integrated human computer solutions for the

advertisement
QUEST: integrated human computer ‘solutions’ for the 'wicked' problems we face
Is It possible to Create a Computer That Feels?
Creating a Cognitive Exoskeleton!
# We shouldn’t speak
of solutions just speak
of subjectively
acceptable responses
to ‘wicked’ problems
Steve “Capt Amerika” Rogers, Ph.D., SES/ST
Senior Scientist for ATR and Sensor Fusion
Sensors and Information Directorates
Air Force Research Laboratory
Profs Bauer, Oxley, Mendenhall – AFIT ATR
Profs Raines, Mills, Baldwin, Kurkowski, Galloway, Okolica, Humphries, Borghetti, Grimaila – AFIT Cyber
Matt Kabrisky, Adam Rogers - Quale LLC
Drs. Wilson, Priddy, Williams, Perlovsky, and Wicker, Maj Weir, A. Bryant ,M. McClure, G. Lukens, R. Malhotra, Vogel– AFRL/RY
Dr. Brian Tsou, Janet Peasant, Catherine Harrison, Dr. Laurie Fenstermacher, Bob Patterson, Byron Pierce – AFRL/RH
Dr. R Linderman – AFRL/RI
Pete Collins – AFIT EMAG
Chuck Sadowski – ACC/A8
Don Wunsch – Univ Mo Rolla
Apr 2010
Hidden Units: Mark Derriso (RB), Col Ryer, Tim Lacey, Maj Eyster, Capt Birrer, Cindy Krieger, Capt Dube
Outline
‘wicked’ problems we care about are often
characterized by interaction between sentient
agents
• 1. Premise – computational intelligence abstraction tie
together mini tracks (representations applicable to
‘wicked’ problems - OODA)
• 2. Define agents – data, information, knowledge and
context – key to ‘representation’
• 3. What Turing meant to say – ‘alignment’ key
• 4. Adaptable foe generalization for cross domain
insight – cyber, asymmetric warfare, and cancer
(adaptable -> wicked)
• 5. QUEST – QUalia Exploitation of Sensor Technology –
general ‘solution’ approach to wicked problems
When one of the interacting agents is continually modifying its representation
to compensate for actions being taken against its goals (changing requirements)
Questions welcome as we go through the material or at the end
Slides – ROE – colors
find something that
interest you #
- conflicting requirements and changing requirements suggest
that as you solve part of the problem, you negatively impact other aspects
Wicked Problems
Term introduced in 1967
By C. West Churchman
Suggest many of the issues we are discussing
in this conference are ‘wicked’ – so why are
we ignoring the ‘wickedness’ and
discussing # point solutions for them?
Defining characteristics of wicked problems:
1. The problem is not understood until after the formulation of a solution.
So we shouldn’t speak
2.
Wicked
problems
have
no
stopping
rule.
of solutions just speak
It may just be
3. Solutions to wicked problems are not right or wrong.
of subjectively
subjectively
acceptable responses
acceptable
4. Every wicked problem is essentially novel and unique.
5. Every solution to a wicked problem is a 'one shot operation'to the ‘wicked’ problem
6. Wicked problems have no given alternative solutions.
#Changing requirements often a result of
- By Jeff Conklin
# ‘Solution’ to n doesn’t necessarily help with n + 1
mismatch between real problem and spec for solution
to current instantiation of the problem
It may help but it DEPENDS!
‘cannot be tackled by the traditional approach in which problems are defined,
analyzed, and solved in sequential steps.’
‘most paradigmatical examples of wicked problems come from the areas of public
planning and policy: these include global climate change, natural hazards, healthcare
in the United States and elsewhere, the AIDS epidemic, pandemic influenza,
international drug trafficking, homeland security, nuclear weapons, and nuclear
Solution to box cutter detection didn’t help with PETN
energy and waste.’
