PREAMBLE
The purpose of this work is not to be right, Lord no, for we may be
wrong.
We do hope to plant a seed however, by singing The Song that we all
once knew, back when our minds were free, so that those who might
benefit from it can perhaps start to think another way than what we have
been taught.
Homer W. Smith Nov 1, 2021 Ithaca, NY.
ABSTRACT
This paper describes the Machine Certainty Theorem (MCT) and it's
ramifications.
The MCT states that a machine can not learn anything with perfect
certainty. (Thus the MCT might better be called the Machine UNcertainty
Theorem.)
This statement depends on the definitions of machine, learn, and
perfect certainty.
A machine is defined as any system of parts interacting via cause
and effect across a space time distance.
A "space time distance" is later generalized to "dimension" which
is later generalized further to "difference", i.e. two different
objects.
OBJECTS
A object is anything that has qualities, and a quality is anything
that describes an object in whole or in part.
Objects have states described by their object quality set, the set
of all qualities of that object, both alone and in relation to other
objects.
The object that learns is the 'learner.'
The object that is learned about is the 'learned about.'
The object that is between the learner and the learned about,
providing the distance or difference between them, is the 'learned
through.'
LEARNING
To learn is defined as coming to know knowledge about.
Knowledge is a state in the learned that represents, renders or
symbolizes a state in the learned about.
Knowledge is gleaned by the learner from the learned about via
cause and effect.
The learned about is the cause and the learner is the effect, the
knowledge learned is represened, rendered or symbolized in the learner
as a change in state in the learner that then indicates its knowledge
about the learned about.
Without cause and effect there can be no learning as the effect IS
the learning about the cause.
PERFECT CERTAINTY
Certainty is defined as perfect certainty.
Perfect certainty is defined as learned knowledge which can not be
wrong, ie knowledge one is sure is true without possibility of error.
We discriminate false certainties where in the learner THINKS there
is no possibility of error, from true certainties where there IS no
possibility of error.
REFERENTS AND SYMBOLS
It is asserted that in the physical universe, all learning results
from being the effect of some cause.
The object which is cause is called the original referent.
The object which is effect is called the symbol of final authority,
meaning the last symbol in line that we choose to learn from.
The cause produced by the referent and its causal qualities
produces the effect which is a change in the qualities of the symbol.
TRACKING
Tracking is the fact that some qualities of the symbol are caused
by and are thus tracking some qualities of the referent indicating a
cause and effect relationship between referent and symbol.
Because of the time involved in causal traversal from cause to
effect, a symbol can only represent a referent as it WAS, never as it
IS.
Thus even a machine looking at itself via cause and effect can only
know how it WAS, never how it IS.
DATA
The effect rendered in the symbol contains some data about the
nature of the cause in the referent, and that effect IS the learning
that takes place in the symbol about the referent.
This is called Learning by Being an Effect.
As effect does not prove cause with certainty, Learning by Being an
Effect is only always theory.
In the absence of being an effect there can be no learning.
In the absence of cause there can be no learning.
Since all learning is from effects brought about by a cause, all
learning is about cause, namely how that theorectical cause lead to the
observed effect in the symbol of final authority.
Thus if cause is not involved, no learning can take place.
This is also called learning by indirect perception.
INDIRECT PERCEPTION
Indirect perception means learning about A by looking at B, ie A's
causal imprint on B. One studies the state of B to determine in theory
the state of A.
In this case B is the machine learning about A.
In the absence of an effect or causal imprint on B by A, there is
no learning by A about B.
It is also asserted that "state does not prove prior state", ie any
state a machine might be in, will never contain certainty that it was
ever in a prior state.
Thus it is concluded that a machine can not learn that it has
changed state with (non theoretical) perfect certainty.
The absence of perfect certainty of a change in state implies there
must also be an absence of certainty that an effect was received, which
implies an absence of certainty on any learning that might have been
deduced from that effect about a cause.
It is further asserted that correlation does not prove causation.
By 'prove' we mean only to demonstrate the perfect certainty of.
We do NOT mean merely to provide possible evidence for.
Thus effects do not prove cause, therefore changes in state 'here'
do not prove prior states 'there', either in the machine itself or
elsewhere.
'Here' and 'there' mean two different points in either space or
time or both.
It will be shown that the key term is 'two different'.
It is thus concluded that a machine can not be built to prove that
there is a cause merely by looking at effects or changes in itself, EVEN
IF IT COULD BE CERTAIN OF EFFECTS WHICH IT CAN'T.
Since effects do not prove cause, and all a machine can do is BE an
effect, no machine can learn with certainty that there is cause.
In other words it is offered that because a machine learns by being
an effect of causes, and because effect does not prove cause, a machine
can never learn with certainty if cause even exists merely by studying
effects, let alone learn the nature of that cause.
The concept of an effect is native to a machine, because machines
work by being an effect.
But the concept of cause is non native to a machine.
Cause and its nature forever remain a theory to a machine, not a
directly perceived perfect certainty.
IN CONCLUSION
Because a machine can only learn by being an effect, and
because a machine can not learn with certainty that it has changed
state, and
because a machine can not learn with certainty that a change in
state was caused,
it is concluded that a machine can not learn with certainty about
anything, which includes cause, effect, space, time, or the existence of
anything including itself.
The important application of the MCT is in the reverse, to wit
since consciousness CAN learn with perfect certainty a number of things,
including its own existence, perception of two different colors, and
personal agency (causation), one is led to conclude that consciousness
is not a machine, that is consciousness is NOT using indirect perception
to learn about itself, but instead is using direct perception by looking
at itself directly.
DIRECT PERCEPTION
Direct perception is an oxymoron in the physical universe, but is
defined as learning about cause by looking at cause rather than by
looking at effects of that cause.
That learning about cause by looking directly at cause is so
incomprehensible to the human physical mind, the ability to do so has
escaped human attention until now.
It also needs to be pointed out the irony that consiousness which
learns ONLY through direct perception of itself, is non the less
enslaved to pretending it lives in a physical mechanical universe that
CAN ONLY learn through indirect perception.
In the examples above where one is learning about A by looking at
B, B is consciousness and A is the alleged physical universe, and
EVERTHING we think we know about A comes from theorizing about what we
see directly in B, our own consciousness.
Homer