[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]
Subject: Re: [huml] another DARPA BAA (better timeframe and perhaps more appropriatetarget)
I don't think they realize that it really can't be done "in the end" with a Turing machine... But there are enough applications that will prove useful and practical which can be. These will emulate many of the "behaviors" of cognitive systems to enough of a degree that they will give one the impression that they are performing these functions. This is a classic example of "we really want something that is hard to define, and of which we have only a vague idea how to approach." That is why I mentioned defining the "path" rather than a supposed "final" architecture. There are a number of assumptions upon which most of the current work is based that will prove to be "problematic", and in the end, "fatal" (in my opinion) to the development of artificial cognitive systems that approach a level of ability close to that of a human being (there are other "approaches" which I believe to be far more promising). However, since not all applications would require this level of ability, the current approaches should provide fertile ground for development. If we can, as Rex suggests, make the case, bullet point by bullet point, and provide a coherent framework to approach the problem, and demonstrate practical application of the framework, we may have a chance at the grant. I'll certainly help you Rex lay it out. Others can contribute as they choose to the initial outline. Once we are happy with the initial outline describing the suggested approaches, we can decide if it is a practical endeavor, and one that will be worth the effort submitting. I think this is the best approach we can take Jim.
Rob
Rex Brooks wrote:
Yikes! I'm glad I didn't try to fit this into my day! Do you suppose they could have written a more scattered and unfocused description of a Turing Machine? Except, of course, it goes a bit farther afield than a true Turing Machine. I will take a whack at seeing if I think this is doable tomorrow morning. Warning: the only way I know how to do that is to answer enough of those bullet points to make the case, or not, so it will probably be a long drink of water. It'll probably come at you in chunks since I can't devote an entire day to this right now because I'm already way behind in just keeping up with stuff because of the hardware foul ups. Ciao,Rex At 8:38 PM -0500 6/3/03, Rob Nixon wrote:
Re: ( DARPA ) http://www.darpa.mil/ipto/Solicitations/Mod4_02-21.html
Cognitive Information Processing Technology - Modification 4
The HumanML initiative could certainly play a major role in the development of systems such as these. The Cognitive/Context (CogEnvML) subcommittee would be the logical nexus of most of the activity related to this particular proposal. With that said we have to be careful that we wouldn't be perceived as "losing" our focus. I am confident that we have the intellectual capital to address this proposal given the expertise, diversity, and skill of our members. The only real concern I have is of identifying (and convincing) a larger "established" corporate partner to be interested in joining the team to lend us an air of credibility. And doing this in time to meet the deadline that is imposed on us. Coordination is key as Jim has mentioned. And much of what Len has said must be kept in mind. The only reason that I'm considering this proposal is that it overlaps a great deal with what the group wants to accomplish, as well as myself and our company. You would be hard pressed to find an application area that is likely to be more complex, so we have to make sure we go into this (if we decide to) with our blinders off. I've been working on this problem for over a decade and half, and will attest that the deeper you go, the deeper it gets ( for obvious reasons ). But I think that if we define the "path" of development to follow, rather than trying to develop the "final" form out of the box we will be far more successful (but you all know this ).
There's nothing more exciting in my book than to work on (and develop ) something others think is impossible (as long as it produces results of practical value).
Rob
Rex Brooks wrote:
Sorry, I couldn't get a reply out on this as I was scrambling around getting a used monitor to fill in for the one that died first thing this morning. I haven't yet had a chance to even do more than bookmark the BAA that Len posted. But, I can assure all that if Rob is for it, so am I. I will get to it before tomorrow morning, but I am still scrambling, and frankly, given my mindset at the moment, no one wants me making decisions about anything until I get back to a more reasonable perspective. Ciao,Rex At 1:21 PM -0500 6/3/03, Rob Nixon wrote:
Re: The real question here, in my opinion, is this: Are we going to make a decision to go after this grant, or are we just going to talk about it and let the opportunity slip away?
Rex and I put together a proposal last year in a very tight timeframe. It wasn't selected because we added a section on security that wasn't requested. With that lesson learned I am sure that we could approach this one appropriately. It would however, have to be a very detailed response, done quickly and thoroughly. I think it would be of great value to us on multiple levels. However, it would, as Jim indicates require significant coordination. I'm up for it, if others will join in the process.
Rob
James Landrum wrote:
Well folks, the big hurdle with the LifeLog program call is the deadline of June 23- and trust me on this when I say compiling, preparing, and submitting a potentially successful proposal to this program will require more time than that. Secondly, the Cognitive Information Proicessing Technology program call has a much better time frame for preparation and submission of a potentially successful grant application. These types of grant proposals are very technical and have significant detail (content) requirements, are very structured in scope, schedules and goals, and require a substantial coordinated effort by everyone involved to prepare as a collaborative undertaking.
This is especially important when one considers that the HumanML TC and Humanmarkup.org participants and collaborators are so spread out in various localities that coordination will need to be well organized and activity will need to be sustained with deliverables (contributions) to the project proposal set within a timeframe that each and everyone can meet. If you have not done so, at least read through one time the guidelines and requirements. That will give each of us an opprtunity to make a more informed group decision about the question I pose here:The real question here, in my opinion, is this: Are we going to make a decision to go after this grant, or are we just going to talk about it and let the opportunity slip away?
If we decide to go after this grant, we'll need to dedicate serious time and effort to it, and that is a serious commitment for each of us. We already have a certain amount of prepared content (boiler plate), but we'd need to develop a broad range of other detailed content and information for inclusion in the total proposal package. At issue here are which unit of HumanMarkup.Org or HumanMarkup Language TC is prepared to administrate the overall grant preparation and submission process, and which unit would be designated as the primary, lead institution, and how would the participants (individuals and organizations, institutions, etc., roles be described and justified in the project description, and what share would each unit (participant) receive in the project fund distribution. Lots of issues to explore and discuss, and these are just a few for your consideration.
Rob Nixon wrote:
Absolutely agree.
Also need to model a number of cognitive agents that "aren't" particularly good
at using their previous "experience" in order to increase the potential solution
space. In other words to allow for enough degrees of freedom for the system to
discover novel solutions to problems, much the same way that some people
"stumble" on to solutions through a series of mistakes. ( related to GA systems
). Another aspect of this (beyond the individual agent, related to layers of
social hierarchy) is the effect that these "poorly" functioning agents will have
on the other agents within their work cluster. The interaction of these
"poorly" functioning agents with the more accurate agents can potentially again
enhance the solution space. In fact you may discover emergent behavior among
the work group clusters of agents that wouldn't be there if you have just groups
of fairly optimized agents working together.
This will be very important for exploring real world "simulations".
Rob
"Bullard, Claude L (Len)" wrote:
It is the harder one but one that requires less data
to test although it is hard to prove it scales. The
first one is closer to what HumanML as a set of categories
for contexts of human communication can easily do.
Systems that can successfully emulate human cognition
must deal with theory formation and how that is
effected by interpretation of observed events which
have emotional contexts.
Ask yourself how one goes about selecting HumanML
categories to create a secondary? Then how does
one markup an observation (think of it as a document?
Then how does the cognitive agent select this document
among many it has to perform an action based on a
new observation?
The hard part for DARPA to accept will be that humans
are not that good at this either without training
and that training being focused on narrow domains
limits flexibility. So the next question is how to
create teams of cognitive agents that share theories
and will act in concert to achieve some mission.
They will discover the same principles of leadership
and group dynamics that work for the military and
sports teams when the actions to be achieved require
highly directed force. In other types of actions,
a different team dynamic is required.
How does a set of episodic events become a genre?
len
-----Original Message-----
From: Rob Nixon [mailto:robnixon@execpc.com]
Sent: Tuesday, June 03, 2003 11:53 AM
To: James Landrum
Cc: huml@lists.oasis-open.org;humanorg@yahoogroups.com; Bullard, Claude
L (Len)
Subject: Re: [huml] another DARPA BAA (better timeframe and perhaps more
appropriatetarget)
Both DARPA projects that Jim and Len have pointed out look interesting.
The one that Jim has pointed out appears to quite parallel to what
my company is currently working on.
Regards,
Rob
http://www.darpa.mil/ipto/Solicitations/Mod4_02-21.htmlJames Landrum wrote:
see Cognitive Information Processing Technology
http://www.darpa.mil/ipto/Solicitations/Mod4_02-21.html
--
>From the Desk of James E. Landrum III
Database Manager
Archaeology Technologies Laboratory (ATL; http://atl.ndsu.edu)
Digital Archive Network for Anthropology (DANA;
http://atl.ndsu.edu/archive)
North Dakota State University, Fargo, North Dakota 58105
Ph: 701-231-7115 (my desk) and ATL 701-231-6434
FAX: 701-231-1047
email: james.landrum@ndsu.nodak.edu
You may leave a Technical Committee at any time by visiting
http://www.oasis-open.org/apps/org/workgroup/huml/members/leave_workgroup.ph
p
--
>From the Desk of James E. Landrum III
Database Manager
Archaeology Technologies Laboratory (ATL; http://atl.ndsu.edu)
Digital Archive Network for Anthropology (DANA; http://atl.ndsu.edu/archive)
North Dakota State University, Fargo, North Dakota 58105
Ph: 701-231-7115 (my desk) and ATL 701-231-6434
FAX: 701-231-1047
email: james.landrum@ndsu.nodak.edu
--
Rex Brooks
GeoAddress: 1361-A Addison, Berkeley, CA, 94702 USA, Earth
W3Address: http://www.starbourne.com
Email: rexb@starbourne.comTel: 510-849-2309Fax: By Request
--Rex Brooks
GeoAddress: 1361-A Addison, Berkeley, CA, 94702 USA, Earth
W3Address: http://www.starbourne.com
Email: rexb@starbourne.comTel: 510-849-2309Fax: By Request
--
[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]