and neither solve airline security http://en.wikipedia.org/wiki/Wicked_problem
Mission assurance mini-track
EX) Real mission is to get out of Afghanistan and have
it not become a base for attacks against U. S. – versus – take
Marjah – casted incorrectly achieving one negatively impacts
other! This mismatch between the ‘real’ problem and the
problem specification based on current instantiation results in
changing requirements! #
• 1.) Mission representation
Missions are always in an environment (battle space) that is
fluid and dynamic – the ‘fog’ of war – no plan survives the
fist shot -> ‘wicked’
‘security must focus on mission assurance, not network defense’
Capt Dube - quest
K. Jabbour (lessons in IA)
These make it ‘wicked’ – network defense
is worthless if I can’t assure my mission
‘its not bolt loose or bolt tight it is can I do this maneuver? Can I
perform this mission?’ M. Derriso (ISHM - quest)
Bolt information is worthless
without the mission implication
2.) Military decision making processes –
‘combat actionable information’ – T Wilson (CHLOED driver - quest)
Even if I could extract the concepts from streaming ISR data how
can I insert them into the human decision processes? What are
the military decision processes? How dynamic are they?
Military decision making processes – related quest work
Beyond the OODA Nonsense – think piece Quest group Fall 2009
Anticipatory (predictive) reasoning
focus on anticipation - the entities in the
world possibly to include
intents to allow anticipation Anticipate
ToM
Asset organization, scheduling –
providing Joey insight into the
‘situation’ and efficient control
of interrogation to achieve any
given mission – like find HVT 1 –
taking the streaming sensor data
(RFID, camera, microphone) to
architect a GISTing process to
access relevant
memory/knowledge
Sense
Contextual situation awareness –
how do we capture a
representation of the world
Act
Observe
Decide
Update
Orient
Joey/Operator (Jack also has OODA)
Human-machine interface
Monitor
CHLOED – a system
for closing the loop –
exploits the whole
set of processes - Wilson
Discerning behavior, activities, events (these
initially will be definable events like
‘casing’ or ‘trip-wire’ – we will
investigate making the vocabulary
more robust)
# 1. Persistent sensing doesn’t lend itself to OODA
2. Observe has to be intelligent
3. Observe = Orient = Decide = Cognition
4. OODA doesn’t allow for intuition
5. Integrated human/computer solutions are key
Mental representation for COIN and
Asymmetric Approaches in IW
• Traditional tenets of war in cyberspace –
especially in COIN and asymmetric approaches
in IW – ‘wicked’ nature of problem space
AIFIFT2EA4 – expanded kill chain (#QUEST group ATR article ‘07)
What I should anticipate will changed based on what I ID
Effects assessment must look beyond first-order effects on networks and
information to higher-order effects on systems and missions
Deterrence changes continually based on the adversary – and their changing situation
We can achieve effective deterrence without attribution by increasing the
cost and lowering the benefit to an attacker (K. Jabbour – lessons in IA)
Twitris – Prof Sheth – ‘sensing public opinion using social networks’
IW great driver problem
using social networks to
estimate public opinion
Vision:
Anticipate, Interact,
Find, Identify, Fix, Track,
Target, Engage, Assess
—
Anything, Anytime,
Anywhere
Not Sequential!
These make the problems ‘wicked’ – missions change
as do the resulting ‘kill-chain’ or representation of effects
Wicked problems require wicked ‘solutions’ # subjectively acceptable responses
QUEST: Integrated human/computer ‘solutions’ are ‘wicked’
• Align the representations of both partners
in the solution Old approach one directional arrow!
QUEST: TWO AGENTS
Predictability –
key is to be able
to GIST your partner
Predictability allows
fusion!
Leonardo = Aligning
# Second Look = Aligning
Prof Sheth – ‘to integrate
human and machine perception
the convergence must occur at
this abstraction level … requires
a shared framework’ #
Fundamental Principle of fusion – get what you are fusing into a common representation!
Data, Information, knowledge and understanding must be ‘alignable’ to integrate! Only way
the computer can provide ‘information’ to the human and vice versa.
Aligning to a constantly changing human representation is by definition ‘wicked’
http://jareddonovan.com/blog/wp-content/uploads/2009/03/mental_models-450px.png
2: Agent, data, information, knowledge defined
Some Agents can use context
We will define context as
stimuli from other
Agents outputs/Qualia
(previously, current or
expected qualia) to
communicate via the
context – the two agents
have to be ‘aligned’ thus
it becomes a means to
pass ‘information’
between agents
Some Agents respond to the stimuli
by acting on environment
There is uncertainty in the representation
that is reduced by the use of knowledge = # information
Representation
Generating impact
on representation uses
Knowledge also –
representor
Data
Sensor(s) – (uses knowledge)
Sensors take stimuli and make data
Not all agents require ‘context’ – the most basic agent the ‘atomic agent’ doesn’t have a context
input
Agents exist in an environment that has
a plethora of stimuli -
Stimuli
A(s,k,r)
Agent consist of
sensors, knowledge,
representation, can
be manmade or natural
There are types of knowledge – coded in the sensors determining what it
measures, determining the impact on the representation of the data and
about what it transmits to agents around it – also about how to use context
Agent can have internal Agents –
Compound Agent
Can have hierarchy
within the pool
context
Internal
Agent 1
Internal
…
Agent 2
Representation
Internal
Agent n
Each of these internal agents have sensors/knowledge and their
respective internal representations – they together make
up the representation for the compound Agent
Data
Context is the means to allow the capturing
of the continually changing requirements
Stimuli
more to come on this – required for
all ‘wicked’ problems
Even if an agent could model self can it imbue that to
others? An autism deficiency.
False Belief Test
Software is the creation of the
minds of HUMANS – Correct extraction
and interpretation requires a # ToM
which requires a solution to
#Self. BAD NEWS ToM is ‘wicked’
Required for Alignment!
between agents
Relevance to IDS, terrorist –
intent from activity? Autistic
solutions won’t solve some
important problems!
In that world model we have to also place
our model of others. That model attempts
to infer beliefs and desires and tries to
predict what they will do from those beliefs
and desires and our assumption of what
they are sensing. As ill-posed as this may
seem – it is no more difficult than the ill
posed nature of what all our sensors do –
assumptions are made that are
indispensable but also indefensible other
than historically they have served us well.
ToM is the key to intent
and intent is the key to
# ‘ANTICIPATE’. Solutions
to IW will require an
expanded ‘kill-chain’.
As pointed out earlier ‘anticipate’ is
‘wicked’
https://sfari.org/commentaries/-/asset_publis...
3. Turing and code breaking and
the Imitation Game
ALAN TURING HAD AN APPROACH TO PROBLEM SOLVING
- HE USED THAT APPROACH TO CRACK THE NAZI CODE – HE USED THAT APPROACH
TO INVENTING THE IMITATION GAME – THROUGH THAT ASSOCIATION WE WILL GET
BETTER INSIGHT INTO WHAT THE IMPLICATION OF THE IMITATION GAME MEANING
IS TO COMING UP WITH A BETTER CAPTCHA, BETTER APPROACH TO ‘# TRUST’ AND
AN AUTISM DETECTION SCHEME – and a unique approach to Intent from activity (malware)
Turing, Alan M. "Computing Machinery and Intelligence," Mind, 59
Rogers, S.K., Kabrisky,
M., Bauer,
K. and Oxley, M.
(1950),
433-460
“Computing Machinery and Intelligence
Amplification,” Computational Intelligence, The Experts
Speak (Chapter 3),” New Jersey: IEEE Press, 2003.
Static codes aren’t necessarily wicked but once the human mind is engaged and constantly
changing the types of messages and thus the symbol implication = ‘wicked’
http://images.google.com/imgres?imgurl=http://www.thetech.org/robotics/universal/images
/p13_turing_breakout.jpg&imgrefurl=http://www.thetech.org/robotics/universal/breakout_
p13_turing.html&usg=__6bpeJsLNKV_XkzurPBm3h7nAZZg=&h=398&w=600&sz=
54&hl=en&start=48&itbs=1&tbnid=T4sJOvdlVD7k6M:&tbnh=90&tbnw=135&prev=
/images%3Fq%3DAlan%2BTuring%26gbv%3D2%26ndsp%3D21%26hl%3Den%26sa%3DN%26start%3D42
Meaning Turing not Ox
Prof Oxley suggested maybe the ‘deceit’ aspect is also universal in his thoughts – he lived a life
based on trying NOT to let people break his personal ‘code’ – his sexuality
Estimating Intelligence
Imitation Game – Turing Test
Imitation game = Turing Test
is a ‘wicked’ task – more to
come on this idea but the essence
is that the human mind that the
interrogator is interacting with can
continually change itself in a
non-algorithmic manner – making
the requirements that the interrogator
is trying to decipher continually
changing!
