Intelligent Computer Aided Instruction

One of the first researchers to advocate the use of AI in teaching systems was 1 aime Carbonell. His milestone paper (Carbonell, 1970) precipitated concerted activity in the area which continues until this day. Although he did not use the term, the field that he opened up is often called Intelligent Computer Aided (or Assisted) Instruction (ICAI). Carbonell maintains that any program that is going to teach a student must have some knowledge of the subject area or domain of interest rather than just storing answers to specific questions. As he observes 'In his teaching process, a human teacher is not reciting specific questions, but he is utilizing and processing knowledge he has stored in the form of a semantic information structure' (Carbonell, 1 970, p 1 95). Carbonell provides this knowledge by adapting the concept of the semantic network that Quillian ( 1 969) had used for natural language comprehension.

The subject that Carbonell chose for his research was South American geography. His program, known as SCHOLAR, stores facts, concepts and procedures in this domain. A simplified version of part of the semantic network, taken from O'Shea and Self ( 1 983), is reproduced as Figure 2. 1 . Argentina Superconcept Location Bordering countries Latitude -22 -55 Country Superconcept Superpart Examples Latitude

The information stored in such networks allows the system to answer a range of questions of different kinds. Instead of questions and answers being pre-stored the system can work them out as necessary from its internal knowledge of geography contained in these data structures. In addition to facts and relationships, heuristic procedures are available for determining information that cannot be specifically inferred from the networks. For example, a procedure is available that estimates climate given local conditions such as latitude and altitude.

In addition to being able to answer direct questions such as 'What is the latitude of South America?' SCHOLAR can deal with queries like 'Give me some information about Brazil'. It can then trace through the semantic network listing facts about the country, applying the heuristic that the further away from the Brazil node the information is stored the less relevant it is likely to be.

Although the idea that information for teaching could be stored in a semantic net was a good one, the program has no underlying teaching strategy other than that the student should learn by asking and answering questions. Sessions are undirected and topics are generated at random leading to a lack of cohesion. The mixed initiative interactive style is typified by the extract shown in Figure 2.2.

Jaime Carbonell died in 1 973 but his work was continued by others including Allan Collins and Albert Stevens. One of the main lines of research aimed to make the dialogue in SCHOLAR more directed. To this end Collins, Warnock and Passafiume ( 1 975) studied protocols of dialogues between human tutors and students to determine appropriate heuristics for directing the dialogues during sessions. Central to the approach is the incorporation of an agenda that is used to determine what topics and sub­ topics should be covered and when.

The model was further enhanced by defining an educational basis for the tutorial approach: the Socratic method. In this approach, originally used by the Ancient Greeks, the tutor seeks to teach students by asking them carefully formulated questions that encourage them to make appropriate deductions. Collins (1977) endeavours to pinpoint the central goals and tenets of the Socratic method. He gives the main goals as attempting to force students to reason for themselves, getting them to derive general principles and encouraging them to apply these principles to new situations. He then shows how these goals can be achieved by using a computer program.

SCHOLAR : THE POPULATION IN CHILE IS APPROX 8500000. PEOPLE. TRUE OR FALSE?

Student : TRUE

SCHOLAR : YOU ARE CORRECT.

Student :

USE ONE OF THE FOLLOWING: SUCRE

FRENCH ARGENTINA SPANISH

TO ANSWER THE QUESTION:

WHAT IS THE LANGUAGE IN CHILE?

SPANICH

SCHOLAR : YOU SHOULD HAVE WRITTEN "SPANISH".

Student :

VERY GOOD.

PLEASE INDICATE IF THE FOLLOWING STATEMENT IS CORRECT OR INCORRECT:

THE AREA OF PARAGUAY IS APPROXIMATELY 47432 SQUARE MILES.

WHAT IS THE AREA OF BRAZIL?

SCHOLAR : THE AREA OF BRAZIL IS APPROX 3600000. SQUARE MILES.

NOW ANSWER THE QUESTION YOU DIDN'T ANSWER BEFORE.

Student : INCORRECT SCHOLAR : VERY GOOD.

Student :

THE AREA OF PARAGUAY IS APPROX 1 57000 SQUARE MILES. APPROX WHAT IS THE POPULATION IN BRAZIL?

60000000.

SCHOLAR : APPROXIMATELY TRUE, BUT A MORE EXACT VALUE IS 70000000.

