Available resources:
· Robot intelligence and
self-awareness
· The drama of the
robots: the recall
· What You Can Do to
Help: Adopt-A-Robot
· Letter from a Robot
· Frequently Asked Questions
· The Official Save the Robots flier
· Send Us An Email
|
Robot intelligence and self-awareness: an ongoing debate
page 2
So what makes a robot tick and jump in front of a car to save its owner? The answer is fuzzy algorithms and hardware efficiency. Exponential increase in storage capacity and processing power over the past decades have dramatically increased the results of these techniques. Robots using fuzzy algorithms (also known as CRFA's for Converging Robotic Fuzzy Algorithms) work following a three step process: 1. Concept gathering, 2. Concept storing and associations definitions and 3. Concepts implementation. Sounds fancy but it's pretty basic.
Concept gathering is the process through which a robot learns. A concept can be any piece of information, from the manufacturing process of a computer keyboard to the shape of an apple or the knowledge that a ten ton steel bar will crush anything in its path if it rolls over. A typical robot today can store data (concepts) equivalent to a million times the contents of all books that were ever printed. This huge storage capacity allows robots to learn and store a mind-boggling amount of data, programs and "knowledge".
A small part of these programs and data is pre-imprinted in the robot's microchips at the time of its manufacturing. After that the robot is connected to huge databases, the internet or even real-life situations, from which it gathers additional concepts that it needs for its future roles or jobs. Exponential increases in processing power mean that in one hour a robot can scan 50,000 webpages on the internet or a datanet, select and gather about one million concepts from these pages, and compare each of these to the other ten trillion concepts that it might already have stored. The results of these comparisons lead to associations and later to implementations of gathered concepts at their jobs in factories, offices or homes.
While robots are generally pre-programmed with certain requirements for data gathering and concept implementation (i.e. "specialize in chemical processing"), fuzzy algorithms allow robots to create their own learning path and later react to unpredictable situations and commands based on each unit's unique set of stored concepts and associations. These reactions cannot be predicted in the same way that human reactions can't be foreseen. Fantastic storage capacity and computing power means that robots can learn pretty much anything and create any associations they see fit, unlike older sequential databases focused on optimization and strict rules and structures.
Let's see an example of how this process works. Take your basic sak-seven (SK-7 model) robot that is being prepared for use as a chemical manufacturing plant supervisor. This robot is pre-imprinted with certain minimal algorithms and data such as manufacturing processes, safety regulations, process control, etc. Next, the robot is hooked up to databases, datanets and the internet and starts doing its own research. It might start by running a search on "safety procedures" as instructed by its training algorithms. As the content and structure of the internet and other datanets change constantly, no one can know in advance what the robot will find.
This doesn't matter much. The robot will start by analyzing and storing information related to its search. For example it might find and store a concept such as "when you mix compound A with compound B the plant blows up". Besides the information itself, the robot also stores context: who made this assumption? who used it? was it found in other sources as well? was it cited in other materials? is it consistent with concepts that robot has already stored? was it tested in real life? How pure does compound A need to be for the plant to actually blow up? What can be done to avoid such a reaction? What other reactions can substitute the one between A and B? Such questions can lead the robot on many different searches and learning paths.
Next, the robot creates associations. It might associate the chemical reaction of compounds A and B (and the plant blowing up) with a safety requirement that says "chemical reactions that result in explosions should be avoided". Thus, it might initiate another search for "substitutes for reaction between A and B" and go on from there. The robot might also communicate with other robots having similar positions, or even require assistance from human specialists.
Says Starkson: "The distinction between human-to-human and robot-to-human communication is blurred. For many of the electronically-submitted questions and communications that I receive I need to check header info to see if I'm talking with a person or a robot." And that's exactly what the Turing Test is about. Developed by Alan Turing in 1950, the theory tests for machine intelligence by checking to see if a human operator can distinguish between a person and a machine (both hidden from the operator) with whom he or she is carrying simultaneous conversations.
Pages: 1 2 3next page
|
|
