Chapter 3 Part E
As an acceptable alternative to brain structure and chemistry,
scientists interested in thought also study patterns of behavior in organisms like
rats, pigeons, and people that are being stimulated in controlled, replicable
ways. We can, for example, try to train rats to work for wages. This kind of
study is the focus of Behavioral Psychology. (See Baum’s 2004 book
“Understanding Behaviorism”.) (7)
As a third
alternative, we can even try to program computers to do things very similar to
the things that humans do. Then if the computers do behave in human-like
ways, we should be able to infer some tentative, testable conclusions about
what human thinking and knowing are from the programs that enabled these
computers to behave like humans. This kind of research is done in a branch of
Computing Science called "Artificial Intelligence" or A.I.
To many empiricist philosophers and
scientists, A.I. seems to offer them their best hope of defining once and for
all a base for their way of thinking, a base that can explain all of human
thinking’s so-called “abstract processes” and that is also materially
observable. A program either runs or it doesn’t. One that made computers
imitate humans so well that we couldn’t tell which was the computer answering
us and which was the human would arguably have encoded what thinking is. At
last, a beginning point beyond the challenges of the critics of Empiricism and
their endless counter-examples. (A layman’s view on how A.I. is doing is in Meltzer’s article in The Guardian, 17/4/2012.) (8)
Testability and
replicability of the tests, I repeat, are the characteristics of modern Empiricism
and of all Science. All else, to modern empiricists, has as much reality and as
much reliability to it as fantasy creatures in a fantasy novel ... amusing
daydreams, nothing more.
Kurt Godel
The most
optimistic of the Empiricists for years were looking to A.I. for models
of thinking that would work in the real world. Their position has been cut down
in several ways since those eager, early days. What exploded it for many was
the proof found by Kurt Godel, Einstein’s companion during his lunch hour walks
at Princeton. Godel showed that no rigorous system of symbols for expressing some
of the most basic of human thinking routines can be a complete
system. (In Godel's proof, the ideas that he analyzed were basic axioms in Arithmetic.) Godel's proof is difficult for laymen to follow, but
non-mathematicians don't need to be able to do that formal logic in order to
grasp what Godel’s proof implies about everyday thinking. (See Hofstader for an
accessible critique of Godel.) (9.)
Douglas Hofstadter
If we take what
it says about Arithmetic and extend that finding to all kinds of human
thinking, then what Godel's proof says is that there is no symbol system for expressing
our thoughts that will ever be good enough to express and discuss all of the
new ideas that the human mind can dream up. Furthermore, in principle – in other
words at the roots of human thinking itself – there can’t be any such system of
expression. (O.W. Holmes said: “No generalization is worth a damn, including
this one.” If this statement is true, it makes itself false. If false, it makes
itself true. Our problem occurs because the statement is self-referencing. It
is a rough example in ordinary English of what Godel is talking about.)
What Godel's
proof implies is that no way of modeling what the human mind does will ever adequately
model or explain that very thing. Not in English, Logic, French, Russian,
Chinese, Java, C++, music, or Martian. We will always be able to generate
thoughts, questions, and statements that we can't express in any one symbol
system. If we find a system that can be used to encode some of our favorite ideas
really well, we will only discover that no matter how well the system is
designed, no matter how large or subtle it is, we will have other thoughts
that, in that system, we can't express at all. Yet we have to make statements
that at least attempt, however inadequately, to communicate our ideas. Science,
like almost all activities of human life, is communal. It has to be shared in
order to advance.
The further
conclusion to be drawn from Godel is that researchers in sciences like
Physiological Psychology, Behavioral Psychology, Computer Science, Sociology,
etc., may all study what thinking and knowing are, with each discipline
approaching the phenomena from its perspective, but no single one of them, and
no set of them taken together will ever completely define what human thinking
and knowing are. No system for representing and expressing the thoughts
generated in the mind will ever be capable of fully defining or describing the
mind or what it is doing as it does that generating.
Other theorems in
Computing Science seem to offer fascinating support to Godel's theorem. For
example, in the early days of the development of computers, programmers over
and over were creating programs with loops in them. After a program had been
written, it would be run and then, sometimes, the program would get stuck in a
sub-routine that kept going over one sequence of steps from, say, line 193 to
line 511 then back to line 193, again and again. Whenever a program contained
this kind of flaw, a human being had to stop the computer, go over the program,
find why the loop was occurring, then either re-write the loop or write around
it. The work was frustrating and very time consuming.
Soon, a few
programmers got the idea of writing a kind of meta-program that they were
hoping would act as a "check" program. It would scan other programs,
find their loops, and fix them, or at least point them out to the programmer so
that she could fix them. The programmers knew that writing such a program would
be difficult, but once it was written, it would save so many people so much
time.
