[ CogSci Summaries home | UP | email ]
http://www.jimdavies.org/summaries/
Conati, C. (2009). Intelligent Tutoring Systems: New Challenges and Directions. Proceedings of the 21st International Joint Conference on Artificial Intelligence, Pasadena, California, 2-7
 


@InProceedings{Conati2009, 
 author      = {Conati, Cristina}, 
 title       = {Intelligent Tutoring Systems: New Challenges and Directions }, 
 booktitle   = {Proceedings of the 21st International Joint Conference on Artificial Intelligence}, 
 year        = {2009}, 
 pages       = {2--7} 
}

Author of the summary : Marc Meyer, 2010, barefootmarc@gmail.com
Cite this paper for :
[p2]"Intelligent Tutoring Systems ... combine research in Artificial Intelligence, Cognitive Science, and Education to create intelligent agents that act as tutors during computer-aided instruction.

Traditional CAI systems need to define a huge number of components ( problems, solutions, actions that may be taken) so are difficult to scale up as the expected tasks become more complex.

CAI can be adaped better to individual user's needs by giving the ITS knowledge of the teaching process itself and the ability to adjust its output based on what it learns about the user. The system needs to have knowledge about what needs to be taught ( domain model ), about the user ( student model), and about teaching strategies that might apply ( pedagogical model). It also needs to know how to present its information clearly ( communication knowledge). A system may include some or all four of these components, and the degree of sophistication of each may vary. Some may employ little more than a simple set of heuristics.

[p2]"Using the domain knowledge and its communication knowledge, the tutor presents the selected problem to the student in the format most suitable for the student's abilities and preferences. Then it monitors the student's solution to the problem and compares it with its known solution ( or set of relevant alternate solutions ) to decide whether the student's solution is appropriate or requires pedagogical interventions."

Each specific, possible solution does not need to be separately defined. They can be generated in real-time based on the problem definition and knowledge of the instructional domain.

[p3]"ITS research has successfully delivered techniques and systems that provide adaptive support for student problem solving or question-answering activities in a variety of domains ( e.g. ... physics, algebra ...)...There are, however, other educational activities that can benefit ... such as learning from examples, exploring interactive simulations, playing educational games, and learning with a group of peers."

Meta-cognition involves developing the skills required for learning, as opposed to being taught specific items of information. Skills that might be taught are how to monitor one's own progess ( self-monitoring),how to clarify material that needs further elaboration before it is understood ( self-explanation), how to use examples effectively ( analogical reasoning ), and the ability to seek help as needed. Within analogical reasoning may be taught the specific skill of min-analogy.

[p3]"Min-analogy involves transferring from an example only the minimum amount of information necessary to enable successful problem solving."

EXBL Description
[p3]"...an ITS that takes into account individual differences in (self-explanation and min-analogy ) skills to provide user-adaptive support to example-based learning."

EXBL Components
[p3]"...the SE-Coach supports example studying prior to problem solving ...the EA-Coach supports the effective use of examples during problem solving ..."

EXBL Architecture
[p4]The user studies examples before starting (SE-Coach) and uses examples to solve problems they are given (EA-Coach). The user's actions are compared to a solution graph, which is the system's self-generated representation of the relevant solutions to the problems and examples that the user has been given. There are various components which work together before the session to generate the solution graph. A model of the student's learning history initializes the rule nodes in a Bayesian Network, which serves as a sort of short-term memory that activates every time a student opens a new exercise, and which contains the system's assessment of the user's current knowledge and skill level.

[p4]'this ...allows the system to generate tailored interventions to foster effective meta-cognitive skills when the modela ssesses the student as having knowledge gaps or requiring improvement in her meta-cognitive behaviours."

[p4]...based on its current assessment of the student's knowledge, the student's reading patterns ... and student explanations on the example that the student can generate ...the SE-Coach guides the student to more carefully explain parts of the example that may not be fully understood ... automatically generates solutions at different levels of detail, and helps students generate the missing solution steps ... to support effective ... problem solving by selecting for each student and current problem an example that maximises both problem solving success and student learning."

The system assesses how effectively a user is exploring, based on their actions, their knowledge, and their self-explanation of their exploring behaviour.

ACE Description
[p5]...an ITS that supports student exploration of mathematical functions via a set of interactive simulations ..."

The student is encouraged to alter certain features of an equation and to observe the resultant effect on how that equation is plotted on a plane.

[p5]Ace monitors the student's interactions with its simulations, and generates interventions to improve those behaviors deemed to be suboptimal...(such as)...which further exploratory actions to perform when a student's exploration of a given activity is incomplete.

ACE uses a probabilistic model of the student's interactions. Its DBN has nodes corresponding to all possible cases the student might explore, and nodes corresponding to the mathematical concepts needed to understand the material. The model also encodes how exploring different examples ( of the former ) will affect the user's knowledge ( of the latter ). ACE monitors how extensively the user is exploring the sample space and intervenes accordingly.

[5]...ACE sometimes overestmated students' exploratory behavior...a student who quickly moves a function plot around the screen, but never reflects on how these movements change the function e[p2]quation, is performing many exploratory actions,but can hardly learn from them."

As a result improvements were made to the system. An eye-tracking system was incorporated which measures not only the duration of each activity but also the users attention to that activity. To this was added the users own information about their explorations ( self-explanation) as gathered by the system.

[p6]As previously mentioned ITS can support a broad range of educational activities, and other investigations include those by Isotami and Misogici ( learning with a group of peers ), Leelawong and Biswas ( use "teachable agents",taught by the user, as peers )Conati, McLaren, and D'Mello ( tutor systems that take the student's emotional affect into account ), Manske, Conati, and Johnson ( playing educational games ) and Lynch ( ITS for poorly defined knowledge domains ).

Building meta-cognitive skills requires ITS that can model knowledge domains and student behaviours that are less clearly defined than those modelled by the earlier generations of CAI systems.

[p6]"Advances in AI techniques for reasoning under uncertainty, machine learning, decision-theoretic planning ...(and)... increasing availability of sensors that can help capture the relevant user states, are promising means ... to face these challenges."

Summary author's notes :