Archive for June, 2010

QUEST Discussion Topics for June 24th

QUEST Discussion Topics

June 24th 2010

Tomorrow’s meeting will be pretty open in terms of material on the agenda, if anyone has anything they would like to talk about we welcome any and all subjects of interest.  If there are no items brought up, here are some areas that we are still actively thinking about and would always appreciate more discussion on.

1.) Sloman architectures – can see the blog for the original piece, also attached to this message.  Email me for slides.

2.) Aleksander work on designing conscious systems – paper attached here.  Again, email if you would like slides.

3.) Color constancy – a subject that was brought back up as we discuss conceptual learning and encoding, can see attached word doc for some background material.

4.) Possible adventure into Second Life if we can get a viewer installed on the machine in the CCR Conference Room.  Could lead us into discussions of what is possible in terms of exercises/tutorials to help kids learn math and science concepts.

QUEST Discussion Topics and News June 24

QUEST Discussion Topics June 18,2010

Quest Discussion Topics

June 18, 2010

1.) Awareness and Consciousness – we want to return to my attempt to distinguish aspects of our internal representation of the world by defining these terms in a way so we can use those definitions to communicate impacts to some of the architectures that we are considering for QUEST solutions. 

a.     We will define consciousness as the generation of qualia within the context of a complete ‘narrative’ – a complete world model (what Prof Barsalou calls a ‘situation’).  Qualia are the introspectively available aspects of the internal representation.  You can’t have a quale out of context.  It has to be integrated within a ‘narrative’ = ‘situation’.  Any aspect of that internal representation may or may not be ‘true’ with respect to the reality of the world.  There is no way for us to know whether our qualia are ‘real’ or not since we only have access to them – we don’t have access to the world just our internal representation of that world. 

b.     We will define awareness as the agreement between the internal representation and the reality of the world – if you will the mutual information between the two. 

c.      By this definition you are always ‘aware’ of things you are ‘conscious’ of as being qualia,  (when I see red on a coke bottle the red I see is the quale evoked by the external stimuli (photons) and so there is always perfect agreement ‘awareness’ of my quale (consciousness of the redness) and the fact that red is my quale associated with the color of the coke bottle top) so my awareness of this as an experience is high) BUT aspects of your conscious representation can be incorrect and thus have misleading ‘awareness’ – so for example in the coke top we could do the Land Mondrian idea and with the right illumination an  surrounding shapes/colors the same top could evoke the quale of ‘orange’ – I would have the conscious impression that the top was orange – but that isn’t true and so my awareness of the FACT that the top of the bottle has the color red is low.  Similarly for the color constancy example we use in our power points – I’m conscious of the two squares on a checkerboard being of different brightness although there is low awareness of that calculation since they are really ARE the same brightness – another example BIIDs (body integrity identity disorder) where you are conscious of a limb (like your leg) being a leg = there is awareness of it as a leg — BUT your consciousness (qualia) of it being ‘your’ leg is incorrect – it doesn’t feel like your leg – there is a misleading quale being formed thus low ‘awareness’ with respect to the ‘self’ quale associated with the leg. 

d.     (we have a powerpoint on consciousness and awareness using these definitions and some web material on color constancy available for those who request it).

2.)  Architectures for QUEST – we want to return to the work of Sloman and virtual machines as a discussion of the characteristics of ‘conscious’ architectures.  The idea that introspection is a perception and as such we are only conscious of those aspects of internal processing the evoke distinct qualia and thus introspection can be deceptive with respect to architecture implications.   Architecture strata of interest are (powerpoint and article):

a.     Reactive

b.     Hybrid

c.      Deliberative

3.)  Also the work of Aleksander on designing conscious systems.  Specifically his five axioms for what consciousness ‘feels’ like:  what important elements of consciousness feel like and how they may by translated into neural mechanisms (power point and article):

a.     presence,

b.     imagination,

c.      attention,

d.     volition

e.     emotion.

