Pentti O A Haikonen
Only
the Impossible Is Difficult Enough
Dr.
Pentti O A Haikonen
Adjunct Professor
Department
of Philosophy
University of Illinois at
Springfield
Machine Cognition – Robot Brains
– Machine Emotions – Qualia –
Conscious Machines – Artificial
Consciousness – Philosophy
_______________________________________________________________________
My Youtube videos can be seen here:
Serious work: Demo videos of my cognitive robot XCR-1, a demo of my quasi-quantum computer that is able to find factors of given binary numbers instantly, and an idea what was before the big bang.
Entertainment: Some season's greetings videos.
_______________________________________________________________________
Books
and book chapters:
Pentti O
Haikonen: Robot Brains; circuits and systems for conscious machines.
Wiley and Sons,
UK 2007
From
associative neurons and neuron groups to perception circuits, cognitive
architectures, machine emotions, natural language in machines, machine
consciousness
Contains a
little bit mathematics and circuit diagrams. Get your soldering iron ready or
write some code!
It
is useful to read the old book first.
http://eu.wiley.com/WileyCDA/WileyTitle/productCd-0470062045.html
These pictures are not in the book:
My first hardware neuron
from 1990’s with digital/analog correlator synapse and match/mismatch/novelty
detection.
This could demonstrate
Pavlovian conditioning. An execution of the block diagram in fig. 3.3. of the book.
This is the timed sequence
learning circuit of fig. 4.28 in the book. The circuit boards are: Top left,
simple audio synthesizer,
top right and bottom left,
the actual circuitry of fig 4.28, bottom right, one octave filter bank (eight
resonance amplifiers) and a microphone.
(Interconnecting flat
cables are removed for clarity.) Experimental associative neuron group
microchips made by
VTT (Technical Research
Center of Finland) were used (the two big chips).
_____________________________________________________________
The good old
one:
Pentti O
Haikonen: The Cognitive Approach to Conscious Machines. Imprint Academic,
UK 2003
My background
philosophy towards the design of conscious machines – Easy reading, no
mathematics here, lots of ideas.
Based on
cognitive sciences, engineer’s insights and common sense.
http://www.imprint.co.uk/books/haikonen.html
http://www.amazon.co.uk/exec/obidos/ASIN/0907845428/202-3266986-3686259
___________________________________________________________________________
Pentti O Haikonen Videotekniikka
1992 - 1994 (In Finnish). Sold out.
___________________________________________________________________________
Book Chapter
in Visions of Mind, Darryl N Davis (editor), Idea Group Publishing, USA
2005
http://www.idea-group.com/books/details.asp?id=4629
____________________________________________________________________________
Useful
links:
The pertinent issues of
artificial consciousness (machine consciousness) explained:
http://en.wikipedia.org/wiki/Artificial_consciousness
______
Excellent site for
conscious robots, good content, frequently updated:
http://conscious-robots.com/en
The very nice site of
Giorgio Marchetti about mind and consciousness:
www.mind-consciousness-language.com
______
Suomen Tekoälyseura –
Finnish Artificial Intelligence Society
______
Mathematics: Why
differential forms are good for the representation of the physical world? Terho
Max Haikonen explains.
Notes on differential forms and spacetime:
http://koti.welho.com/thaikon2/spacetime3.pdf
_________________________________________________________________
Essays and other stuff
Five
Easy Steps Towards Conscious Robots
Consider a robot with environmental sensors for the
sensing of the environment and self-sensors that sense the states of the robot
itself. This robot could be implemented with various degrees of complexity.
Depending on this complexity the following cases can be
distinguished:
1. Simple
reflex. A robot detects an obstacle, backs off, turns a little and goes forward
again until another obstacle is encountered. Would this be a conscious act?
After all, the robot has detected the obstacle and reacted to it, so it must be
“aware” of the obstacle? However, even toy robots can act like this and they are
definitely not conscious.