Student : Q/A

(student changes mode to question/answer) Student : WHAT IS THE PRINCIPAL RELIGION IN URUGUAY? SCHOLAR : THE PRINCIPAL RELIGION IS CATHOLICISM. Student : THE PRINCIPAL RELIGION IN PERU IS - - - . SCHOLAR : I'M AFRAID I CANT ANSWER THAT.

Student : IS IT TRUE THAT THE AREA OF URUGUAY IS 70000? SCHOLAR : THAT'S ABOUT RIGHT. THE EXACT ANSWER IS 79000. Student : THE PRINCIPAL RIVERS IN ARGENTINA ARE - - -.

SCHOLA R : THE PRINCIPAL RIVERS IN ARGENTINA ARE THE RIO DE LA PLATA, THE PARANA, THE URUGUAY RIVER, THE SALADO, THE PARAGUAY RIVER, AND THE PILCOMA YO.

In a related project, Stevens and Collins (1977) investigate the use of the Socratic method in a meteorological tutor called WHY. The main aim of SCHOLAR is to teach concepts and relationships and so the semantic net is a convenient form of representation. For WHY, which seeks to teach the student about the processes that lead to rainfall, another type of structure that could better represent these processes was needed. Stevens and Collins recognised the significance of the parallel research that Schank and Abelson (1 977) were carrying out on narrative understanding, particularly the concept of a script. Scripts give a method of representing stereotypic sequences of events including the actors, the scenes and the likely actions. Stevens and Collins adapted this idea to store knowledge about rainfall including temporal and causal relationships.

In WHY, scripts are used as a basis of the questions to be asked and to assess the students' responses. The scripts are structured into hierarchies where general scripts give an overall view and the lower level ones supply detailed information about causal chains. Figure 2.3 illustrates the more natural dialogue achieved in this system compared with the early version of SCHOLAR.

WH Y: Student: WHY: Student: WH Y: Student: WH Y: Student: WH Y:

Do you think the Amazon jungle has heavy rainfall or light rainfall? (Rule 1: Ask for a prediction about a particular case.)

Heavy rainfall

Why does the Amazon have heavy rainfall? (Rule 2: Ask for a prior cause.)

Because it's near the Atlantic

Yes, the Amazon jungle is near the equatorial Atlantic. How does that affect rainfall in the Amazon jungle? (Rule 3: Ask for intermediate cause.)

The water in the current is warm

Do you think that any place with mountains has heavy rainfall?

(Rule 6: Form a general rule for an insufficient cause.) Yes

Southern California has mountains. Why doesn't Southern California have heavy rainfall?

(Rule 7: Pick a counter-example for an insufficient cause, and Rule 2: Ask for prior causes.)

The SCHOLAR and WHY projects were ambitious but both lacked some important aspects of human teaching, notably the ability to pitch material at a level appropriate to the student's knowledge and to deal effectively _with misconceptions. Most importantly, it is very difficult to conduct a sensible conversation on any topic without a great deal of real world knowledge. The technology was not available in the 1 970s to deal with these problems but, even so, these programs demonstrated that by storing information in a coherent and integrated form it can be used in a very flexible way.

A third influential pl'oject during the 1970s was SOPHIE (Brown et al. , 1975; Brown et al. , 1982). SOPHIE (SOPHisticated Instructional Environment) was the name given to a series of grograms developed between 1973 and 1 978 for teaching electronic troubleshooting, initially at the University of California and later at Bolt, Beranek and Newman Inc. (BBN) in Cambridge, Massachusetts.

Rather than just getting students to find faults in given circuits, the project team aimed to improve their conceptual understanding so that they could develop diagnostic procedures of their own, assimilate new information from technical manuals and troubleshoot unfamiliar equipment. A sample dialogue from an early version of the program is shown in Figure 2.4. In this example, the student is presented with a circuit that is not working and checks various components before suggesting possible faults. The system provides results of the tests and helps the student think through the ramifications of his/her hypotheses. As can be seen, the interaction style is very flexible.

Drawing on experience from SCHOLAR and WHY, the developers store knowledge in structures inside the computer. Unlike these earlier programs, however, knowledge is not just stored in a single form. SOPHIE uses three methods of representation: a simulation model of the domain, procedural specialists that contain logical skills and semantic nets for storing time-invariant factual knowledge.

