Archive

Archive for July, 2014

Weekly QuEST Discussion Topics and News, 1 Aug

The first topic is our colleague Lizzy S has to provide a presentation on her summer QuEST related work next week so we want to provide her with a forum to get feedback from the group. So we will start by helping her formulate her message associated with Event based content management systems.

The next topic is a couple of articles provided to us by our colleague Robert P. –

The mind’s best trick:
how we experience conscious will
Daniel M. Wegner

• We often consciously will our own actions. This experience is so profound that it tempts us to believe that our actions are caused by consciousness.
• It could also be a trick, however – the mind’s way of estimating its own apparent authorship by drawing causal inferences about relationships between thoughts and actions.
• Cognitive, social, and neuropsychological studies of apparent mental causation suggest that experiences of conscious will frequently depart from actual causal processes and so might not reflect direct perceptions of conscious thought causing action.

Johansson, P., Hall, L., Sikström, S., & Olsson, A. (2005). Failure to detect
mismatches between intention and outcome in a simple decision task.
Science (New York, N.Y.), 310(5745), 116–9. doi:10.1126/science.1111709

• A fundamental assumption of theories of decision-making is that we detect mismatches between intention and outcome, adjust our behavior in the face of error, and adapt to changing circumstances. Is this always the case? We investigated the relation between intention, choice, and introspection. Participants made choices between presented face pairs on the basis of attractiveness, while we covertly manipulated the relationship between choice and outcome that they experienced. Participants failed to notice conspicuous mismatches between their intended choice and the outcome they were presented with, while nevertheless offering introspectively derived reasons for why they chose the way they did. We call this effect choice blindness.

WeeklyQuESTDiscussionTopicsandNews1Aug
WeeklyQuESTDiscussionTopicsandNews1Aug

Weekly QuEST Discussion Topics and News, 25 July

QuEST July 25, 2014

The first topic this week is to allow our colleague Sandy V to present her 10 min talk for the upcoming Computational Models of Narratives conference – then open the floor for 10 min questions she might get asked – then come constructive comments on possible changes to her slides/presentation –

Abstract for ‘Narratives as a Fundamental Component of Consciousness’

In this paper, we propose a conceptual architecture that models human (spatially-temporally-modally) cohesive narrative development using a computer representation of quale properties. Qualia are proposed to be the fundamental cognitive components humans use to generate cohesive narratives. The engineering approach is based on cognitively inspired technologies and incorporates the novel concept of quale representation for computation of primitive cognitive components of narrative. The ultimate objective of this research is to develop an architecture that emulates the human ability to generate cohesive narratives with incomplete or perturbated information.

In a related topic area:

Working with Sandy / Lizzy this week it became obvious we need to carefully revisit our definitions of consciousness and qualia – so I review of the material that most closely aligns to our use of the terms is useful – and for that we commonly have used the dissertation of the John Searle student Cowell –

The next topic is a couple of articles provided to us by our colleague Robert P. –

The mind’s best trick:
how we experience conscious will
Daniel M. Wegner

• We often consciously will our own actions. This experience is so profound that it tempts us to believe that our actions are caused by consciousness.
• It could also be a trick, however – the mind’s way of estimating its own apparent authorship by drawing causal inferences about relationships between thoughts and actions.
• Cognitive, social, and neuropsychological studies of apparent mental causation suggest that experiences of conscious will frequently depart from actual causal processes and so might not reflect direct perceptions of conscious thought causing action.

Johansson, P., Hall, L., Sikström, S., & Olsson, A. (2005). Failure to detect
mismatches between intention and outcome in a simple decision task.
Science (New York, N.Y.), 310(5745), 116–9. doi:10.1126/science.1111709

• A fundamental assumption of theories of decision-making is that we detect mismatches between intention and outcome, adjust our behavior in the face of error, and adapt to changing circumstances. Is this always the case? We investigated the relation between intention, choice, and introspection. Participants made choices between presented face pairs on the basis of attractiveness, while we covertly manipulated the relationship between choice and outcome that they experienced. Participants failed to notice conspicuous mismatches between their intended choice and the outcome they were presented with, while nevertheless offering introspectively derived reasons for why they chose the way they did. We call this effect choice blindness.