Predicted: after 5 minutes
only 70% chance to correctly
determine interactions with
computer
Turing, Alan M. "Computing Machinery and Intelligence," Mind, 59 (1950), 433-460
Fogel, David B. Blondie24: Playing at the Edge of AI. San Francisco: Morgan
Kaufmann. September 2001
Estimating Intelligence/What
Midway example shows need for
interaction
Code Breaking as an alignment issue
German language is the data into the Enigma encryption system agent!
Nazi
Thoughts
(qualia symbols)
German
language
instantiation
(words symbols)
Enigma
encoded
symbols
RF
transmissions
from sub
The idea is that the Allied code breaker (human) has a thoughts – in their private qualia vocabulary
This person has the desire to ‘understand’ the thoughts that the Nazi Submarine human is trying
to communicate to another Nazi human in a command center (probably in terms of location, date, …)
To extract from the stimuli at any point of this set of sets and processes requires
‘knowledge’ of the symbol set and processes – requires ‘alignment’ with symbols/processes
resident in the entity trying to extract – the goal is to extract the symbols and deduce
their meaning to the transmitter of the symbols by ‘aligning’ with symbols
resident in the extracting entity – so for example when some covert Allies entity captures
the RF transmissions they have to use their knowledge of RF encoding/communication schemes
used by the Nazi subs to be able to extract the symbols being transmitted by the sub – thus
the meaning of the RF transmissions in the sense of which EM wave component is associated
with a given symbol from the sub (meaning in this sense is what the RF transmission
represented) (another example use of the word ‘Eins’ (one) – and its frequency of use in Naval messages).
The result of application of this knowledge and some brute force application of potential other symbol alignment is a
translation of the encoded message back into German language words
Imitation Game #‘aligning’
• The interrogator has thoughts in its private
mentalease = qualia
• Its job is to align its symbol set with those of
the entity ‘A’ (man or computer) and ‘B’
(woman) --- that is to deduce the ‘meaning’ of
the symbols being used by those entities to
those entities and from that deduction
determine which entity is the woman!
Meaning of symbols being transmitted to the
entity that is transmitting them is the
Connection BETWEEN CODE BREAKING AND
IMITATION GAME! If the meaning is consistent
with what I would expect of a woman … or of a trusted node! (Lacey = quest)
Mental Symbols and Representation
Don’t use physics based sensory data to represent experiences
• Mental symbols encode sensory
Turing solution:
experiences both real and imagined,
Would program feel?
Does that belong on list?
forming the basis of a rich and complex
system of communication. Such symbols
We would contend that
‘feeling’ is critical to
can be kept to oneself or expressed to
‘thinking’ – since we will
define thinking as the
others as words or pictures. ‘Mentalease’
generation, maintenance,
and manipulation of mental
symbols = qualia, and that is – language of thoughts
Allows representation of abstract thoughts to be ‘fuseable’
the language of ‘feelings’
• Abstract thought permits contemplation
of things beyond what we can see, hear,
touch, taste or smell.
‘alignment’ = KEY TO INTENT from activity!
Midway example shows need for
interaction
What is unique about human intelligence? By Dr. Marc Hauser
In Scientific American, September 2009
TWO WAY ALIGNMENT CAPTCHA
Completely Automated Public Turing Tests
to Tell Computers and Humans Apart
(CAPTCHAs)
Tie between the imitation game and code breaking – part of
uniqueness of humanity is flexibility of our ability to align
Captcha security – a case study
Yan et al
Thus malware detection is an exercise in code breaking –
the essence of code breaking is deducing the meaning of
the symbols to the entity that is transmitting them
# two way alignment captcha = quest
Animated Captchas
http://blog.case.edu/singham/2009/11/08/dalmation.jpg
4. ‘Magic Bullet’ fallacy for cyber,
Asymmetric war and cancer
• Nobel Laureate P. Ehrlich introduced the concept of #
‘magic bullet’ over 100 yrs ago (compounds to
selectively target/kill tumor cells or disease causing
organisms without negative impact on normal cells).