2. Simple reflex
with memory. The robot is able to react to obstacles and can record its history.
For instance, the robot could learn to negotiate a maze by memorizing correct
turns at each obstacle. Now the robot will be “aware” of the obstacles and the
layout of the maze. Is it conscious or merely executing mechanically a string of
operations?
3. Perception
with meaning and associative memory. The robot perceives the world and can
“learn” from experiences. The approaching obstacle evokes “images” of past
encounters with similar obstacles and the consequences of the same. Some kinds
of “pain” for dangerous objects and “pleasure” for useful objects are also
involved. The evoked “imagery” and the “pain” trigger avoiding actions. In this
case the perceived obstacle has a meaning to the robot –something to be avoided
in certain ways; something that has “pain” associated with it. Perception is not
simple recognition, instead it is an active process of searching and detecting
opportunities and threats. Perceived entities are not recognized, instead they
remind of things and evoke “imagery” of action, which may or may not be actually
executed. The robot is looking for “satisfaction”. The focus of perception and
action is controlled by needs and good/bad criteria; attention process.
4. Perception
with associative memory and report. The robot can declare what it is perceiving
and doing while doing it and afterwards. The robot can associate meaning with
the declarations of its peers. The robot can also report its self-sensor
percepts. This calls for the use of representational symbols and symbol systems;
“words” and “language”.
5. The robot
perceives itself perceiving. The robot can perceive its own declarations and
understand them. At first this may happen via environment sensors, later on the
declarations may be looped back to the sensory circuits internally. The robot is
able to make the distinction between the percepts that are caused by the
environment and those that are caused by internal processes. The perception of
internally caused “images” and the associative evocation of further “inner
images” lead to the ability to “imagine”. The flow of the “narrative inner
images” and their related good/bad value evaluation would constitute a kind of “
mental content”. On the other hand the robot would have needs to satisfy and
active as well as latent “imagery” of tasks to be executed. All these would
constitute the “mind” of the robot. At this point the robot seemingly has most,
if not all hallmarks of consciousness.
It should be noted that none of these steps involve
the creation of “specific conscious circuits”. The steps 3 and 4 are crucial,
but clearly within the realm of engineering and can be realized by proper
application of associative memory and system architecture. The final step 5 does
not actually involve any additional hardware, instead an additional way of
operation is required; the inner feedback that enables the robot to perceive its
mental content silently, without the need to act it out for the environmental
sensors. It should also be noted that the resulting phenomenon that has the
characteristics of “real consciousness” is not a circuit, it is not an observer,
it is not a causal agent; instead it is a content level way of operation. It is
the way, how the system perceives itself perceiving.
Summary
of the Haikonen Model of Machine
Consciousness
The Haikonen
cognitive machine is characterized by: Distributed signal representation,
associative processing and learning, perception process, sensory attention and
inner attention, the flow of inner speech, inner imagery and the equivalent for
other sensory modalities, evaluation of significance, basic system reactions,
machine emotions, motivation, imagination, “mirror neuron” action. This machine
utilizes an architecture that is characterized by: Sensory preprocessing
circuits that derive distributed representations from sensors, very large number
of introspective feedback loops that detect sensory information and broadcast it
to other loops, associative cross-coupling of these loops, attention control via
large number of variable thresholds. The key concepts are summarized
here.
Reception. The system receives signals from
sensors.
Detection. The system
discriminates received signals from noise and interfering signals (non-linear
threshold mechanisms etc.)
Perception.
The system augments signal
detection by attention, experience, context and expectations (feedback
mechanisms) A percept is the “official” result of perception process. Percepts
have direct causal meaning: They depict the physical entity that caused them.
