Archive

Archive for October, 2010

QUEST Discussion Topics, 10/29

October 29, 2010 Leave a comment
QUEST Discussion Topics, 10/29

1.)  The initial discussion topic for this week will cover Capt Amerika’s talk on Flexible Autonomy (see notes below)

2.) With any remaining time, we will move on to a presentation on designing ‘conscious’ systems and what we envision a QUEST solution to look like.


Quest for Flexible Autonomy  Summary Notes:

Oct 2010

Capt Amerika

Issue:  Swimming in sensors – drowning in data (myth of ‘unmanned’ ISR – layered sensing, cyberspace has a human dimension, Integrated systems health monitoring, medical, …)

Current issue: myth of coming of intelligent machines/autonomy (resulting from great advances in processing throughput) all are based on the assumption that some miraculous solution to concept encoding (abstracting from pixels to information for example) will appear out of maturation of currently known machine intelligence approaches.  In all of these areas we are swimming in sensors and drowning in data.

Premise of this talk:  missing fundamental science (and math) for integrating multiple agents (some human some machine) will be required to solve these wicked (continually changing requirements) problems.   The second aspect of the proposed solution is that concepts are only definable in context (situations are the fundamental unit of cognition).  Situations can only be extracted from the environment with a closing of the loop around the sensor (query to sensing, proposed hypothesis situation, to new query – requires interaction).

Major points of talk:

1.)  RPA – CAP requires ~160 people, in space turning around a re-useable spacecraft takes 1.2k people and takes months (need to get to couple hundred and reduce turnaround to days), cyber Host based security services data is routinely discarded after recording due to the mismatch on PED capacity.

2.)  Integrated human machine solutions required – integration requires an ability of the human to ‘gist’ what the computer is doing / why it is doing it / and what it will do next – as well as the computer agents having a representation that answers these same questions about the human – requires a common framework.

3.)  Common framework – generalized agent definition (information processor – with precise definition of data, information, knowledge that is agent specific), extension of that formulation to definition of quest agents (incorporates what is known about human information processing) to include Theory of Mind (ability of humans to represent other human’s representation of the stimuli – thus the ability to predict the other agents ability).

4.)  Concepts in quest agents have a level of the representation that include a subjective (situation based) aspect.  ‘red is a concept not a wavelength nor a range of wavelengths’.

5.)  There exists a common set of math to describe the combined human – machine system of systems.

6.)  The ultimate goal of the common math is to predict and bound performance and allow decisions on investment in sensing, algorithmics or people to solve a given mission requirement.

 

Advertisements

QUEST Discussion Topics and News, 10/22

October 22, 2010 Leave a comment

QUEST Discussion Topics

10/22

This week we will continue our discussion on architectures, specifically how our old qualia-generating kernel diagram could be updated to more accurately reflect our current ideas.  Included in this discussion we will attempt to nail down our definition of some key terms, such as blobbing and gisting, and where and how a Qualia system sits on top of a standard pattern recognition solution to create a two system representation.

QUEST Discussion Topics and News, 10/22

architecture

Blobs

 

 

Virtual Discussion following the ‘QUEST for Flexible Autonomy’ presentation

October 15, 2010 Leave a comment

Everyone

Thanks for including me.. the discussions look interesting to me… wish I were there.  The paradox you point out is the crux of the issue.  And it is huge.

In an application that is close to me, S&T can give the user more data.  But in fact, more data makes the user less effective.  What he needs is more actionable info, which is usually a combination of technology outputs, schmoozed together with SME human judgment… not easy to do today, so the user isn’t getting it!  What he gets is even more data than he got yesterday… makes him even less effective than he was yesterday… and sours him on S&T!  The user has come to view S&T as part of the problem, not part of the solution.

I suppose the clearest real example of that is non-AF example, 911.  S&T was providing the data to the user that could have prevented 911.  Problem:  the data was not in the form of actionable info, so the data was lost, mixed into the tidal wave of other, meaningless data being piped to the user.  The solution after 911?  Collect even MORE data on more people and give THAT to the user.

We need to focus less on ‘automated dot generation’ and more on ‘automated dot CONNECTION’.

V/R
Guns Atoll

Charles Sadowski Jr., Contractor

Chuck

Well said!  I agree totally.  What the user needs is the ability to ingest/create actionable information.

Remember that when all you have is a hammer everything is a nail.  Data is the hammer today.

Kevin Priddy

Hi Kevin and Chuck,

I am following your thread of thought a little bit and agree totally with your arguments. I have the following to add:

What we need, I think, is the ability to package information in a way so that the human can chunk it into meaningful units and patterns. As noted in a famous study in the 1950s by George Miller, humans have the ability to chunk a lot of low-bit pieces of information into a smaller number of higher-bit pieces of information, which occurs in large part based on meaning. de Groot in the 1960s showed that master chess players can reproduce almost exactly a complicated chess position when it was exposed to them for only 5 seconds, yet the chess masters were no better than novices when trying to reproduce a chess board composed of randomly placed pieces. It was the meaningful arrangement of the pieces in the former situation that allowed the chess masters to encode the positions of the pieces in meaningful chunks and patterns and then remember all of the positions. Follow up work by Chase and Simon in the 1970s showed that methods could be developed to actually measure the size and number of the chunks that the chess masters employed in reproducing from memory various chess arrangements.

