use has an old
enough philosophical pedigree to allow me.

And by saying "qualia exist" I can draw attention to phenomena that
I believe Dennett did not analyse adequately and did not explain at
all (or, more precisely, did not add anything to the embryonic
explanations implicit in pre-existing work in AI, robotics, etc.)

Choosing labels is just a matter of marketing.

[Henry]
> >But in the case of an intense and enduring experience what
> >that experience is IS what it seems to be,

What it seems to be BEFORE you've studied philosophy and become
corrupted by bad theories is different from what it seems to be
AFTER you have done so.

Of course you no longer remember what it used to seem to be.

[Henry]
> > ...for that
> >experience is a sustained process of knowing the qualia
> >within a background field that allows the person to know
                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> >that he know it.
   ^^^^^^^^^^^^^^^
This is getting close to the information processing account, which
requires a metamanagement (reflective) layer in the architecture.

But it goes way beyond what experiences seem like to a three year
old, or a novice philosophy student. Henry's kind of seeming is a
result of much theorising, and far from universal.

[Pat]
> Dennett is not arguing against the reality of experiences.  The
> 'properties' to which he refers, and is denying the existence of, are
> those
> "which have been banished from the "external" world by the triumph of
> physics: 'raw feels', 'sensa', ... and of course 'qualia', the term I
> will use."
> (Dennett, same page.)

Well physics achieved no such banishment of the entities I am
talking about.

Henry thinks they are banished from classical physics, but he still
hasn't convinced me that CP has any such nothing-buttery
implications. It doesn't talk about them, but it also doesn't talk
about transfinite sets, computer operating systems, crime, poverty
or genes either. It doesn't banish things just because it lacks the
concepts needed to refer to hem.

Ignorance is not denial.

[Pat on Dennett]
> He is arguing against what might called the qualiar theory of
> experience - that experiences are somehow comprised of a certain kind
> of nonphysical essense which comes in atomic units called 'qualia'.

Note: I am not referring to "atomic units", and certainly not atomic
units with an enduring identity. There's nothing I have said which
implies that there's a unique, correct, way of carving up
experiences into qualia.

There MAY be a unique decomposition for certain simple sorts of
qualia produced by certain sorts of discrete computational
mechanisms. But other kinds of qualia may be as lacking in clear
boundaries as the hills and valleys in rolling terrain.

[Pat, to Henry]
> ...Your assertion that "experience is a
> sustained process of knowing the qualia within a background field" is
> an elegantly compact summary of the core of that theory, so we can
> presume that you accept it. But one can reject that theory without
> rejecting the reality of experience.

Yup.

[Pat on Dennett]
> ... in fact: he gives color a *physical*
> significance (color is a property of the reflectance spectrum of a
> surface.) That is Dennett's point here: we do not need to introduce
> 'qualia' to serve as the bearers of color properties, since colors
> are properties of the external physical world, not of experience.
> Colors are real, and surfaces really have them, and we see those
> colors just as we see the surfaces.

Sure: just as shapes are real and surfaces really have them, and we
see those shapes just as we see the surfaces.

But besides the shapes of the physical objects there are also the
"shapes" we can attend to when we notice our qualia. I am not
claiming that EXACTLY the same concept of shape is applicable to
both (hence my use of quote marks).

E.g. the internal shapes are probably inherently indeterminate to a
far greater degree. Likewise the colours. It's a commonplace that
something experienced may be vague in ways that are impossible for
physical objects, like the *number* of experienced points of light
when I look at a starry night sky. (There's more on colour
indeterminacy in my map example below.)

[Pat's qualia:]
> Here is the picture as Dennett (and I and many others) see it. Things
> exist in the world, and have many complex properties. The process of
> perceiving the world involves the construction of an internal
                                                    ^^^^^^^^^^^
> description of it (of a certain kind and in a certain functional
  ^^^^^^^^^^^^^^^^^
> relationship to other parts of the self); one which, while itself
> very intricate, is of necessity both incomplete and in some sense
> idiosyncratic, both to the individual and to his species. In many
> respects it may be inaccurate, and even systematically in error;
> nevertheless, it is a description, and it is of the external world. A
> person has an experience (of something in the world) just when an
> appropriate such description (of that part of, or thing in, the
> world) is appropriately constructed in its brain in an appropriately
> functional way.