Introspection. The system can
only acquire information via sensory circuits. Therefore it is not inherently
aware of the inner thinking processes. There is an exception, though; certain
match/mismatch conditions must be detected at the neural level and this
information must be made available without actual sensors. The winning results
of the actual thinking processes may manifest themselves as motor actions, which
in turn may be perceived via sensors. Thus the system may become aware of the
products of its thinking processes via external feedback. This is awkward and
therefore a method that enables the system to introspect its mental content
internally is required. This can be achieved by internal feedback that returns
the mental content into the equivalents of sensory percepts like this: Auditory
percepts; inner speech, inner music. Visual percepts; inner imagery. Body
sensors; imagined movements. Inner sensors; emotional states. These inner
"representations" are not independent of each other; inner speech is related to
inner situational representations etc.
Attention. Only a small
number of entities may be actively processed at a time. The selective process is
called attention. Sensed entities are selected by sensory attention, mental
entities are selected by inner attention.
Affordance. A percept that
evokes possibilities for use and action; this necessitates the activation of
cross-connections between various modalities.
Cognition. The association
of auxiliary meanings with percepts, the use of percepts as symbols, the
manipulation of these, reasoning, response generation,
language.
Imitation. Imitation is
the ability to reproduce seen action or heard sounds. This is achieved via
“mirror neuron action”; sensory signals activate proper motor
neurons.
Imagination. Imagination is
the forming and manipulation of conscious mental representations of actions and
entities, which are not sensorily present. Actions and their consequences may be
imagined and evaluated. Imagination and imitation utilize same motor neuron
connections; in imagined “mirror neuron action” the evoking representations have
inner cause instead of sensory origin.
Learning. The acquisition
of mental entities and connections between these. Motor routines
included.
Emotions. Emotions are
seen as combinations of basic system reactions (accept, reject, approach,
withdraw ...) triggered by emotional criteria (good, bad, pleasant, painful
etc). Emotions operate as attention control, motivation, short-cut templates for
style of action. Emotions affect learning. The Haikonen systems reactions theory
of emotions is used.
Language. Language is
seen as a description method for external and internal states of affairs.
Perceived situations may be translated into linguistic description, linguistic
descriptions may be used instead of direct sensory percepts. Inner speech is the
system’s internal interactive narration. The Haikonen multimodal model of
language is used.
Consciousness.
The content of
consciousness is created by perception. The awareness of the environment is
created by the perception of environment. Self-consciousness is created by the
perception of body and its processes and the perception of mental content
(introspection). There are no special circuits or locations that would turn
entering signals into conscious ones. Each circuit operates basically in the
same way whether the overall operation is “conscious” or “non-conscious”. The
contents become “conscious” when the various circuits operate in unison, focus
attention on the same entity as perceived by the various sensory modalities.
This involves a wealth of cross-connections and subsequently the forming of
associative memories. Therefore the “conscious” event can be remembered for a
while and can be responded to and reported in the terms of the various
modalities such as sounds, words, gestures, drawings, written text, etc. A
conscious percept appears as an affordance.
Phenomenality. It is proposed
that the subjective experience “the feel of being conscious” is produced by the
way of operation of the perception-centered system; “the system perceiving
itself perceiving”. This subjective experience is “apparently immaterial”, which
in turn is the direct consequence of the absence of any inner observing core.
The material innards are not inspected for the detection of the inner states,
instead these states connect directly to other states and cause the evocation of
responses. Therefore the material nature of the system is not observed and does
not enter into the mental content. The machinery remains transparent and the
subjective experience would therefore be a content-level phenomenon (like
information carried by modulation) that is related to the dynamics of attention
and the flow of inner representations.
Robots
Just Want to Have Some Fun
Why can't
we explain consciousness? I think that this is because we have approached the
problem from a wrong direction.
True machine
consciousness and robots with conscious minds would seem to be beyond our reach
as long as we cannot answer to the basic questions about human consciousness;
what is it, is it really an immaterial entity, what is this phenomenal part of
it, what exactly is the feel of pain or pleasure. Until we know this there is
little hope to reproduce consciousness in machines in any plausible
way.
