Decision Structures
Demonstrations
Emulating Organized Behavior
This demonstration shows our LINIAC™ artificial neural net technology in action. You assume the role of a rogue submarine pirate with control of a quantity of nuclear warheads. The neural net implements a virtual defensive command, using their knowledge to stop your missiles with their limited arsenal. Can you outfox the computer?
Decision Structures
A Method Developed by Pathfinder Systems to Emulate Human Decisions
Easy and Difficult
With fast digital computers, we can easily do complex computations, compose animations, create games, build office software networks, etc., but we have great difficulty in emulating the simplest human decisions when they are intuitive or judgmental, especially when they represent interactions. To overcome this, we have Artificial Intelligence.
Artificial Intelligence?
Artificial intelligence—is it just software? Of course it is. We call it Artificial Intelligence (AI) when software needs to do something humans do with ease, but we cannot quite figure out how to do it with a computer. A decision structure emulates human functions when we cannot develop a closed-form algorithm. We refer to this as “AI”.
The Neural Net Model
Our approach is based on the “neural net” model. This model emulates how a brain works, but can only roughly emulate the “how”:
-
Massively distributed, multi-layer processing is used.
-
Each layer performs very simple computing (in essence, sums inputs then forwards an output).
-
Each layer is non-linear.
It does not even approximately do the “what”:
-
The human brain’s hardware is billions of times more capable than a high-end PC.
-
Every second, the brain accomplishes the equivalent of writing a million lines of good code.
“Knowledge” is stored in connection strengths within each layer—i.e. essentially in the coefficients of a very large matrix. The net does not have rules; it “learns by experience” (changes its connections).
How One Layer Works
Example from our early demonstration program, Krasniy Oktyabr
This is just one layer. The inputs are estimates based on imperfect sensor data. Launch Decision controls the decision to launch ABM’s. Processing takes a mere microsecond on a PC.
Interactions in an Organization
We postulate a human organization model:
-
The organization can deal with many different problems.
-
It has multiple specialists.
-
Different problems may require different groups of specialists to form a solution.
-
In different order, using different data sources
-
With different procedures selected from a group of standard operating procedures
-
-
The specialists may interact with each other.
The Virtual Organization
Implementation at Pathfinder Systems:
-
Each specialist is a LINIAC very high-speed artificial neural net (ANN).
-
The standard operating procedures are correlated to the incoming problem.
-
The result is a dynamically reconfigurable neural net structure, tailored to each problem.
-
We refer to this as the decision structure.
Implementation
LINIAC ANN’s have been implemented as compiled code modules for many environments as well as Microsoft Excel spreadsheet add-ins. Execution vectors may be implemented as user-developed code, standard execution vector modules, or spreadsheet equations. Each LINIAC is trained by the TEACHER training software.
Current Status
Today, you can contract with Pathfinder Systems to add this capability to your software. You can add, change, and extend the operating procedures to suit your needs, and add the knowledge of your experts. You don’t need any computer expertise.
Summary
It is feasible to add emulation of human decision making to software code or spreadsheets. It is also feasible to emulate collective decision making by specialists having different areas of knowledge. In addition, this technology is extremely fast (hundreds of thousands of decisions per second on a PC).