Friday, July 1, 2011

The Decade Of The Mind & Your New Cyborg Brain 3b

We continue with the second part of how DARPA is changing the face of the military and eventually your brain itself.

DARPA according to a 2005 report, has the following areas of interest and focus:
• Detection, precision identification, tracking, and destruction of elusive targets
• Location and characterization of underground structures
• Networked manned and unmanned systems
• Robust, secure self-forming tactical networks
• Cognitive computing
• Assured use of space
• Bio-revolution
• Urban area operations
These goals do not neatly line up to the 2006 presentation by Dr. Tether, but they are rough equivalents.
Kindle eBook
Common Tech Problems,
Windows, OS X

Biometric Computing
In June of 2001, Dr. Tether made a report to subcommittee on Military Research and Development.  At this time Dr. Tether was the director of DARPA.  In August 2000 DARPA according to this report began work on the Human Identification At A Distance program.  The purpose of the program was to identify "...humans at a distance using different biometrics techniques such as face and body parts identification, infrared and hyper-spectral imagery, gait and temporal human dynamics, non-imaging physiological based-biometrics, and remote iris scan."  This project is still progressing.
"Computers, which are more and more central to our society, are already mediating an increasing proportion of our spoken and written communication-- in the telephone switching and transmission system, in electronic mail, in word processing and electronic publishing, in full-text information retrieval and computer bulletin boards..."

Personalized Assistant that Learns (PAL)
The program manager for this project under DARPA is Dr. Robert Kohout, formerly a research scientist at Strategic Analysis part of General Dynamics, which is now Intelligent Automation, Inc.  If we go to Dr. Kohout's former employer, we can look at a PAL system called ARTeMUS which acts very much like a moving, learning network to assist combat troops in urban environments.  If you would like to see it in an ideal scenario we post this short film from the company's website.

To further elaborate on this technology we let DARPA themselves tell us what they are permitted to say.
The Personalized Assistant that Learns (PAL) program is developing machine learning technologies to make information understanding and decision-making more effective and efficient for military users. The program is creating robust software assistants that can help users perform a wide variety of tasks while adapting to the environment and the user’s goals without programming assistance or technical intervention. PAL technologies will reduce the need for large command staffs, thereby enabling smaller, more mobile, less vulnerable command centers.
This program was supplemented by another program called MARS (Mobile Autonomous Robot Software).  Tether explains this technology, "In FY 2001, the program is demonstrating on-line learning techniques that can automatically generate desirable,adaptive behaviors without human intervention. The ultimate goal is to allow the warfighter to task a robot in the same terms as he or she might task a human."

We also present for your perusal a series done on the History Chanel That's Impossible Real Terminators.  If you cannot see the embedded video, here is the link:

Transfer Learning
This is a form of AI (artificial intelligence).  It is not the same as machine learning. Sinno Jialin Pan in a lecture for DARPA defines transfer learning as "...the ability of a system to recognize and apply knowledge and skills learned in previous tasks to novel tasks."  Examples he gives, are crossing over knowledge from Chess to Checkers, Mathematics or Computer Science.  He presented this chart:

Or this chart
This transfer learning is related to machine learning, but also the core technology for the aforementioned PAL program of DARPA.  If you are getting tired of all these acronyms, forgive us, but there are more.  Transfer Learning was funded under an acronym CALO (Cognitive Assistant that Learns and Organizes, which was inspired by the Greek word Calonis meaning "solider's servant") from 2003-2008.  It is claimed that this was the largest AI project in history with more than 300 researches from the top 25 universities and commercial institutions.  The goal is to produce "cognitive assistants" that can "...reason, learn from experience, be told what to do, explain what they are doing, reflect on their experience, and respond robustly to surprise."  This program was incorporated in the PAL program, being its
VIGIL camera
predecessor.  This program is now being spearheaded by SRI International.  On June 21, 2011 the company was hired by DARPA to develop a related technology called VIGIL (Visual Intelligence Grounded in Learning).  Vigil's goal is develop autonomous AI combat robots.
VIGIL will be integrated with other components of the Mind’s Eye program that are being developed by research teams around the globe. The result will be a smart camera system that needs minimal human supervision, can be deployed rapidly and cost effectively in unmanned vehicles (UVs) stationed in areas under automatic surveillance, and can interpret behavior and identify potential threats from video data.
This is a vast undertaking still with the following institutions taking part:

Boeing Phantom Works
Carnegie Mellon University
Sybase iAnywhere
Fetch Technologies, Inc.
Georgia Institute of Technology
Harvard University
Massachusetts Institute of Technology
Oregon Health & Science University
Oregon State University
Radar Networks, Inc.
Stanford University
State University of New York - Stony Brook
University of California at Berkeley
University of Massachusetts at Amherst
University of Michigan
University of Pennsylvania
University of Rochester
University of Southern California and its Information Sciences Institute
The University of Texas at Austin
University of Washington
University of West Florida’s Institute of Human and Machine Cognition
Yale University

Integrated Learning
This work is part of the POIROT project inside DARPA.  POIROT is the catchy acronym for Plan Order Induction by Reasoning from One Trial.  An algorithm was developed by a Dr. Fusan Yaman, who in 2008 was working for Raytheon Corporation.  The purpose of this algorithm which she named WIT (Workflow Inference from Traces) was to learn
...a hierarchy of planning procedures from an observed plan. WIT combines Grammar Induction techniques with effective heuristics and compared to existing learning algorithms for planning knowledge, WIT uses no background knowledge about the target domain. Also for the same project Dr. Yaman, developed and implemented the algorithm, Cool Hybrid Adaptive Ranking Model, (CHARM) that learns the ranking of a collection objects based on the selections an expert demonstrates while solving a planning problem. CHARM assumes the underlying preferences of the expert can be represented as lexicographic order on the object attributes and can learn the target model using very little number of demonstrations.
An example of such a "learning system" is another acronym developed by DARPA called GILA (Generalized Integrated Learning Architecture) developed in 2008 for the purpose of actually the air traffic control of military aircraft due to the increasing amount of military drones flying.  The prediction by Lockheed Martin the military contractor who developed the technology is that eventually the AI will surpass a human air traffic controller by 125%.  This article states that,
Such software has to combine limited observations with subject expertise, general knowledge, reasoning, and by asking what-if questions. The integrated learner also will have explicit learning goals, keep track of what it does not know, what it needs to know as well as track and reason about its uncertainties. The software will attempt to figure things out, as well as tolerate errors and missing information by using whatever information or reasoning is available."
If this is not a real AI, we will confess we do not know what is.  How developed is it?  We assume very developed.

Architectures for Cognitive Information Processing (ACIP)
You might wonder what cognitive information processing is.  It is a theory of learning.  It is defined by Reiser & Dempsey in a 2007 book titled, Trends and Issues In Instructional Design.
The cognitive information processing theory looks at the role of the three stages of memory (sensory, short-term, and long-term) in retrieving information and then transferring it to store and then recall in memory. Sensory memory allows the learners to organize groups of information or patterns in their environment; learners recognize and then process these patters. Short-term memory allows the learner to hold and to understand small amounts of information. If the information is effectively connected to previous knowledge, it is stored in long-term memory. Long-term memory allows the learner to remember and then apply knowledge across learning environments; and, remember the information for large amounts of time after it is learned. Encoding and retrieval also play key roles in the cognitive information processing theory
In our next part in this series, we will continue with Dr. Tether's outline of the use of neuroprosthetics for all this military technology.

No comments: