MET:Intelligent Tutoring System

This page was originally authored by Tracey Best and Shelby Budd (2007). This page has been revised by Paul Darvasi (2008), Bridget Perry-Gore (2011) and Mackenzie Moyer (2018).

An intelligent tutoring system (ITS) is a computer system based on artificial intelligence designed to deliver content and provide feedback to its user. The main goal of ITS is to interact with students one on one, similar to a human tutor. The program personalizes the instruction based on the background and the progress of every individual student. Students are given the freedom to ask questions and complete complex problems and assignments with effective and efficient feedback provided by the ITS.

The development of modern ITS authoring systems allows for the rapid development of ITSs. The ease of using these systems means that educators can quickly develop ITS content for any type of subject. For an example of an ITS authoring system, see Cognitive Tutor Authoring tools. The development of rapid e-Learning tools has seen the proliferation of ITS principles (such as computer-based branching and feedback) move beyond ITS authoring tools.

History

The concept of using computers to assist learning was originally known as "computer assisted instruction"(CAI) or "computer assisted learning"(CAL)[[1]]and has been around since the 1950's. The main problem with the older systems is that they did not provide feedback nor could they been individualized for the students (Nwana, 1990). These systems have now evolved into the "Intelligent Tutoring Systems".

Intelligent Tutoring Systems have their base in the Artificial Intelligence (AI) movement that occurred during the 1950s and 1960s. Many innovative thinkers like Alan Turing, Marvin Minsky, John McCarthy and Allen Newell believed that computers that could think just like humans would soon be developed.

It was not until the 1980s that the first intelligent tutoring systems were recognized. During this time those working in the realm of artificial intelligence realized the underlying problem with creating thinking computers - the assumption that people thought like computers. (Lurain, 2007) The focus shifted to creating expert systems that allowed for multiple solutions to a problem instead of one expert solution.(Nwana, 1990)

1950s

The linear programs created by B.F. Skinner guided the student in a linear fashion using very simple questions which slowly advanced the student. Incorrect answers were not expected if the designer had done the job properly and the student moved ahead. Students proceeded regardless of their answers.

1960s

Crowder incorporated the learner's responses by using them to chose the next question. There were still a limited number of questions. His branching program used pattern-matching techniques allowing for partially correct answers.

1970s

A breakthrough comes in that computers could be used to generate and solve problems:"generative CAI". This was especially useful in mathematics as it allowed allowed for systems to use less memory as questions did not need to be pre-stored. It is now possible to adapt to the level of ability to some degree (Nwana, 1990).

1980-2000

Improvements in ITS feedback methods.

Present and future

ITSs can now read user moods: Beverly Woolf, a researcher at the University of Massachusetts Amherst, is one of a few pioneers developing a computer that can identify its user’s mood and respond appropriately with encouragement, empathy or advice. In order to determine an individual's emotions they to use sensors and cameras to record heart rate and read facial expression. At the moment they can determine the mood with up to 80% accuracy (Gulli, 2011).

Some ITSs can identify and categorize students' off-task behaviours as productive or unproductive, a data type that can be useful for educators. (Baker et al., 2007)

The capabilities of ITSs have also expended with the advancement of AI more generally. For instance, ITSs capable of natural language have now been developed. (Litman & Silliman, 2004; Graesser, 2001)

Benefits and Limitations

Main article: Educational Benefits & Limitations

Meta-analysis has shown that ITSs can improve academic performance. In addition, "the difference in learning outcomes between ITS and...human tutoring was not statistically significant." (Ma & Adisope, 2014) As others have argued, human tutors are generally hindered by a lack of instructional training and subject matter knowledge. (Graesser et al., 2001) ITS is associated with significantly higher achievement outcomes than large-group instruction, individual work books, and non-ITS forms of computer based instruction.(Ma & Adisope, 2014)

It has been suggested that ITSs used in K-12 education have the best outcomes with students without learning challenges and when used for no more than a year. The effectiveness of ITS use begins declining after a year of use. (Steenbergen-Hu & Cooper, 2013) Additional work has shown that ITSs for computer science are more effective when students students learn in pairs.(Harsley, 2017)

