Artificial Intelligence

Overview

Artificial Intelligence (AI) utilizes programming algorithms to simulate thought processes and reasoning that produce behavior similar to humans. A successful implementation of AI could be tested using a Turing Test approach, in which a human interacts with an interface that could have either a human or computer on the other end. The test is considered successful if the human is unable to determine whether there is a computer or a human on the other end.

Intelligent Agents

AI is made up of intelligent agents that perform functions within an environment. Intelligent agents have sensors that are used to perceive changes in the environment and utilizes effectors to produce actions on the environment. The agent is typically programmed to perceive certain changes in the environment and can learn from those perceptions by testing its actions on the environment. In this manner complete AI entities are typically made up of multiple intelligent agents that have specific functions within its own domain.

AI in E-Learning

The application of AI within e-learning can produce the potential of creating realistic environments with which students can interact. The student essentially would interact with the intelligent agents which in turn effect or perceive changes in the simulated environment. The intelligent agents would then communicate perceived changes in the environment back to the student who then makes decisions based upon their own perceptions of the environment.

For example, in technical troubleshooting applications intelligent agents could be used to produce an environment in which the learner must diagnose a problem. The learner is provided with all the resources that would normally be available on the job, such as a voltmeter, oscilliscope, tools, clipboard, etc. The User Interface would provide access to these resources, but not provide any clue as to the order in which each resource should be used. The resources could be simple objects that simply interact with the environment, i.e. the voltmeter would simply receive the voltage measurement and display the results to the learner. The intelligent agents would be programmed to manipulate the environment based on the usage of the resources, and to monitor how the resources are used in order to measure performance. This is different than just having the student follow a procedure to diagnose a problem because the simulated environment could react differently to different approaches to the problem. The learner is not given any indication as to what to do next, so it truly monitors how the learner reacts to different situations and how the student adapts to unexpected output.

This produces the most realistic environment in which a student can safely learn technical skills as well as soft skills such as problem solving. Such an environment provides a valuable opportunity to acquire experience before applying in real world situations.

Call Center Training Example

AI engines are basically software components that follow rules, and once a sufficient AI engine or framework is built, you just need to write the rules to fit any case situation. The rules could follow along the same lines as the old Eliza program from the 1960s. This was a natural language processor that scanned a user's response and produces a response of its own by scanning the input for speech patterns. In a Call Center environment, the Eliza-like intelligent agent could represent a customer. Additional intelligent agents would be employed to not only manage the responses but also measure the student on how they respond and interact with the customers, thus measuring their level of proficiency.

The rules of the AI component would be designed so that the customer is attempting to reach a specific objective (i.e a satisfied customer), and would measure the student on their performance in reaching his/her objective. Many case studies could be implemented in this type of simulation, including customers attempting to defraud as well as customers with legitimate issues. Additionally, intelligent agents could be used to interject emotional responses based upon the performance of the call center analyst student.

For example, here is a simplified logical approach to consider with an intelligent agent representing the customer. In practicality, there would be much more code related to producing the virtual environment and perceptions of the environment and what this does not plainly show is that another intelligent agent is consulted to determine the emotional state of this agent based on the semantic analysis of the input. Additionally the Knowledge Base (KB) exists for other agents to recognize repetitive behavior, references to earlier statements, or to just simply learn from the exchange of information during the interaction.

Input: "As I mentioned to you before I cannot do that"

Response: "That is completely unacceptable! This is so frustrating, I want to speak to your supervisor!"

VoiceXML comes in as a processor to take the text generated by the output of the intelligent agents into a synthesized voice, and also to receive voice input from a microphone into a voice recognition server. The student would use a headset connected to the computer and would interact with this e-learning component as if they were actually talking to a customer.

<?xml version="1.0" encoding="UTF-8"?> 
         <vxml version="2.0" xmlns="http://www.w3.org/2001/vxml"          
         xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" 
         xsi:schemaLocation="http://www.w3.org/2001/vxml 
         http://www.w3.org/TR/voicexml20/vxml.xsd">
         <form id="Call Center">
         <block>Incoming Call.</block>
         <field name="greeting">
         <catch event="help">
         Please speak the greeting to the customer.
         </catch>
         </field>
         .
         .
         .
         <field name="response">
         <prompt>
         <submit next="/servlet/customer-agent namelist=input">
         </prompt>
         <grammar src="english.grxml" type="application/srgs+xml"/>
         <catch event="help">
         Please speak your response to the customer.
         </catch>
         </field>
         </form>
         </vxml>

This is just an idea of an approach, and it could be quite expensive to develop so it is important to research existing AI engines that could be used for this approach and to determine the long-term benefit compared to the cost in time and budget, however once an AI engine is built to follow interpersonal communication rules, it could have a large application base in e-learning for multiple disciplines such as customer service, management, sales negotiation, etc.

The Challenge

Ultimately the challenge for using e-learning for this type of training is practice. That is, practice in order to raise the level of proficiency. Some people might argue however that it draws dangerously close to removing true human interaction in training and therefore produces less human employees. Joseph Weizenbaum who is the author of the Eliza program even agreed with this when he wrote "Computer Power and Human Reason". In that book he pointed out that AI might induce the thought that humans are automatable and thus could result in a loss of humanity if we pursue the research and development of AI. The purpose of creating AI simulations of these types of interactions is to base the rules and scripts on actual human cases and therefore provide an abundant library of case studies for a trainee to practice without the need to consume additional instructional resources. It also provides an asynchronous availability of the practice so that schedules do not prevent an employee from getting the practice they need in their learning path.

Applying AI to new types of situations still requires detailed programming which makes it expensive to implement in certain e-learning applications. It can be worthwhile however if a framework can be constructed that allows adaptability to multiple situations simply by changing content, but leaving the intelligent agents themselves intact.

AI does have the potential to be applied in learning programs that can accelerate the learning process.

Copyright ©2006 E-Learning Engineering
Last modified: April 22, 2006
Terms of Use