The only difference between this and my story is that by using the
singular form "an ... description .. one which ....itself ..etc..."
Pat (unwittingly?) obscures the important disinction between the
biologically important description produced of what is out there
in the environment and the many other descriptions that may be
produced of intermediate structures in the sensory array, and which
even if produced in all animals, may be "centrally" inaccessible to
most of them, and in humans are generally not accessible by high
level decision-making processes without special training.

Multiple intermediate descriptions are typically required in all
interesting AI vision systems, language understanding systems, etc.
etc. (though not in the more simplistic neural net models). The
reasons have been understood for at least 40 years. (E.g. Minsky's
1962 paper: Steps towards artificial intelligence.)

I also disagree with Pat on the nature of the descriptions: the
descriptions are themselves typically structures in virtual
machines, NOT physical objects.

We might also disagree as to the diversity of forms required for
such intermediate descripions, in human-like perceptual mechanisms.

[Pat]
> ... the key point is that in this kind of an
> account, there is no need to hypothesise any kind of nonphysical
> 'qualia' in order to account for experience. Everything in this
> account is physical: the things experienced and the representations
> which describe them.

Pat has here lapsed into the kind of mind-brain identity theory
which I think is mistaken, and which in other contexts I think he
has also argued is mistaken.

[Pat]
> .... I think that qualia are a really bad idea,
> and it might be interesting to recast your ideas in a way that avoids
> them.

I think they are a really good idea. Paying close attention to them
(doing phenomenology well) helps us to define far more precisely
than ever before the requirements for a robot (or zombie) that is to
have human-like internal information processing capabilities. If you
ignore them you'll just say that the robot can perceive the
roundness of pennies, the greenness of grass and the coldness of icy
water.

But a robot which could ONLY do those things would lack some of your
and my experiences (qualia), though it might to that extent be like
unreflective animals (and human infants) which are sentient but not
self conscious.

To be like me it would need far more.

Moreover: it's not generally understood that that "far more" need
not be behaviourally manifestable. The internal information
processing may be so rich that none of the available physical
behaviours have sufficient bandwidth to manifest them!

Moreover, even if they could be manifested in principle, the
architecture need not support mechanisms for such external
manifestation, for the richness is required only for intermediate
stages of complex internal process.

Hence my kind of virtual machine functionalism does not require that
every feature of the evolved mechanism had to be directly selected
for by some environmental constraint requiring specific behaviour.

The people who say they can visualise zombies with information
processing capabilities just like ours but lacking our experiences
don't realise that they have failed to specify the "information
processing capabilities just like ours".

[Pat]
> PS. For the record, I wasnt persuaded by Dennett to reject 'qualia';
> I have found the concept ridiculous ever since I first heard it, and
> find it hard to see how anyone can take it seriously.

Maybe you were looking at the wrong "it". Try my version.

[Pat]
> Far from being
> a convincing account of my experience, I find it inaccurate to the
> point of lunacy. When I look at a tree what I see is the tree, not a
> tree-quale; and what I come to know as a result of looking is
> something about the tree, not about a quale.

When you go to the optician to have your eyes tested and he asks you
which of two lines is sharper, etc. you are not being asked about
the thing on the wall, but about your qualia (in my sense, the
original philosophical sense).

What's more I am sure that you understand that very well in that
context. You don't answer: "Sorry I can't tell you which line is
sharper because I need new glasses and only with them will I be able
to see the line accurately."

> I have no idea what it
> would be like to know anything about a quale, and nobody is ever able
> to tell me.

I have.

But please don't assume that "a quale" refers to the unique result
of a process of decomposition of experience into enduring
re-identifiable parts. That's just another bad theory.

