OVERVIEW OF INTELLIGENT TUTORING SYSTEMS
Intelligent Tutoring Systems research—more broadly known as the field
of “AI and Education” [1]—has been concerned with individualized
instruction in many different domains, with a variety of
representational media and interactive methods [27]. It should be
mentioned at the outset that ITS systems work, and they can work quite
well at teaching STEM topics compared to teacher-to-student tutoring
[1]. Furthermore, it is important to note that ITS systems generally
have not been intended as replacements for human teachers or tutors, but
rather designed as tools to assist in classwork and for independent
learning.
ITS work can be traced to the 1960s with the development of AI programs
that represent knowledge in structured models , especially
semantic nets, production rules, and schemas/frames. Programs using such
models can solve problems, such as answering factual questions, proving
theorems, and manipulating mathematical equations. Subsequent research
in the 1970s added a reasoning module that interprets the structured
model in particular situations for professional tasks, such as
diagnosis, planning, design, and process control; these programs were
called “expert systems.”
In general, an intelligent tutoring program contains such
an AI problem-solving program, using it to interact with and instruct a
student. Thus ITS is contrasted with computer-based instruction programs
that do not have a built-in capability to solve the problems that are
presented to students.
Most intelligent tutoring programs engage the student in a learning
activity in which the program serves as an instructor; they use distinct
models of the domain, the student, and curriculum; and the interactive
design is based on a theory of the pedagogical process [13, 23, 24,
27].
ITS research has been concerned with teaching mathematics [1] and
basic science, as well as professional expertise relating to complicated
systems, such as electro-mechanical troubleshooting [3], engineering
operations [16], and medicine [9]. Insofar as machine learning
programs are complicated systems whose capabilities and, to some extent,
methods we want users to understand, the techniques and lessons from ITS
research and development over nearly 50 years are worth considering for
adoption in XAI research.