Our thoughts and
conscious mind seem to be immaterial. Our everyday experience seems to prove
this beyond any suspicion; we cannot perceive any material processes taking
place when we think. We can see and perceive things and actions out there
directly as such, without any apparent material medium. Likewise we can hear
sounds coming from our environment, again as such. Moreover, our apparently
immaterial percepts have phenomenal qualities. We can feel the heat of the sun,
the wetness of the rain, we can feel pain and pleasure. The apparent immaterial
nature of all this has so far prevented plausible explanations of
consciousness.
How can anything
immaterial arise from the material brain? Many contemporary theories of
consciousness try to equate the processes of the mind to biological brain
processes, to patterns of electrical and chemical neural activity. There has
been some success there, as today brain activity can be monitored by various
methods like magnetic resonance imaging and positron emission tomography. Also
certain neural transmitter chemicals like endorphins have been associated with
the alleviation of pain and the generation of pleasure. However, in this way no
answer may necessarily emerge to the question why the mind appears to be
immaterial, why the mental entities are about something and have phenomenal
flavors, subjective feel.
Why can't we
explain consciousness? I think that this is because we have approached the
problem from a wrong direction. Instead of asking what this immaterial
consciousness could be we should ask why do we perceive the mind as immaterial
in the first place. A successful answer to this question will evaporate all the
other problems and make them redundant.
Our knowledge
about the physical world comes from our senses; seeing, hearing, smell, taste,
touch, etc. Our explanation of consciousness must begin with this; the way in
which the brain represents external information to itself. Indeed, researchers
like Prof. Igor Aleksander and others have seen the ability to create suitable
inner representations from sensory information as an essential prerequisite for
consciousness. But, are these representations immaterial depictions of the
actual entities? If they were not, then surely we would see some material
carrier for them, like some "neural blackboard" or "theater" for our inner eye
to observe, but we do not.
It is known that
the receptors of the senses generate neural signals in response to their
stimulation; these and nothing else are forwarded to the brain. For instance,
each photoreceptor on the retina generates a neural signal that corresponds to
the illumination of that receptor. However, our visual experience does not
consist of odd collections of retinal stimuli or the corresponding neural signal
patterns, instead we appear to see actual visual objects out there, without the
awareness of any related neural processes. Somehow these neural signals seem to
be able to convey the information content only while remaining transparent in
themselves and consequently our percepts and thoughts become to be about real
word entities instead of the neural signal patterns that actually carry them.
This effect, I think, is the key to the essence of our conscious mind. Surely
the brain must be performing a complicated, perhaps even supernatural trick
here, how else could this be explained?
In fact, this
trick is not a complicated one. We can consider a simple experiment that
illustrates this point. What would we feel if we scanned a rough surface with a
rigid stick? It so happens that we would not perceive the vibrations of the
stick as such, instead we would perceive the groove patterns of the surface.
There are no nerve fibers going through the stick and into our brain, therefore
common sense would say that we could only feel the vibrations of the stick
against our fingers. This, of course, is what happens, but these vibrations are
caused by and contain information about the actual roughness of the surface and
this is what we perceive. The rigid stick remains transparent, not because of
any complicated trick, but by the sole virtue of its rigidity, the ability to
preserve vibration information. In a similar fashion the neural signals from the
senses are "rigid". They are able to convey sensory information in a transparent
way. This sensory information is carried like modulation on a radio wave; it is
the music that we hear not the carrier wave. However, the carrier is necessary;
neurons are needed to carry and switch neural signals even though we cannot
perceive these without external means. The
"phenomenal" information is carried by the neural signals, the system will
operate on this information only and not on the physical nature of the carrying
medium. Therefore our thoughts are about something instead of being mere neural
firings. Thinking and reasoning are thus based on the interaction of the carried
information content, the modulation patterns of the neural signals. The brain as a
higher level system does not have to be able to perceive its material basis; the
actual nature of the carrier medium and hardware does not enter into the logic
of the thought flow.