Teacher knowledge and decision-making have been shown to be influenced by ITS use. In a quasi-experimental study based on 8 classes of grade 5 students, when teachers were able to access ITS data on their students (in the form of a dashboard), these teachers adapted their lesson plans and what they covered in class. (Xhakaj, Aleven & McLaren, 2017)

While ITSs can be effective instructional tools, there are two bottlenecks that limiting the growth of ITS implementation: a shortage of experts of ITS authoring tools (authoring tools allow for the most rapid development of ITS content, as opposed to programming an ITS directly, which would be even more time consuming), and the amount of time these experts take to produce ITS lessons. For reference, an expert ITS author requires several hundred hours to produce a single hour of ITS lesson content using an authoring tool.(Olney & Cade, 2015)

Mostly due to the limits of technological affordances, intelligent tutoring systems are often subject to educational benefits and limitations. The fact that we are close to creating a system which can effectively read emotions means that we are closing in on the idea that systems can gauge and respond to a student's emotional state. With more research and development we should get to a time where ITS agents will model human behaviour and consider the students emotional state and proceed accordingly.

Initial research suggests that there are three main emotional spectrums to which an ITS could ideally respond:

1) Tutorial Session: from boredom to curiosity

2) Student's own results: from distress to enthusiasm

3) Attitude towards learning: from anxiety to confidence

Determining the student's emotions must be done before the system is called into action to react and there are various ways that this can be done. Kalisz (2006) suggests three methods by which the ITS can determine the student's emotions:

1) Watching the sequence of tutor-student interactions.

2) Tracking movements and physiological responses using cameras and tracking devices. This has been done by Beverly Wolf at the University of Massachucetts.

3) Studying answers provided by the student.

Structure

ITS Components

ITS models typically have the following four components:

Domain model: the area where the expert knowledge and behaviour is located
Student model: assesses the student's level of knowledge and is where the AI functioning of the system is essential so students are directed properly in their learning needs
Teaching model: houses the lessons and activities for completion
Learning environment: the interface with which the student interacts.

Different from other computer-based learning situations ITSs use highly interactive learning situations with simulations. The student's performance instructs the ITS; instruction is tailored for specific learning outcomes.

Examples in Mathematics

Mathematics is a unique subject in that the material accumulates throughout elementary school and up through high school. It is necessary to have a good source of feedback and help in order to correctly build knowledge. There are some interesting examples of Intelligent Tutoring in Mathematics such as the following.

Active Math [2] was developed by the University of Saarland in Germany in conjunction with the German Center for Research in Artificial Intelligence:it is a complete ITS for mathematics education. This system is adaptive to students, diagnoses errors made by students and diagnoses misconceptions. The site is aimed at university courses such as calculus. Active MAth is free and it is also possible to integrate it into a link title Moodle.

Assistments [3] was built in response to the No Child Left behind policy in the US. In Massachusetts, standardized tests in are administered to all grade 3-10 students in the public funded system. These tests in English, math, history and science are rigorous and are taken very seriously; students need to pass math and English portions of the tenth grade test to graduate with a high school diploma. In 2003, 10% of high school seniors were predicted to fail (after 4 tries) and the state singled out grade 8 mathematics as the weak spot. The push was on to not only improve the level of mathematics but also to predict the outcomes on the final test. Recorded formative evaluation became expected with teachers knowing that this took up time in the classroom which is normally used for instruction (Feng et al., 2010)

Various companies got busy creating programs which would evaluate automatically, however a math teacher by the name of Neil Heffernan saw a different angle. His team would create a different kind of program. Funded by the Department of Education, a group at Worcester Polytechnic Institute (WPI) built a web-based tutoring system that would "assist" and "assess" at the same time. With these two integrated into the same system, students are offered instruction and are also provided with a more detailed evaluation of their abilities. Students are given scaffolding and hints when they ask and teachers can evaluate and monitor their progress.

Cognitive Tutor [4] is an ITS from Carnegie learning Inc and is a well-researched and highly regarded system. It uses form of blended learning which combines textbooks and MATHia software. They concepts are well-researched and they are aware of efa ct that motivation is a big factor in learning. By tailoring the instruction to the interest of students and giving them real life problems to solve, they tap into this motivation The combination of their textbooks and software provide formative assessment, relevant problem-centered activities which help develop mathematical reasoning all in a personalized learning environment (Ritter, 2010). They want to get students to think, not become rote learners.