However, progress on the writing of this
"check" program seemed to be running into difficulty after
difficulty. Eventually, someone really good with computer languages (Alan
Turing) published a proof which showed that writing a check program was, in
principle, not possible. A foolproof algorithm for checking other algorithms is,
in principle, not possible. (See
“Halting Problem” in Wikipedia.) (10)
This finding in
Computing Science, the science which many people see as the bridge between the
abstractness of thinking and the concreteness of material reality, is, I
believe, Godel all over again. In another kind of proof, it confirms our
deepest feelings about Empiricism. It is doomed to remain incomplete. No
completely effective check program has ever been found. Some check programs
which are able to catch the simpler mistakes that beginning programmers make have
been written, but no foolproof one has ever been created in any of the many
programming languages that have evolved in the field over the years.
There are some
computer scientists who believe that by using math theorems not even known yet,
one day it may be possible for programmers to write computer programs that keep
iterating closer to doing what human programmers do, even checking other
programs. But they admit that no program so far can do what a human programmer
does when she fixes a faulty program … or writes a song.
The possibilities
are fascinating, but for our purposes in trying to find a base for a
philosophical system and a moral code, the conclusion is much simpler. The more
we study both the theoretical points and the real world evidence, including
evidence from Science itself, the more we are driven to conclude that the Empiricist
way of seeing or understanding what thinking and knowing are will probably
never be able to explain itself. If Godel's proof is right, and nearly everyone
in Math and Computing Science thinks it is, and if it is extended to human
thinking in general, Empiricism's own methods have ruled out the possibility of
an unshakable Empiricist beginning point for epistemology.
If I think that I have found a way to
describe what thinking is, then I will have to express what I want to say about
the matter in a language of some kind … English, Russian, C++ or some other
sort of language for encoding thoughts. But there is not, nor can there be, a
code that is capable of capturing and communicating what the thinker is doing
as she is thinking about her own thinking. It is a mental conundrum with no
solution. (What is the meaning of the word “meaning”?)
Diagram of the human brain
A single neuron, showing its branching structure
Of course, the
last few paragraphs are only describing the dead ends that have been hit in
A.I., but other sciences searching for this same holy grail – a clear,
evidence-backed model of human thinking – haven’t fared any better. Neurophysiology
and Behavioral Psychology also keep striking out.
If a neurophysiologist could set up an MRI
or some other similar imaging device, then predict in advance which networks of
neurons in his own brain would be active when he turned the device on and
watched the machine give its information on his own brain activities as he was
studying them himself, in real time, then he and his science could say that they
had formulated a model of what consciousness is. But on both the theoretical
and practical sides, neuroscience is not even close to being so complete.
Patterns of neuron firings that are mapped
on one occasion when a subject is performing even a very simple task unfortunately
can’t be counted on. We find different patterns of firings, apparently, depending
on what we assume consciousness is and different researchers’ definitions
differ widely. A human brain contains one hundred billion neurons, each one
capable of connecting to as many as ten thousand others. Philosophers looking
for a solid base for Empiricism are disappointed if they go to neuro-physiology
for that base. (11)
Similar problems beset Behavioral
Psychology. The researchers can condition rats and predict what they will do in
controlled experimental situations, but endless ad hoc add-ons and exceptions
have to be made to their explanations of what humans in everyday life do.
In a simple example, alcoholics who say
that they truly want to get sober for good can be given a drug that makes them
violently, physically ill if they imbibe even very small amounts of alcohol,
but that does not affect them as long as they do not drink alcohol. This would
seem to be a behaviorist’s solution to alcoholism, one of society’s most
intractable problems. But alas it doesn’t work. Thousands of alcoholics have
kept on with their self-destructive ways while on disulfiram. (12) What is going
on in these cases is obviously much more complex than any explanation given by
behaviorism’s best theories can account for. And this is but one simple example.
I, for one, am not disappointed to learn
that the human animal turns out to be an enormously complex piece of work, no
matter the model under which we analyze it.
Notes
7. Baum, William;
“Understanding Behaviorism: Behavior, Culture, and Evolution”; Blackwell Publ.;
2005
8. Meltzer, Thomas;
“Alan Turing’s Legacy: “How Close Are We To Thinking Machines?”; The Guardian,
June 17, 2012.
9. Hofstader, Douglas;
“Godel, Escher, Bach: An Eternal Golden Braid”; Basic Books; 1999.
10. “Halting Problem”;
Wikipedia; 2012.
12.
Fuller, Richard K.; “Does Disulfiram Have A Role In Alcoholism Treatment
Today?”; Addiction; Dec.