To
understand radio programming we must investigate the contents of the modulation,
the meaning of the transmitted program. The mere inspection of the radio
circuitry will not do; the meaning is on a higher level. In order to understand
speech we do not have to study air molecules. Again it is the modulation of the
sound that matters even though without air molecules there would be no sound. In
a similar way the mental content arises above the actual physical machinery of
the brain. The system will perceive
the apparently immaterial mental content only and may therefore arrive at the
naïve conclusion of the immateriality of the mind.
The concepts of
modulation and circuit transparency are well understood in electronics
engineering. Consequently the consideration of mental entities as modulation
patterns carried by neural signals gives an acceptable explanation to the
problem of the apparent immateriality and aboutness of these entities. However,
the strong phenomenal properties like pain and pleasure cannot be readily
explained in this way.
What is pain?
When we hurt ourselves, a neural signal is transmitted from the affected pain
receptor to the brain and we feel pain. However, this neural pain signal is
actually similar to the other neural signals in the brain. Why then would this
signal be felt as painful when the other similar neural signals like those
originating from the eyes or ears do not very much feel like anything? What is
the specific feel of pain?
The meanings of
sensory neural signals are causally grounded to the outside world, to the
properties of the sensed entities. However, the feel of pain is not grounded in
this way to sensed entities because pain is not a property of a sensed entity.
Pain receptors do not sense pain, they simply sense cell damage and the caused
signal indicates only that pain is to be evoked. The pain signals themselves do
not carry the feel of pain, instead the feel arises from the effects that these
signals have on the system and this in turn depends on the way how the signals
are connected to the system. Thus the feel of pain is not a representation,
instead it is a system reaction. However, we can very well label pain and
describe it verbally, we can associate a linguistic representations with
it.
What kinds of
system reactions would feel like something to the system? Here we must consider
one cognitive mechanism that according to researchers like Prof. John G. Taylor
and others is closely related to consciousness, namely attention. The external
world offers numerous stimuli, but in order to respond properly we have to focus
our sensory attention on the most pertinent set of stimuli at each moment.
Likewise we must focus our inner attention so that a coherent train of thoughts
can arise. Pain and pleasure affect attention strongly. Pain demands attention;
it disrupts any attention that is focused on any on-going task. Obviously pain
signals are transmitted to every part of the frontal cortex where they try to
stop whatever is going on so that something else that might stop the pain could
be initiated. This global disruption of attention is necessary as the pain
signal itself does not know what should be done to stop the damage and therefore
it has to broadcast its message to everywhere and disrupt the attended processes
there. It is exactly this general broadcasting that makes us moan and writhe
when in pain. I consider this disruptive broadcasting as a fundamental property
of pain and I would dare to go to as far as to propose that the subjective feel
of pain is indeed caused by this attention disruption. Thus, if you were the
system this disruption in its various forms would be what you would report as
pain. This proposed link between attention and pain would also explain why pain
may be alleviated by focussing attention heavily on unrelated matters and
suppressing the disruption in this way.
Pleasure's
effect on attention is different. Pleasure signals indicate good conditions that
should be sustained. Therefore no shift of attention is required or desired,
instead attention will be more and more focussed on the activity that produces
the pleasure signals. This would also suppress the initiation of any alternative
actions.
There is nothing
mystical in the apparent immateriality of our minds. The mind appears immaterial
as it operates directly on the carried information and cannot perceive the
material carrier basis. Each piece of information appears different from the
others as the causally grounded meaning is different. The phenomenal pain and
pleasure have their specific feel because they are system reactions, not
representations. Once we understand this we can begin to consider how the mind
works as a system; how the mind utilizes inner speech and imagery, how emotions
arise and values emerge, what motivates action.
This approach to
consciousness opens up the way for the design of sentient robots with apparently
immaterial minds; robots that have the flow of inner speech, inner imagery,
sensations and emotions; robots that feel pain and pleasure; robots that just
want to have some fun.