Artificial intelligence: Solving problems for the real world
Editor’s note: This is the second part in an occasional series in which the fantastic realms of science fiction are compared to those of real world science fact.
For a typical American, the mention of artificial intelligence may conjure up nasty images of a robot wielding a plasma rifle atop a pyramid of human skulls.
Mention artificial intelligence to Berthe Choueiry, though, and she thinks of problems.
Choueiry, an associate professor of computer science and engineering, conducts artificial intelligence research at the University of Nebraska-Lincoln.
And out of the wide field of artificial intelligence, her research focuses on constraint processing. This involves developing techniques to solve decision problems and applying them to real world uses, Choueiry said.
But artificial intelligence hasn’t always been what it is now. The field has its roots in the philosophical field of logic, tracing back to when Aristotle first defined a reasoning process called syllogism. The syllogism works on a simple “if, then” operation using two premises to come to a conclusion.
For example, take this premise: “If this can of beer has been lying open on my coffee table all night, then it probably doesn’t taste too good the next day.”
The next premise could be that there is a beer that has been lying open on my coffee table all day, and a conclusion, based on the two initial premises, can be drawn: The beer, in fact, doesn’t taste too good; it’s flat and someone has ashed a cigarette in it.
Artificial intelligence processes can be divided into three stages: perception, reasoning and action.
Perception is the software acquiring data from a source, be it a user inputting data or a predetermined database. It also could be sensors measuring data in real time, such as in a weather station, called an “accessible environment.”
That data then moves into the reasoning stage, where the software must process the input data into usable, quantifiable information. After processing, the software moves into the reasoning stage, where the program must decide what course of action to take or data to output.
Think of a robot navigating its way around a room. Sensors could measure distance from walls and furniture to the robot. The robot then reasons where it can and cannot move, then acts on that reasoning by traveling.
Choueiry’s own research doesn’t involve enabling robots to become self-aware and resist their human enslavers; instead, it focuses on something not only far milder but also entirely software-based. She works to generate “solutions that hopefully apply to real world problems like resource allocation, airline times and natural language processing,” Choueiry said.
Although her tools are elaborate mathematical functions, her goal is to keep them as simple as possible.
“We’d like to develop tools for you to use constraints without even thinking of them or having to learn what they are,” she said.
Constraint propagation, which is Choueiry’s speciality, is just one method of reasoning for artificial intelligence. Constraint-based reasoning is a deductive process in which the program looks at a group of data by considering which responses are not acceptable. These unacceptable responses are constraints.
Suppose Choueiry were to develop software to determine a student’s entertainment options for the weekend. Certain constraints could include the student’s financial limitations or transportation options. Driving to a casino in Iowa would be an unacceptable result, or a constraint, for a student with only $5 and a bicycle.
With the help of students, Choueiry created a program to illustrate the concept of constraints. For the program, the computer science and engineering classes had to find graduate students to be teaching assistants. Constraints like class schedules, prerequisites and qualifications were input by the prospective teaching assistants. Eventually a teaching schedule, which was actually used by the department, was derived.
Choueiry said one of the most difficult aspects of constraint processing is the concept of combinatorial explosion. One problem may have a relatively simple solution, but once the constraints of that problem increase, the time required to solve that problem increase exponentially.
Choueiry used the example of a sliding tile puzzle, where a grid with numbered tiles must be arranged in numerical order.
“You can manage a three by three and probably solve it pretty quickly,” Choueiry said. “A five by five might take you a whole day, and a twenty by twenty might take you the life of the universe.”
http://www.dailynebraskan.com/vnews/display.v/ART/2005/03/21/423e9361b0419