• Problems in # asymmetric war, cyber warfare or cancer
are related – Is it reasonable to think that there exists
some technological solution that will allow us to have
a warfare ‘magic bullet’ in any of these wars?
• Even worse does the thought of a ‘magic bullet’
dominating our work cause us to not pursue
technology avenues that would bring great value!
Metastasis – ‘magic bullet’ issue
• As a cancer tumor undergoes metastases it changes – and thus tumors
and many diseases like Aids requires a cocktail of drugs – thus with our
analogy to fighting a war against a foe like Al Qaeda you can imagine a
country/countries like Iraq/Afghanistan/Pakistan is as complicated as a
body and as new ‘cells’ metastasize they change – Ex) al Zarqawi the
Prince of Iraq wasn’t following the dictates of Bin Laden et al – thus the
strategy to fight those cells had to be different than the ones in
Pakistan/Afghanistan
• BOTTOM LINE WE SHOULDN’T BE TALKING COUNTER INSURGENCY OR
COUNTER TERRORISM – WE SHOULD BE TALKING ALL OF THE ABOVE – AS
A COCKTAIL OF SOLUTIONS
• In Asymmetric Approaches in IW
PETN detector – chasing the adaptable foe? Does NOT solve
the airline security problem! Airline Security requirements as
we state them will always be continually changing.
Telomerase analogy
Is there something in how body
fights cancer that is useful in our
‘wicked’ fights?
The red ends show regions of telomeres in a chromosome (structure in the cell nucleus containing DNA, histone & nonhistone proteins)
http://openpit.wordpress.com/2009/10/12/immortality/
•
•
•
A pathologist, once remarked that when you do cytology on cancer cells the first thing that strikes
you is how chaotic they look: like someone set off a bomb. In the billions of cells in any critter, that
undoubtedly happens all the time, yet hardly any of the bombs in us turns into a killer because no
matter what they can do, even in the matter of uncontrolled reproduction, unless their particular
chaos includes starting up the telomerase gene, they rapidly senesce, and the malignant cell line
dies out as their telomeres are lost
That’s how nature solved the cancer problem: NO TELOMERASE FOR ANY CELLS EXCEPT FOR STEM
CELLS AND GERM LINE CELLS. If that fails, and it does once in a while, you die. Fighting an immortal
cancer cell line, which can make telomerase, is like jousting with chaos. Thus we have begun to
realize that immortal cancer lines can at best be relegated to the statues of just a chronic, though
ultimately fatal, disorder like ageing (see any life expectancy table.)
Terrorist attacks and the intrinsic metast-abilty of terrorist organizations resemble the animal
cancer system and its possible self defense strategies. But the fact is that animals are successful
most of the time when cellular chaos occurs in some cell, only because of the telomerase time
bomb built in to each cell. That is not the case in terrorist "tumors". So the analogy is that fighting
terrorism will always be jousting with chaos, and history shows that this is what happens. The
equivalent for a terrorist cell to the DNA is the ‘message’ – it is immortal – you can kill off the cell
but the message still exists: The Moving Finger writes; and, having writ, Moves on: nor all your
Piety nor Wit Shall lure it back to cancel half a Line, Nor all your Tears wash out a Word of it -- Omar
KhayyamThe lymphatic system is the web Suggest only hope we have is Information Warfare!
NY times article 29 dec 2009
• December 29, 2009 FORTY YEARS' WAR
• Old Ideas Spur New Approaches in Cancer Fight
– By GINA KOLATA
• Although seemed like a radical idea: Gene mutations
are part of the process of cancer, but mutations alone
are not enough. Cancer involves an interaction between
rogue cells and surrounding tissue. As are all
adaptable foes – the key is preventing the enabling of
the ‘# normal population’ – in cyber or in Asymmetric
war OR CANCER!
Anbar Awakenings worked because
of this realization
5. Layered Sensing challenge
The system is initially expected to
provide a main full-motion video and
12 pre-programmable sub-views. Air
Force officials have previously said that
their goal is to provide up to 30 sub-views
in future generations of the sensor.