Thus, the key, I think, to handling the explosion of information is to get a handle on how to present the information in meaningful ways that allows individuals to easily chunk the information in working memory and thereby encode it into long term memory for later use. Methods developed by Chase and Simon might be useful for designing such a research project.

Cheers,

Robert Patterson.

Robert

Thank you for the insight.  You’ve now destroyed my thoughts of Fisher having superhuman memory recall.  Maybe it’s true but I suspect the study you highlighted will debunk the theory.  I agree that we scientists and engineers need to discover a way to chunk the data into manageable “information chunks.”  We all know that humans can only handle a limited number of chunks 7 +/- 2 at one time, but they can be extremely complicated in content.  That’s probably why chess masters could do well for non-random arrangements of chess pieces.

I look forward to working with everyone on finding better ways for us to present/represent information to the warfighter.

Kevin Priddy

Great meeting guys – what you are discussing in this chain is one of the S/T gaps that we will point out in the Flexible Autonomy talk – we need a theory of alignment – we will define alignment like we did in today’s meeting – the ability of one agent to provide ‘context’ to another agent – we will use the definition of context to be the interagent communication to improve performance (see Oxley pubs) …
Capt Amerika
Hello all,

I’ve been working the other end of this problem for a few years and have some ideas as to how to proceed. As part of the Robust Decision Making project, our team is trying to develop a cognitive model of the sensemaking process in the reverse engineering problem domain. One of the main differences between this and some of the early reasoning work in blocks world, cryptanalysis, or Tower of Hanoi problems is the amount (and role) of background knowledge and the amount of state that exists in the problem environment, which causes the problem-solver to have to select and focus attention on the environment state that is relevant.

I have a conceptual theory of how this happens, but I’m trying to develop a more robust theory through empirical analysis of reverse engineers performing the task (my prospectus/research plan for that work is in review right now).

One thing that I’d like from the group, if possible, is to gather some additional operational use cases that share these features, with which I could organize a cognitive task analysis. I understand the same type of abstraction problems exist in cyber, intelligence analysis, and target recognition, but I don’t have access to anyone that actually performs those tasks. Would anyone care to help me make these connections back to operational contacts?

Thanks and have a great Friday!
Adam B.

Adam,
When I was at university of Minnesota, in the Carlson school of Management there was a prof who was very much into decision sciences and computational models of the way people did stuff so that he could analyze and find flaws or explanations for why, or predictions of when and where people-based systems were going to have problems.. and how to avoid these.  His work included both the doctor-patient domain for diagnosis and treatment of type 2 diabetes, and the Auditor-Business domain for determining what kinds of situations would lead to auditors missing mistakes and misleading accounting documents, plus probably a lot more

His name is Paul Johnson – Carlson School of Management @ UMN

http://www.csom.umn.edu/faculty-research/faculty.aspx?x500=johns021
http://www.bmhi.umn.edu/aboutihi/people/faculty/pjohnson/index.htm

tell him I said Hi!

-Brett Borghetti

QUEST Discussion Topics and News, 10/15

October 15, 2010 Leave a comment

QUEST Discussion Topics

Oct 15, 2010

 

The main discussion point will be a talk Capt Amerika will be giving (on behalf of the quest group) next week ‘ quest for flexible autonomy’.

 

Autonomy paradox – autonomous systems designed to reduce the human operator footprint are requiring increased manpower support requirements.

 

Premise of our presentation:

 

–      Premise: current approaches to ‘automate’ fail to consider the important issue of fundamental science and math of integrating multiple ‘agents’ (some human some electronic) to solve the wicked driver problems in areas like

•       ISR / RPVs (layered sensing)

•       ISHM (Integrated Systems Health Management)

•       Cyber

  • Medicine

 

There is a large ppt file put together for this topic, please email me if you would like a copy of it before tomorrow’s discussion.

 

Secondary topic for tomorrow is the summary presentation on designing conscious systems – come excerpts from the 3 papers we had used in discussing ongoing efforts – and then into a discussion on implications of the papers to our 4 current driver problems – cyber, ishm, virtual world learning, layered sensing (WAAS and/or blue devil exploitation)

 

Again, please just let me know if you would like the slides accompanying this topic.

 

QUEST Discussion Topics and News, 10/15

 

QUEST Discussion Topics and News, Oct 8

October 8, 2010 Leave a comment
QUEST Discussion Topics, 10/8
The major topic for this week is to address qualia as an architecture
based concept for application to our driver problems (Integrated systems
health monitoring, cyber, layered sensing, STEM education, flexible
autonomy for reducing the manpower tail and medicine)