|
| Scientific Foundation |
|  
Relevant Publications |
|  
Theory Presentation |
|
Contact Us
|
|
|
PMM Artificial Intellect Systems
人造智力系统
|
Our technical base is universal Proto-Mind Machine (PMM) - a general human intellect model.
Proto-Mind Machine accepts mixed
Text-Video-Audio stream
as input and generates a corresponding Text-Audio stream as an output.
Rules for output generation are being created automatically when PMM processes its input stream.
The main PMM algorithm for information processing is of a neural network type that makes PMM reactions looking like human ones - through the resemblance of current and all known contexts.
PMM works in two modes: Listening when it accumulates knowledge and Answering when it generates the output using its acquired knowledge.
|
We design artificial intelligence systems for various domains through setting up the relevant manuals in combination with learning-with-teacher method.
After learning is done PMM has all necessary knowledge and begins to apply it to advice users without external help.
PMM applies its knowledge, searches for data, asks questions and even makes
logical inferences
if it was taught to do that.
|
PMM can be used in any system to eliminate extra personnel that conducts relatively simple intellectual tasks.
Various training systems are good examples of such applications.
In general, any
autonomic robotic systems
which require the elements of
artificial general intellect
are in PMM application domain like PMM
Assistant.
|
******************************************************************************************
|
|
***************
|
|
|
|
|
|
|
Proto-Mind Model Basis
|
|
| Our method: |
Creating the context connections
|
PMM virtualy connects any input natural language message with its messages-predecessors.
These previous messages are also used as the templates of some kind for the future dialogues.
PMM accepts all input messages and resolves the possible contradictions in a human-like manner.
After gaining a necessary experience, PMM's answers become stable and ready for practical use.
|
|
| Background: |
Sophisticated model of brain network
|
|
Mathematical basis of PMM is a result of scientific research in the field of neural network models.
PMM works in the same way as human brain does.
The main research results are as following:
|
| -
The internal human language is universal and independent of any particular external form it can accept.
It is some seed or proto-language.
All natural languages spoken by humans are its external manifestations.
So are English, Greek, French, etc.
The formal language of modern mathematics is among them as well.
|
-
Neural networks of human brain are the origin of proto-language, and their properties predetermine the generic proto-language structure.
|
|
| Artificial Intellect: |
Making an Intellectual Model of a Person
|
PMM is a sophisticated tool that helps to create a personal assistant for a permanent intelligence support.
In order to create PMM assistant we consider a human as an abstract intellect and apply PMM to create a copy of it.
|
|
EXAMPLES
|
Following examples contain a set of PMM responses to the questions and corresponding situation descriptions (contexts) which are relevant to some fiction combat environment.
The manual for all examples was entered in PMM once prior the tests. Find nine test cases bellow.
The general structure of the examples includes:
• part 1: the original situation descriptions which comprise the expected PMM reaction marked with exclamation sign,
• part 2: the description of situation and a question to PMM,
• part 3: a real response of PMM.
Red
color highlights the phrases and/or their parts which were intentionally omitted in parts 1 and 2.
Blue
and
green
colors show the difference between original situation description in the part 1 and the corresponding description in the part 2.
Brown
indicates a changed sentence order.
Square brackets enclose the internal PMM responses (PMM internal speech).
The test synopsis:
1. Tests A and B have the same question asked in the different contexts.
2. Test C shows the simplest case of logical inference made by PMM through using its knowledge.
3. Tests D, E and F demonstrate a variety of context recognition like missing/alternated words, omitted or inserted phrases, modified questions, etc.
4. Test G shows the PMM behaviour in the case it doesn't have enough knowledge to response properly.
5. Finaly the test H demonstrates the usage of the context for some kind of template creation.
|
• Test-A: the question "what's going on?" (the first context)
Original text:
halfway
down the road the
lead
vehicles came to a
sudden
halt
Campbell couldn't see what the
problem
was through the Camel's
starred
windshield
Marchie? Tam?
what's going on?
bots see a checkpoint !
came Marchand's voice from the back seat
looks like the local constabulary !
Provided context:
down the road the vehicles came to a
jerky
halt
Campbell couldn't see
clearly
what the
reason
was through the Camel's windshield
what's going on?
PMM answer:
bots see a checkpoint
[ came Marchand's voice from the back seat ]
looks like the local constabulary
|
|
• Test-B: the same question "what's going on?" (the second context)
Original text:
Daz swore under his breath
and at length
just then
however the gun fire
that
had become a monotonous din,
trailed off in
echoes
simultaneously Marchand spoke in Campbell's ear
hear me?