Concept encoding is a wicked problem because the
data is so variable and the sensors never exactly
http://www.wired.com/images_blogs/dangerroom/images/2009/02/17/gorgon_stare.jpg replicated – point being the specs are changing
"We are going to be swimming in sensors and drowning in data," Gen Deptula
told reporters at a July 7 Pentagon briefing. (09)
Processing this data requires abstraction to concepts
- current approaches process the raw sensory data
- How to map the # data to concepts?
- How to do that in a general way without having
to predefine in every detail (which results in loss of
flexibility and robustness) –
- Close the Loop! Query to sensing to hypothesis
to new query
http://www.armytimes.com/xml/news/2007/10/gns_spytechnology_071002/100207gotcha_800.JPG
Alternatives to solve problems
• Lots of people – even if we could resource the people they
can’t handle the data flow (#50 bits/sec throughput)
• Autonomous computer solutions – after 50 years we have
failed in automating conceptualization and processing in data
space won’t work
• Answer: integrated human/computer provides a ‘wicked’
solution! => as per prior discussion integrated = ‘alignable’
Cyber, Automatic Target Recognition, Integrated Systems Health Monitoring, Nurse shift
facilitator
http://www.etproductions.com/images/ai.gif
QUEST
Our very definition of what we want for QUEST is
‘wicked’ since it doesn’t imply there is a specifiable
set it has to work over = ‘general purpose’ recognition
ability – QUEST solutions have to be ‘wicked’ solns
There has been over half a century of effort into making intelligent
computer programs. One example of such a program would be
one that could understand the world that it is embedded in. BUT
A GENERAL PURPOSE (one that can respond to the unexpected
query in a variety of applications) RECOGNITION ABILITY DOES
NOT EXIST except in nature
QUEST is a unique approach to representation to provide the ability to
anticipate, detect, distinguish and characterize entities in the real world
via integrated human / machine exploitation.
Entities – IEDs, malware, suicide
bombers, aging a/c and other ISHM,
Conceptual Encoding
cancer, …. ‘wicked’ – incomplete,
QUEST is attacking the
contradicting and changing information
# DATA – CONCEPT GAP
TO GAIN FLEXIBILITY AND
ROBUSTNESS LOST WHEN
PROCESS IN SENSORY SPACE
QUEST
To integrate we have to be able to
align – requires replication of
concept encoding
QUalia Exploitation of Sensor Technology – creating a cognitive exoskeleton
Don’t restrict problem down to ‘selected’ well defined targets and confusers and conops – be able to respond to the
unexpected query – unintended reuse of representation.
To account for the ‘wicked’ nature of generating an internal
representation of a changing world? Qualia = concepts
‘might acquire qualia in the process of being economically encoded or “prepared” into manageable chunks as they are delivered to a
central executive structure higher up in the brain’ – Ramachandran – QUEST view chunks used for meta-representations for sensory
and motor experiences. The
goal is to build a world model
that is consistent and useful not
necessarily an accurate one in
a physics sense.
What ‘perception’ is
subjectively acceptable?
What it is like to
undergo an experience
Why do you ‘feel’ pain?
Individualized!
#Pudding doesn’t have
a taste until you eat it.
Pheung – individualized
basis
Why do you ‘see’
red? # Color is in
your head – light
has NO COLOR! Do
Martians see red?
LSOC important
Redness is wicked
Subjective element
of the experience:
Qualia Theory of
Relativity.
Chords a great example – what you hear is evoked in your mind it isn’t just the notes! Headphones!
QUEST Theory of Relativity
We have spent all our traditional bandwidth on trying to faithfully capture the physical
versus capturing a representation that can be ‘understood’ by a computer. Concept
of transposition (singing in a different key or at twice the speed is irrelevant – still song).
Physical World
Autistic Representation
Red as a wavelength!
Hue - physical
attribute preferential
reflects .7 um
Entity One
Subjective World Model
QUEST Representation
Red as a # concept!
Subjective Representation
Entity One
Hue - physical
attribute preferential
reflects .45 um
Entity Two
Absolute pitch –
1 in 10k
Color quale
link that captures
the relative
redness between
these two entities
captures relative
saturation,
intensity,
hue – axes of
experience
Redness is ‘wicked’ since as I encode that concept
it changes the requirement for ‘blueness’
Horn Theory -
LSOC
Subjective Representation of
Entity Two
Theory of binding – issue
of crowding!
