Intelligent Agents

“The mind is to the brain as software is to hardware” (Rapaport, 1996)

In the early days of artificial intelligence, despite hardware needing basic software to operate as the quote above illustrates, intelligent agent implementation attempts were often hardware based where some form of physical sensor and actuator were involved. Over time these became more and more reliant on intelligent software agents to operate to the point where it became possible to create an intelligent software agent with no physical, or hardware, attributes at all.

In order to be considered an “agent” (either hardware or software) an intelligent agent must autonomously perceive through sensors and behave through actuators.

To demonstrate the use of an intelligent software agent, I will draw from my own experience. I was involved in a project in the late 1990s where we had 40 people page setting directory pages on computers and we produced around 100 directories per annum. Each directory took 3 months to prepare for print with this staffing level. We had tried many ways of improving the speed of production by supplementing users with automated graphical and colour management routines, pre-flight checkers for spelling errors, swearwords, illegal advertising claims and so on, however the best we managed were small percentage increases in efficiency. As a result we setup a study group with a group of LISP programmers at Edinburgh University to examine if we could develop an intelligent software agent that could layout the book based on a set of “morals”, i.e. rules regarding boundaries but not regarding actual graphical placement (adverts). The results were astonishing.

Within a year we had directories being produced in 3 minutes by 1 person (pressing “go”) and all page setters were re-assigned or made redundant. The system essentially read the contents of a folder (specially SGML tagged graphics files) and produced a print ready file. However this time saving was not the only benefit to emerge, because the system could make its own judgement on the use of page space, it also produced the same directory contents in fewer pages, saving £millions in printing costs and paper on an ongoing basis and because the directory took minutes to produce, the sales window for advertising gained by 3 months. Had this been a hardware based system there would have been some form of physical page setting hardware that interacted with the page setting process as the people who made the pages previously did, using its own software experience to make the same decisions as the software only model did.

It could be argued in this case that the intelligent software agent was actually based on a set of rules and did not operate autonomously, however the actual decision about which square was placed where on which page, unless controlled by a “moral” was totally the software’s decision.

References

Glenn, J. (2009) Computer Science: An Overview. 10th ed. Boston: Pearson Education.

Rapaport, William J. (1996): Cognitive Science [Online]. Available at: http://www.cse.buffalo.edu/~rapaport/Papers/cogsci.pdf (Accessed 24 December 2009).

Wikipedia: Intelligent Agent [Online]. Available at: http://en.wikipedia.org/wiki/Intelligent_agent (Accessed 24 December 2009).