4.)  STEM project discussion

Unmanned article

unmanned proving a misnomer

Dr. Barsalou’s slides

From his presentation at AFIT, 6/4/2010, “Grounding Knowledge in the Brain’s Modal Systems”

@Barsalou – GKBMS- 06-10 – SitCon

(file is ~11MB)

Link to more recent Barsalou paper

Titled ‘Grounded Cognition’, this one is more recent and thus a little more relevant than the old ’99 paper on PSS that we were going off of.


Walkaways from Dr. Barsalou’s presentation

Some walk-away Barsalou QUEST tenet impact points:
1.)  All concepts are only perceived ‘situated’!  this is what we have stressed (or stated in not a clean a manner as Barsalou) as all qualia have to be integrated within a ‘narrative’.  Never in isolation.  This is a challenge to enforce and architect.  Concepts are not captured or represented in a vacuum always situated.  BASIC UNIT OF COGNITION IS A SITUATION.  When accessing a concept the links cause the imagination of a ‘situated’ narrative.  That can include other features not in the sensory data and other entities and even mental affect states.  Interesting aspect of situating is that if the concept is of a concrete object the situating becomes fairly specific in terms of time and space where abstract concepts seem to fairly distributed.
2.)  Barsalou suggest a single narrative – ‘simulation’ explanation versus what we’ve suggested as a set of multiple plausible narratives competing for attention – the winning ‘narrative’ becomes the illusory cartesean theater in our view.
3.)  The idea of a simulation – versus a playing in a theater of the data from the sensory surface – wasn’t clear where he would say the ‘experience’ lives – we would say in the ‘simulation’ representation versus the connecting to the existing sensory motor activation.
4.)  Interactive AI – versus open loop concept formulation, interchange with prof flach, ‘closing the loop’ via the action / decision resulting is a common agreed upon principle, we would emphasize that one of the most critical ‘actions’ that the result can be observed by  the generating critter is the concept forming action.  The point being instead of concepts being formulated via an open loop data in concepts out process it requires a continual integrated if you will ‘situated’ concept formation.  The tentative conceptually based narrative is checked for consistency within the ‘situated’ rest of the ‘narrative’.  Based on closing that loop some uncertainty in how well it ‘fits’ might cause a re-categorization.
5.)  The conceptual system to Barsalou is the language that is used to represent knowledge (associated with gathered experiences in the world).  He also suggest that in online processing it is used for ‘high-level’ perception — we would say it is the language for ALL perception (even down to redness). 
6.)  Offline processing – the conceptual system is the language used for memory, language and thought – conceptualization of entities not present.
7.)  Learning – the conceptual system provides the language for interpretation of novel material.
8.)  Memory – getting to knowledge that is stored – done through simulations that may be below the level of consciousness (Libet), but his point of always being partial and distorted when compared to the actual prior experience (specifically talks about exemplars or averages of exemplars – this we might not completely agree with as simulators have far more flexibility than recall of stored templates).
9.)  Affect qualia are also accomplished by simulators in his formulation – so the simulators are not just for external reproductions but also for internal states.
10.)                     Abstract objects are also represented via simulation. 
11.)                     Links and embodied cognition – the experiments he went through for example the work of Zwaan et al 2002 – where you measure reaction time and you have the subject read a sentence like ‘the bird sat quietly in the tree’ – note the state of the wings is not in the sentence – but then you flash an image of a bird either in flight or just a bird in isolation with its wings folded – you then query whether the image is consistent to the sentence just read – quicker recall when consistent – we would say that when you read the sentence you generate a ‘narrative’ – situate the sentence – embody it into an acceptable narrative – and if then the picture fits in your narrative can be quickly determined – the extra parts of the narrative that weren’t in the sensed data was accomplished via the links.  This effect seems to be across modalities – reading words excites motor cortex or somatosensory cortex as appropriate – means to get to narrative is to excite the cortex as appropriate.