‘What is done by what is called myself is, I feel, done by something greater than
myself in me’ – James Clerk Maxwell on his deathbed, 1879 (user illusion)
Cartesean Theater
Thoughts
Discernible aspects can be used
internally to ‘think’ or
communicated via language
to an aligned agent -
Pretending to be in charge of things
beyond its control – user illusion
Although claims to be making decisions – it
isn’t even there when the decision is made
Qualia associated with representation
of the physical environment and to
represent the thoughts you are having
all in a stable, consistent and useful way
Discernible meant to capture that there is
a difference between one state and the
alternative (blue versus brown) and the
fact that aspects are introspectively
available
For the purpose of this presentation I will define qualia as any discernible aspect
of the illusory Cartesean theater = any aspect of your world model that you are
aware of (meaning you know is part of, meaning you experience that aspect) as being
part of that world model, the fact you can ‘see’ the redness of a car means that
red attribute of your world model is a quale red – any sound you hear (the attributes
of the sound that evoked JND aspects to them are each a quale at that moment) –
any thought you have at that moment you are thinking it is in your world
model as a thought in your mind so there is a quale of thought associated with it AND
thoughts in fact are composed of qualia (the primitives of all thought)
Qualia provide a language to capture the ‘wickedness’ inherent in everyday problems – everything is inter-dependent!
Required for Alignment!
Can you function without Qualia?
What do you lose without qualia?
Physical actions by the critter
Real World
11 mbits/sec
Sensor Data
Libet Soup, #
Intuitive and
autistic
Note related to idea
of Thalamo-cortical loop
Externally evoked via
Libet populated
Physical reality
Qualia based
Cartesean
Theater
Access to memory
via associative est
of similarity,
parallel processing,
fast, (soccer pass),
biases (priming)
50 bits/sec throughput b/w of consciousness
amazing code book vector capability!
Internally evoked via this link
(thoughts)
Logical analysis,
generation and initiation
of empiric test, Veto rights,
controls communication,
maintains thoughts
Pretending to be in charge of things beyond its control – user illusion
Recent article ‘Custer in Cyberspace’
by Gompert and Kugler
Internal representation
No conscious free will
Blindsight Navigation
full reference - de Gelder et al. "Intact navigation skills after bilateral loss of striate cortex."
Current Biology, Volume 18, Issue 24, 23 December 2008, Pages R1128-R1129
news article link - http://www.timesonline.co.uk/tol/news/science/article5385633.ece
youtube video of him going down the hall - http://www.youtube.com/watch?v=nFJvXNGJsws
No buffer,
no multiple alternate responses
Possibly context could be the source of the ‘inference task’ – that determines what an Agent is trying to accomplish – for example
‘context’ could prompt an Agent to ‘perceive’ some aspect of its environment – thus the agent has to gather the appropriate data
to update that aspect of its representation – once this is accomplished the Agent could communicate it out to become stimuli to
another agent
context
Context is the stimuli available
for this agent that are the
outputs of other Agents,
not just their currently
active qualia but also their
previously active qualia
and their expected to be
active qualia --- there has
to be a Kc – a knowledge
associated with how to
process context into
impacting this agent’s
representation
Required for Alignment!
QUEST Agent
Part is introspectively available
part is not = reflexive (also no buffering)
Libet
Generating impact
on representation uses
Knowledge also
Do QUEST agents
require a human?
Do all ‘wicked’
problems require
QUEST agent
solutions ?
Qualia
Data
Sensor(s) – (uses knowledge)
Aq(s, Rq, Rl, k)
The agent can pass along aspects of its
internal representation from both the
Libet (intuition) or Qualia aspects of
its representation?
Stimuli
Knowledge also has types
Ks – sensor knowledge (this
is what controls what stimuli
that gets mapped to data)
Kd – controls how data impacts the
internal representation
Kq – knowledge about other agents so
the appropriate qualia are provided as
output thus stimuli for other Agents
QUEST Cognitive Exoskeleton
Computer aides are limited to well
defined environments and only for
single applications.