One of the strengths of the later versions of SOPHIE is the modular design employed. An expert troubleshooter capable of locating and diagnosing a wide range of faults in electronic circuits provides the domain expertise. A coach monitors the student's attempts at diagnosis and fault finding, intervening as necessary. Finally, a sophisticated natural language processor is used in the interface. This separation of the different activities of the tutor was something that was being carried out on a regular basis in Britain at the time but, in America, the general emphasis was on domain representation rather than looking at the overall pedagogical strategy. The two approaches eventually converged and led to what are now generally called intelligent tutoring systems (IT_Ss).

SOP HIE : WOULD YOU LIKE TO START WITH A FAULTED INSTRUMENT WinCH NEEDS

FIXING?

Student : YES

SOP HIE : THE INSTRUMENT HAS A FAULT AND ITS PRESENT CONTROL SETTINGS

ARE:

CC 1 .0 (CURRENT CONTROL-FROM 0.0 (MIN CURRENT) TO 1 .0)

CR IDGH (CURRENT RANGE SWITCH, IDGH=l AMP, LOW=O.l AMP) LOAD 20 (LOAD RESISTANCE IN OHMS)

VC 1 .0 (VOLT AGE CONTROL-FROM 0.0 (MIN VOLTAGE) TO 1 .0)

VR IDGH (VOLTAGE RANGE SWITCH, HIGH=30 VOLTS, LOW=IO VOLTS) Student : WHAT IS THE OUTPUT?

SOP HIE : THE OUTPUT VOLTAGE IS 1 1 .7 VOLTS. Student : WHAT IS IT IN A WORKING INSTRUMENT?

SOP HIE : IN A WORKING CIRCUIT THE OUTPUT VOLTAGE IS 19.9 VOLTS. THAT VALUE

IS INCORRECT.

Student : _{WHAT IS THE BASE EMITTER VOLTAGE OF} THE VOLTAGE LIMITING TRANSISTOR?

SOP HIE : THE BASE EMITTER VOLTAGE OF Q5 IS -0.037 MILLIVOLTS. Student : WHAT ABOUT FOR THE CURRENT LIMITING TRANSISTOR? SOP HIE : THE BASE EMITTER VOLTAGE OF Q6 IS 0.349 VOLTS. Student : WHAT SHOULD IT BE?

SOP HIE : IN A WORKING CIRCUIT THE BASE EMITTER VOLTAGE OF Q6 IS 0.593 VOLTS. Student : REPLACE Q6

SOP HIE : I AM GOING TO ASK YOU SOME QUESTIONS ABOUT HOW Q6 IS FAULTED.

ARE ANY JUNCTIONS SHORTED? Student : I DONT KNOW

SOPHIE : ARE ANY TERMINALS OPEN? Student : YES

SOPHIE : WHICH TERMINALS ARE OPEN? Student : THE BASE

SOPHIE : Q6 IS NOT FAULTED IN THAT MANNER. Q6 NOT REPLACED.

Student : IS IT POSSIBLE THAT Q6 HAS AN OPEN BASE?

SOP HIE : THAT FAULT DOES NOT EXPLAIN ANY OF THE SYMPTOMS YOU OBSERVED BUT IT IS CONSISTENT WITH THE VALUES FOR THE BASE EMITTER VOLTAGE OF Q5 AND THE VOLTAGE AT N I .

Student : WHAT IS THE OUTPUT OF THE CONSTANT CURRENT SOURCE?

SOP HIE : THE OUTPUT OF THE CONSTANT CURRENT SOURCE IS -0.59 MILLIAMPS.

Student : WHAT SHOULD IT BE?

SOP HIE : IN A WORKING CIRCUIT THE OUTPUT CURRENT OF THE CONSTANT CURRENT SOURCE IS -0.6 MILLIAMPS.

Student : COULD THE PROBLEM BE THAT THE BASE OF Q3 IS OPEN? SOP HIE : HMMM, LET ME THINK ABOUT THAT FOR A MINUTE.

UNDER THE SETTINGS:

CC= l .O, CR=HIGH, LOAD=20, VC=l .O, VR=HIGH

THE FAULT WHICH YOU PROPOSED WILL PROPAGATE. LIKELY FAULTS WHICH MIGHT OCCUR ARE:

R22 OPEN

WOULD YOU LIKE ME TO CONTINUE TESTING YOUR HYPOTHESIS IN A CIRCUIT IN WHICH R22 IS OPEN ALSO?

Student : NO