Weekly QuEST Discussion Topics and News 25 July

Weekly QuEST Discussion Topics and News, 18 July

QuEST 18 July 2014

1.) We will continue discussing some work by our colleagues in the Sensors directorate associated with QuEST applications to SAR and EO data exploitation and specifically tracking. The QuEST position going in is that a dual process tracker (Type 1 and Type 2 components) can out-perform a single process data driven tracker – but this week I want to caveat this statement – it is not clear to me that a Qualia based tracker should outperform a data based tracker on a measure of track accuracy – maybe what we should expect is a qualia based tracker is more robust (holding onto tracks under dramatic variations in lighting / weather and that it requires far less computational bandwidth (recall the 50 bits/sec throughput sub-tenet)) – … the point being we need to discuss what we would expect to gain by a Qualia based tracking more carefully. I suggested that advances like feature aided tracking (FAT) and context aided tracking (CAT) support the position that more than kinematics is required in more complicated operating conditions. But I would suggest that a better Type 2 representation, more QuEST like could do even better since it would not be so bound to the variations in the data – and thus the need to define these measures – the idea being like we’ve pushed for QuEST solutions as an approach to reduce the space of the unexpected query – so for tracking we need to define the UQ – and that will be those operating conditions where current approaches to tracking don’t generate acceptable responses / outputs – then blend these with the qualia based tracking to demonstrate improved performance through these previously problematic operating conditions thus demonstrating the reduction in the UQ space. But current approaches using those ideas are still focused in the measured data world versus attempting to generate a hypothetical representation simulation based / situation / structurally coherent representation and blending that with the data driven tracking. Work that attempts to track through obscurations were a start consistent with this idea– but again they do not attempt to use our guiding tenets for the add-on representation / processes.

a. The SAR / EO fusion aspect should also be considered – the fact that at the ‘cartoon’ level entities and their relationships are all that matters and the sensing modality that the source information came from is washed away – at the cartoon level qualia are defined by their relationships with and how they can / do interact with other qualia – and even this representation should be squeezed to force maximal compression while still providing acceptable responses for the operating conditions experienced – so in an EO image the quale that is associated with a given vehicle moving down the road at a particular Geo location at a particular time but keep in mind in qualia space aspects like geo – location and even time are qualia and are only defined by how they are related to other qualia and how they can interact with other qualia – and they are only experienced when they are the source of the experience -> they are the source of the qualia used to achieve the goal of tracking – so let’s take an EO image and define all the qualia that could be evoked from it (then squeeze down to a subset that allows maintaining track – keeping in mind this image is not taken out of context – it is taken as one image in a sequence of data (EO and SAR) – but within the frame there are blobs – we have defined blobs before as parts of a sensory steam that are submitted for qualiarization based upon their brightness / color / motion / texture for example in the visual stream – since one of the means to blob part of the sensory data is to attach to a portion of the stream a cohesive movement relative to parts of the data stream adjacent to it

2.) The discussion above also suggest we need to return to our previous discussions on ‘meaning’, ‘blobs’ and computing with qualia

3.) Another topic that came up this week is work within the Human Effectiveness directorate on ‘synthetic team-mate’ – we want to have a discussion about that data and its potential use in our research.

news summary (4)