[Pat]
> When I go back to the philosophical sources I find only
> trivially simple 'monadic' examples like redness-quales. I can't even
> get qualiar enthusiasts to agree on how many quales there are, or how
> long they last, or what their individuation criteria are.

If they try to agree, or to disagree, then they too have missed an
important feature of the concept of "qualia" -- noticed long ago by
some other philosophers: I can remember these points about
identity being discussed when I was a student in the late 50s,
though I don't recall sources. Maybe I learnt all this from Ryle.
(Or Waissmann, or Wittgenstein?)

[pat]
> (If I look
> at a coke can, do I have a coke-can-quale, or a can-quale plus a
> red-quale?  If the light changes, do the old quales die and get
> replaced by new ones, or do they just kind of quale-morph?)

I am partly colour blind. One of the manifestations of this is that
when I look at a map of the London underground, especially when the
lines are fairly thin, I can ask of two pieces of line: "are they
the same colour" yet if they are far apart I find myself totally
unable to answer even though when I fold the page to bring them
together I may decide they are different, or if I serially follow
the route to see if they are part of the same line I conclude they
are the same.

In my youth I sometimes went to musical summer schools and played
chamber music for a week. After several days of non-stop playing I
could play the A on my flute and tell whether it was sharp or flat,
and adjust it without having to compare it with a tuning fork or
piano A. After coming home and playing less often, that ability went
away again.

Was the experience of hearing the A the same when I temporarily had
"perfect pitch" different from the experience when I didn't have it?

Well as with identify of locations: it depends on your frame of
reference.

In relation to some things I could do it was the same, and in
relation to other things I could do it was different.

But was it REALLY REALLY the same???

Answer: the question is nonsensical, like the question whether
the ancient Greek REALLY did or did not return the axe he borrowed,
after he had first replaced the broken blade, then replaced the
broken handle.

One type of philsophical maturity is learning to reject demands for
identity criteria!

[Pat]
> If I am
> having more than one quale at once, how do I keep them appropriately
> linked to each other?

YOU don't. But you have plenty of (ill-understood) information
processing mechanisms which keep doing (and undoing) such things.

> (Think of a red can on a blue table. Is there a
> relational network of quales, maybe?)

Yup. Or something like that.

If you remember trying to design visual systems (as you once did in
your youth) you'll understand why. But there doesn't have to be a
unique relational network into which every experience can be (or is)
decomposed.

Nor a unique one synthesised from any given input sensory array.

[Pat]
> If one takes it seriously things get worse. For example, it is
> claimed (eg by the Friends of Mary) that coming to know something is
> an experience (since there is something that it is like). Well then
> if experience is a sustained process of coming to know a quale, then
> presumably this coming to know, being an experience, itself has a
> quale, and the coming to know of quale-1 is a having of quale-2. Lets
> call quale-2 the know-quale of 1, or know-quale-1 for short. Now,
> this immediately suggests a rather worrysome infinite recursion,
> since it is trivial to show that there must be quales of the form
> know-know-....-know-quale for any sequence of know-s.  The only way
> out of it is to truncate the sequence by claiming that at some point
> a quale is identical to its know-quale.

It's not the only way.

Just assume that there are limits to the (simultaneous)
self-reflection capabilities in an intelligent information
processing architecture. Assuming such limits is plausible given
various general assumptions about how such architectures need to be
implemented in physical mechanisms.

Logical omniscience is for that reason (among others) a discredited
requirement for a sensible axiomatisation of knowledge in
implementable agents.

[Pat]
> This would be a kind of
> fixedpoint of the coming-to-know-the-quale process; a quale which one
> would somehow know all about purely by virtue of having it; the quale
> of an experience which is total in its immediacy, in the sense that
> it cannot be further known by any process of contemplation. I don't
> remember having any of these, myself, but then I'm only in my fifties.

We can speculate about all sorts of mathematical constructs without
assuming that they have models in any particular physical system or
fully implemented virtual machine.

Maybe that's one way to give Stan's rabbi his god, if he really
wants one.

Aaron