Marchie, we hear you
what's going on?
we've rerouted into infrared through the swarmbot net, Sarge !
the EM bands are all full of noise
I think the insurgents are bombing the crap out of the city's communications grid !
but our equipment is shielded
great I meant what's going on here, right now?
why's the firing stopped?
oh! the strikebots put down the enemy bots, Sarge
Provided context:
Daz swore under his breath
however the gunfire had become a monotonous din
followed
by echoes
simultaneously Marchand
muttered
something
in Campbell's
headset
what's going on?
PMM answer:
we've rerouted into infrared through the swarmbot net, Sarge
[ the EM bands are all full of noise ]
I think the insurgents are bombing the crap out of the city's communications grid
but our equipment is shielded
|
|
• Test-C: A kind of logic inference
Original text:
five meters ahead of the Camel, some
children were laughing as they chased
two palm sized
helicopter drones
the drones were part of the Camel's standard kit
it would have been odd to see a Camel that didn't have one or two of them circling around it
these ones bounced
up into
the air
,
out of
the children's reach
that's momentarily putting them too far from the ground to sniff out any nearby explosive devices
the drones are to be back and followed Camel !
Provided context:
children were laughing as they chased
tiny
helicopter drones
these ones bounced
in
the air
far from
the children's reach
PMM answer:
[ that's momentarily putting them too far from the ground to sniff out any nearby explosive devices ]
the drones are to be back and followed Camel
|
|
• Test-D: General Questions
Original text:
it was offset from the road by wire fencing and a row of grotesquely rusted oil tanks
a
good
sixty feet of empty lot
spread out around
the plant
itself, which was two stories high and largely windowless
they pulled in behind a screen of sun withered trees and everybody began
piling
out of the vehicles
Campbell activated the strikebots
an action that was relayed automatically to the operations situation display !
they rose up in whirs of dust
Provided context:
sixty feet of empty lot
surrounded
the plant
they pulled in behind a screen of sun withered trees and everybody began
getting
out of the vehicles
Why Campbell activated the strike bots in this situation?
PMM answer:
an action that was relayed automatically to the operations situation display
|
|
• Test-E: General Questions
Original text:
the swarmbots had rolled and scuttled their way into positions
where they could
deploy
their Doppler radar and terahertz sensors
she registered
a number of heartbeats inside the generator building, only
a half dozen or so
in the main plant beyond
the plant personnel must be hidden somewhere deeper inside, out of reach of her sensing gear !
It was a maze of metal in there, after all, so the radar didn't work at all well
Provided context:
the swarmbots had rolled and scuttled their way into positions
where they could
use
their Doppler radar and terahertz sensors
Desai recorded
a number of heartbeats inside the generator building
only
six or seven
in the main plant beyond
where the plant personnel is hiding?
PMM answer:
the plant personnel must be hidden somewhere deeper inside, out of reach of her sensing gear
[ It was a maze of metal in there, after all, so the radar didn't work at all well ]
|
|
• Test-F: General Questions
Original text:
Campbell looked more closely
the two trucks were cabless
behind them was a
herd of squat
multi wheeled platforms piled with
crates and equipment
these low slung cabless vehicles were aptly named scarabs
each
one
was surmounted by a strike bot
in sentry mode
Ops, where are
our
reinforcements?
reinforcements should be arriving now !
said PMM in his ear
only the bots and automated vehicles got through !
Provided context:
the two trucks were cabless
Campbell looked more closely
behind them was a
set of
multi wheeled platforms piled with
cargo
each
scarabs
was surmounted by a strike bot
Ops, where are
those damn
reinforcements?
PMM answer:
reinforcements should be arriving now !
|
|
• Test-G: Unknown Question
Provided context:
Where could Kabadi see the circling UAVs?
PMM answer:
{No answer}
Loading a new text:
Mastan Nouria and Idris Kabadi had found their progress blocked by one of their own flash mobs
Kabadi stepped out onto the running board of the truck to see if he could locate a path through
below him Nouria was chewing a nail nervously
Kabadi could see a flock of UAVs circling the white tower they had so recently left !
the British had snuck around their roadblocks and were at the compound now
the streets were clear between them and the place where Kabadi and Nouria found themselves
Provided context:
Where could Kabadi see the circling UAVs?
PMM answer:
Kabadi could see a flock of UAVs circling the white tower they had so recently left
|
|
• Test-H: Parameter substitution
Original text:
as with all the buildings
and streets
in the industrial sector
the water treatment plant was dotted with smart dust sensors
they were supposed
to be able
to tell if anything bigger than a
dog
wandered into the area
right at the moment they were
obstinately
reporting all clear
even as insurgents with guns ran right over them
it's some kind of
_worm
tailored to sensor nets
it impedes the system functions properly
the problem is that smart dust components use a
worm
like method tailored to sensor nets !