Weekly QuEST Discussion Topics and News, July 11

1.) We will start by discussing some work by our colleagues in the Sensors directorate associated with QuEST applications to SAR and EO data exploitation and specifically tracking. The QuEST position going in is that a dual process tracker (Type 1 and Type 2 components) can our perform a single process data driven tracker. I would suggest that advances like feature aided tracking (FAT) and context aided tracking (CAT) support this position. But I would suggest that a better Type 2 representation, more QuEST like could do even better since it would not be so bound to the variabilities of the data. But current approaches using those ideas are still focused in the measured data world versus attempting to generate a hypothetical representation and blending that with the data driven tracking. Work that attempts to track through obscurations uses this idea and then often confirms the results using features / context – but again they do not attempt to use our guiding tenets for the add-on representation / processes.
2.) We also want to discuss some of the ideas in the current draft of the Erik Blasch led effort for an article for NAECON. *** we might defer much of this discussion as Eric will be out on Fridaywelcoming into the world his new child! ** His bold attempt to get down on paper how what we are doing is related to fusion and narratives inspired some dialogue that the group may want to chime in on. The first topic is where does QuEST fit in the mission space when we are talking PCPAD. Specifically can we achieve an artificially conscious decision aid if we don’t allow that agent / agents the ability to interact with the sensing apparatus. Another topic is where does QuEST fit in the levels of information fusion – there are some obvious places but maybe we need to articulate how it could fit at all of the levels and thus be more integrated in the system of systems. Then there is the discussion of activity based intelligence versus object based production versus a far more general approach to ‘event based content management’ or what I would like to call ‘Qualia based content management’. Then there is the discussion on mapping our construct onto the situational awareness levels. Then explaining the QuEST tenets in a manner a non-QuEST person could get the gist of them. Eventually leading to a QuEST model for information fusion.
3.) The third topic is a continuation of the discussion of implementing QuEST solutions using existing cognitive architectures and their infrastructures. Last week we discussed briefly a recent publication by one of our colleagues Robert Patterson and Kennedy – Modeling Intuitive Decision Making in ACT-R. the reason I want to discuss that article is it has many of the facets we’ve been discussing associated with the proposed work of our colleague Sandy V. specifically we want to use the Patterson approach to interfacing a virtual world to ACT-R and some of their use of the existing ACT-R modules.
a. Abstract: One mode of human decision-making is considered intuitive i.e., unconscious situational pattern recognition. Implicit statistical learning, which involves the sampling of invariances from the environment and is known to involve procedural (i.e., non-declarative) memory, has been shown to be a foundation of this mode of decision making. We present an ACT-R model of implicit learning whose implementation entailed a declarative memory-based learner of the classification of example strings of an artificial grammar. The model performed very well when compared to humans. The fact that the simulation of implicit learning could not be implemented in a straightforward way via a non-declarative memory approach, but rather required a declarative memorybased implementation, suggests that the conceptualization of procedural memory in the ACT-R framework may need to be expanded to include abstract representations of statistical regularities. Our approach to the development and testing of models in ACT-R can be used to predict the development of intuitive decision-making in humans.
4.) We want to extend the discussion this week to include another Kennedy article that implements dual process solution using ACT-R. ‘Integrating Fast and Slow Cognitive Processes’
a. Abstract – Human reactions appear to be controlled by two separate types of mental processes: one fast, automatic, and unconscious and the other slow, deliberate, and conscious. With the attention in the literature focused on the taxonomy of the two processes, there is little discussion of how they interact. In this paper, we focus on modeling the slower process’s ability to inhibit the fast process. We present computational cognitive models in which different strategies allow a human to consciously inhibit an undesirable fast response. These general strategies include (a) blocking sensory input, (b), blocking or interrupting the fast process’s response, and (c) slowing down or delaying processing by introducing additional task. Furthermore, we discuss an approach to learning such strategies based on the inference of the causes and effects of the fast process.

Weekly QuEST Discussion Topics and News July 11

Weekly QuEST Discussion Topics and News, 3 July

1.) We want to start this week discussing modern approaches to collaboration – specifically we’ve asked our colleague Dan Uppencamp to come in and discuss with us his Google embedding time and the tools he used (Google apps) while working in the Earth Engine group out at Mountain View. The reason we want to have this discussion is I would like to encourage us to attempt to use those tools in the development of the QuEST white paper and some of our QuEST research probably using the RDT&E network versus NIPR.
2.) The Second topic is a continuation of the discussion of implementing QuEST solutions using existing cognitive architectures and their infrastructures. This week I’m excited to discuss a recent publication by one of our colleagues Robert Patterson – Modeling Intuitive Decision Making in ACT-R. the reason I want to discuss that article is it has many of the facets we’ve been discussing associated with the proposed work of our colleague Sandy V. specifically we want to use the Patterson approach to interfacing a virtual world to ACT-R and some of their use of the existing ACT-R modules.
a. Abstract: One mode of human decision-making is considered intuitive i.e., unconscious situational pattern recognition. Implicit statistical learning, which involves the sampling of invariances from the environment and is known to involve procedural (i.e., non-declarative) memory, has been shown to be a foundation of this mode of decision making. We present an ACT-R model of implicit learning whose implementation entailed a declarative memory-based learner of the classification of example strings of an artificial grammar. The model performed very well when compared to humans. The fact that the simulation of implicit learning could not be implemented in a straightforward way via a non-declarative memory approach, but rather required a declarative memorybased implementation, suggests that the conceptualization of procedural memory in the ACT-R framework may need to be expanded to include abstract representations of statistical regularities. Our approach to the development and testing of models in ACT-R can be used to predict the development of intuitive decision-making in humans.

Weekly QuEST Discussion Topics and News 3 July