I have done research with novice tutors myself and am aware of expert tutors' use of strategies and their interactions with (and the contributions of) tutees. I am aware that many current ITS are modeled after the behaviour of human tutors. However, how this actually works, how they select problems, how they model the tutees' understanding, how they choose the appropriate feedback, etc. is beyond me. I guess this is a big question as I assume that this involves natural language processing and generation as well as very sophisticated analysis of the tutees' actions and messages.
Where can I start in understanding the artificial intelligence involved? I am moderately comfortable with math and have done some basic programming (mostly Python). I would be very grateful for suggestions from the community.
It's a very big field as intelligent tutoring involves a whole host of artificial intelligence, knowledge representation, and interactivity techniques, with the exact techniques varying from application to application.
I suggest you start with the following:
"A review of recent advances in learner and skill modeling in intelligent learning environments" by Desmarais and Baker gives a good round up of current work including many algorithm families, and covers them in enough detail to give you the general idea without drowning you.
The book "Advances in Intelligent Tutoring Systems" by Nkambou, Mizoguchi, and Bordeau is also worth a look - it includes several chapters that give an overview of the field's main areas, then many more chapters addressing major approaches, illustrated with case studies.
The book "Building Intelligent Interactive Tutors" by Woolf also looks good, but I've not read it well enough to say much more.
Both of these are recent works, and should give you enough to answer the questions you need answered or, at the least, help you find the authors and research that you need.
If you're looking for a broad overview of the field for historic purposes or for conversational rather than technical knowledge, you might also get some mileage out of various older papers - the field of building intelligent agents for teaching has changed a lot over the years, and gone through several names.
"Teaching machine programs that generate problems as a function of interaction with students" by Uhr is one of the seminal early papers in the field (and gives a good intro to the sorts of things you have to think about in designing software to teach). I also found the article at http://www.cse.msu.edu/rgroups/cse101/ITS/its.htm to be very helpful as a summary of the whole field, and it counterbalances the more recent Nkambou book quite well (it's relatively light on older approaches).
ITs is really best thought of as a cluster of related problems rather than a cluster of solutions - most people who work in the field can agree on what the field's trying to achieve, but my experience has been that they're not so great at telling you what the preferred approach is - I don't think that kind of consensus exists yet, and it may be that context is so important that there will never be one 'ideal' approach.
I assume that you know the general structure of ITS i.e. it has four components namely tutoring/pedagogy module, expert module, problem solving environment (to which student interact) and student module. There are several ways to model expert knowledge. Mostly effectiveness in tutoring systems is shown by effective building of student models. There are several techniques to create student models: 1) Overlay model in which students' knowledge is compared with experts' knowledge to build a student model. 2) Perturbation technique which represents possible misconceptions together with missing concepts of the student. c3 Sterotypes technique which includes grouping of students according to several characteristics. 4) Model-tracing techniques, which use procedural rules/ production rules i.e. IF c1, c2 are true THEN take the action a. 5) Constraint based technique, which is more focused on the problem state in which student is and not what student is doing. 6) Machine Learning Techniques, include Bayesian belief networks, reinforcement learning, supervised learning, hidden Makov models, fuzzy logic and decision theory. 7) Expert systems techniques or Ontology based modeling in which ontology is used.
The answer to this question is quiet vast and at present I can cover this much in short.
You might like to look at my dissertation that 1) does a decent review of the literature and 2) and created a system that was roughly based on human tutoring dialog. But I have gotten away from model tracing tutors to make things easier for teachers and researchers to do work. See examples of how folks use ASSITments for psych sort of research here https://sites.google.com/site/assistmentsstudies/home
Apart from the general ITS characteristics, much of the AI is domain specific. If you are targeting ITS using language, latent semantic analysis might be an important concept for you. There are many tutoring programs in the field of text comprehension and automatic essay evaluation that are based on this statistical approcach. You might have a look at AutoTutor, Summary Street or in German language conText etc.
It's a very big field as intelligent tutoring involves a whole host of artificial intelligence, knowledge representation, and interactivity techniques, with the exact techniques varying from application to application.
I suggest you start with the following:
"A review of recent advances in learner and skill modeling in intelligent learning environments" by Desmarais and Baker gives a good round up of current work including many algorithm families, and covers them in enough detail to give you the general idea without drowning you.
The book "Advances in Intelligent Tutoring Systems" by Nkambou, Mizoguchi, and Bordeau is also worth a look - it includes several chapters that give an overview of the field's main areas, then many more chapters addressing major approaches, illustrated with case studies.
The book "Building Intelligent Interactive Tutors" by Woolf also looks good, but I've not read it well enough to say much more.
Both of these are recent works, and should give you enough to answer the questions you need answered or, at the least, help you find the authors and research that you need.
If you're looking for a broad overview of the field for historic purposes or for conversational rather than technical knowledge, you might also get some mileage out of various older papers - the field of building intelligent agents for teaching has changed a lot over the years, and gone through several names.
"Teaching machine programs that generate problems as a function of interaction with students" by Uhr is one of the seminal early papers in the field (and gives a good intro to the sorts of things you have to think about in designing software to teach). I also found the article at http://www.cse.msu.edu/rgroups/cse101/ITS/its.htm to be very helpful as a summary of the whole field, and it counterbalances the more recent Nkambou book quite well (it's relatively light on older approaches).
ITs is really best thought of as a cluster of related problems rather than a cluster of solutions - most people who work in the field can agree on what the field's trying to achieve, but my experience has been that they're not so great at telling you what the preferred approach is - I don't think that kind of consensus exists yet, and it may be that context is so important that there will never be one 'ideal' approach.
Thank you very much for all your helpful and comprehensive answers! I expected that this was a big question and I don't expect to become an expert on this any time soon... But I have many interesting starting points and I will be back with more questions if I come up against anything difficult.
Thank you especially, Wolfgang, for the German reference. As of now, I was of the impression that there is not a lot of research being done here. I was aware of the group around Niels Pinkwart, and of Gerhard Weber, but that's about it.
- Badges
- Science topic
- Similar topics
- Education