Provided context:
as with all the buildings in the industrial sector
the water treatment plant was dotted with smart dust sensors
they were supposed to tell if anything bigger than a
cat
wandered into the area
right at the moment they were reporting all clear
We have some kind of
virus
tailored to sensor nets
PMM answer:
[ it impedes the system functions properly ]
the problem is that smart dust components use a
virus
like method tailored to sensor nets !
|
|
About Us
|
|
|
|
Alexander Borzenko
|
Architect/Developer
|
| 1985 |
Got a Master degree in Applied Mathematics and Computer Science |
| 1985 - 1989 |
Worked on various neural network models for visual image recognition |
| 1987 |
Defended Ph.D. thesis related to mathematical model of reliability of missile control systems |
| 1987 - 1991 |
Worked on mathematical neural network model of symbolic data processing |
| 1989 - 1995 |
Worked on visual image recognition and processing system for space satellite |
| 1990 |
Defended a post-doc thesis on symbolic data processing through neural networks |
| 1991 |
Got Ph.D. in Mathematic Modeling |
| 1995 - 1998 |
Worked on variety of visual recognition systems like face and signature recognition |
| 1995 - 2000 |
Worked on computer methods for post-doc result implementation |
| 2000 - 2005 |
Work on practical implementation of the symbolic data processing methods |
| 1998 - 2008 |
Worked on general theory of symbolic data processing in human brain |
|
|
|
| Andrew Borzenko |
Software Developer
|
| 2004 - 2009 |
Undergraduate student at the University of Toronto |
| 2009 |
B.Sc. in Computer Science, University of Toronto |
|
|
|
|
|
E-mail:
|
alex.borzen@gmail.com
|
|
PMM and Artificial General Intellect
|
|
PMM creates a naturally formed logic system with only substitution rule (modus ponendo ponens).
|
Generally any sequence of phrases can be stored in PMM language database. This feature serves as a means for the inference machine setup. Being properly set, PMM works like some kind of artificial general intellect.
|
|
For example, PMM can process a common syllogism.
|
|
|
Major Premise:
|
|
All humans are mortal
|
|
PMM’s context
|
|
(1)
|
|
|
Minor Premise:
|
|
Socrates
is a human
|
|
PMM’s key phrase
|
|
(2)
|
|
|
Conclusion:
|
|
Socrates
is mortal
|
|
PMM’s inference
|
|
(3)
|
|
This syllogism PMM’s interpretation really works for anybody’s name (in italic) mentioned in the phrase of type
(2)
within the context
(1)
.
Broadly speaking, PMM considers every text fragments like
(1-3)
as some “syllogism” allowing us to vary the syllogism structure relatively its formal definition.
For instance, we can say “Socrates belongs to the human race” instead of
(2)
without causing a failure.
On the contrary, the phrase “Socrates is
not
a human” won’t produce the conclusion
(3)
because of automatically detected "not" semantics.
|
|
The syllogism
(1-3)
is not an exception, and PMM can implement all known others.
Syllogism selection by PMM depends on the context in which a key phrase has been entered.
No random syllogism selection normally happens.
But if it does happen the PMM reaction should be corrected (preferably at the learning stage).
|
Relevant Publications
|
|
| [1] | A. Borzenko,
 
“A Neural Mechanism for Human Language Processing”.
To appear in Neurocomputing, special issue "Artificial brain", 2010 (June-July).
|
|
[2]
|
A. Borzenko,
 
“Neuron Mechanism of Human Languages”.
In: IEEE Proceedings of International Joint Conference on Neural Networks (IJCNN), Atlanta, 2009, ISBN: 978-1-4244-3548-7.
(reference)
|
|
[3]
|
A. Borzenko,
 
“Language Processing in Brain Networks”.
In: Proceedings of the First AGI Conference, IOS Press, Volume 171 Frontiers in Artificial Intelligence and Applications, Edited by: P. Wang, B. Goertzel and S. Franklin, February 2008, 520 pp., ISBN: 978-1-58603-833-5.
(reference)
|
|
[4]
|
A. Borzenko,
 
“Formal Neuron System for the Natural Language Analysis”.
In: IEEE Neural Networks Proceedings, 1998, p.2561-2564.
|
|
[5]
|
A. Borzenko, “Associative Recognition of Signals by a Network of Formal Neurons”. In: Automatics and Mechanics, No. 1, 1985, Moscow, Science, p. 95-100.
|
|
|
|
|
|