•Variations in application or location in
kill-chain (AIFIT2EA4) require reengineering the solution.
•Solutions are reflexive/‘autistic’, with
no ability to learn concepts, adapt to
operating conditions or bound
performance
What are the key
new insights?
(REPLACETHIS BOX AND INSERT
DIAGRAM(S))
Input is never considered without context
(subjective representation of the situation)
and only has value relative to prior, current
or expected experiences (qualia)
• General purpose intelligence is possible with a
representation optimized for exploitation versus
fidelity with physical reality.
•Interaction with the environment is critical.
MAIN ACHIEVEMENT:
Ability to use both memory and knowledge to
hierarchically represent instances to close the loop
around the kill-chain, in a variety of applications, as
relative concepts (qualia) and map them into
plausible narratives (in context) and manipulate
(reason) with that representation to create a stable,
useful and consistent world model.
HOW IT WORKS:
Instead of only using exemplars, prototypes or
models for representing concepts we also use
subjective knowledge of the environment to
populate the representation capturing the
relative characteristics (qualia). QUEST solutions
consist of a quale generating kernel and a set of
processes to manipulate/reason with that
representation. The representation is
characterized by gists and links between the
qualia.
ASSUMPTIONS AND LIMITATIONS:
• Link based representation where the links have
hierarchical relationships can represent any
concept.
• There exists a universal kernel for extracting the
qualia (concept representation).
• A qualia representation can represent itself
within its world model.
QUANTITATIVE IMPACT
Qualia Exploitation of Sensor Technology
• Within any given application
area performance improvements
will be demonstrated and insight
into performance bounding will be
generated.
•The ability to extend the solution
to unexpected queries.
END-OF-PHASE GOAL
STATUS QUO
NEW INSIGHTS
Note characteristics of
QUEST solutions!
Exploitation in a layered sensing
environment
•Computation of ‘self’ and inclusion
in world model.
•Introduction of a general Theory of
Mind facilitating anticipation.
An integrated human/computer solution with the ability to capture and understand concepts around the
kill chain will offer a dramatic engineering advantage over current ‘autistic’ solutions.
QUESTIONS?
‘wicked’ problems we care about are often
characterized by interaction between sentient
agents
• 1. Premise – computational intelligence abstraction tie
together mini tracks (representations applicable to
‘wicked’ problems - OODA)
• 2. Define agents – data, information, knowledge and
context – key to ‘representation’
• 3. What Turing meant to say – ‘alignment’ key
• 4. Adaptable foe generalization for cross domain
insight – cyber, asymmetric warfare, and cancer
(adaptable -> wicked)
• 5. QUEST – QUalia Exploitation of Sensor Technology –
general ‘solution’ approach to wicked problems
When one of the interacting agents is continually modifying its representation
to compensate for actions being taken against its goals (changing requirements)
Questions welcome as we go through the material or at the end
QUEST: integrated human computer solutions for the 'wicked' problems we face
Is It possible to Create a Computer That Feels?
Creating a Cognitive Exoskeleton!
Steve “Capt Amerika” Rogers, Ph.D., SES/ST
Senior Scientist for ATR and Sensor Fusion
Sensors and Information Directorates
Air Force Research Laboratory
Profs Bauer, Oxley, Mendenhall – AFIT ATR
Profs Raines, Mills, Baldwin, Kurkowski, Galloway, Okolica, Humphries, Borghetti – AFIT Cyber
Matt Kabrisky, Adam Rogers - Quale LLC
Drs. Wilson, Priddy, Williams, Perlovsky, and Wicker, Maj Weir, A. Bryant ,M. McClure, G. Lukens, R. Malhotra– AFRL/RY
Dr. Brian Tsou, Janet Peasant, Catherine Harrison, Dr. Laurie Fenstermacher – AFRL/RH
Dr. R Linderman – AFRL/RI
Pete Collins – AFIT EMAG
Chuck Sadowski – ACC/A8
Apr 2010
Hidden Units: Mark Derriso, Col Ryer, Tim Lacey, Maj Eyster, Capt Birrer, Cindy Krieger, Capt Dube
Download