At a cost of approximately $99/student/year it can be looked at as expensive as a school or cheap for an individual compared to the price of a tutor.

Realise-IT [5] was created by an Irish company called CCKF. They have constructed a program that will assess a learner’s prior knowledge and optimize their learning path and objectives mapping in an area that they are about to study.

Wayang Outpost [6] is a system designed to prepare middle and high-school students in their preparation for standardized math tests, such as the SAT, MCAS and CA-Star. It uses multimedia and animated adventures based on an outpost location called Wayang, in the rainforest, to help the student progress through various math concepts. The program includes tutoring, videos, hints a variety of support for the student. and the math problem presented. Wayang Outpost adjusts instruction, using individualized strategies that are effective for each student.

Wayang Outpost is free to teachers, schools, after-school programs, and for use from home.

Association for the Advancement of Artificial Intelligence

The website of the AAAI (Association for the Advancement of Artificial Intelligence) [7]describes itself as a “scientific society devoted to advancing the scientific understanding of the mechanisms underlying thought and intelligent behavior and their embodiment in machines.” One of the primary research areas identified by the AAAI is Intelligent Learning Systems (sometimes referred to as Intelligent Tutoring Systems). No other area more fundamentally highlights the cognitive interaction of man and machine than this research area(CCKF 2010).

References

Baker, R.S., Corbett, A.T. & Koedinger, K.R., Detecting Student Misuse of Intelligent Tutoring Systems. Retrieved February 15, 2007, from http://www.psychology.nottingham.ac.uk/staff/lpzrsb/BCK2004MLFinal.pdf

CCKF Website. http://www.cckf-it.com/where-artificial-intelligence-meets-human-cognitive-science/#more-125

Davidovic, A. (2001). Learning benefits of structural example-based adaptive tutoring systems. Phd. University of South Australia, School of Computer and Information Science. Retrieved February 19, 2007, from http:\\ariic.libary.unsw.edu.au/unisa.

Feng, M., Heffernan, N.T., & Koedinger, K.R. (2010). Addressing the assessment challenge in an Online System that tutors as it assesses. User Modeling and User-Adapted Interaction: The Journal of Personalization Research (UMUAI journal). Volume 19, p.243-26.

Graesser, A., VanLehn, K., Rosé, C., Jordan, P. & Harter, D. (2001) AI Magazine, 22(4), 39-52. Retrieved from https://ocs.aaai.org/ojs/index.php/aimagazine/article/view/1591

Gulli, Cathy (2010). ‘Emotional’ Computers High-tech tutors encourage, empathize with math students. Macleans.ca Retrieved on June 28th, 2011 from http://www2.macleans.ca/2010/04/01/%E2%80%98emotional%E2%80%99-computers/

Harsley, R., Di Eugenio, B., Green, N., & Fossati, D. (2017). Collaborative Intelligent Tutoring Systems: Comparing Learner Outcomes Across Varying Collaboration Feedback Strategies. Philadelphia, PA: International Society of the Learning Sciences. Retrieved from: https://repository.isls.org/bitstream/1/302/1/95.pdf

Intelligent Tutoring Systems (a subtopic of Education). Retrieved February 15, 2007, from http://www.aaai.org/AITopics/html/tutor.html

Intelligent Tutoring Systems (n.d.). Retrieved February 17, 2007, from http://pachome2.pacific.net.sg/~auddrick/its.html

Kalisz, Eugenia, Florea, M.A.(2006). Could Intelligent Tutors Anticipate Successfully User Reactions? Retrieved January 25, 2008, from Academic Search Premiere.

Lane, Chad. (2006). Intelligent Tutoring Systems: Prospects for Guided Practice and Efficient Learning. Institute for Creative Technologies, University of Southern California. Retrieved on June 10th, 2011 from http://tinyurl.com/6bxt72q

Litman, D. J., & Silliman, S. (2004, May). ITSPOKE: An intelligent tutoring spoken dialogue system. In Demonstration papers at HLT-NAACL 2004 (pp. 5-8). Association for Computational Linguistics.

Leddo, J. & Kolodziej, J. (1997). Distributed interactive intelligent tutoring simulation. ERIC Document Access Code ED416319.

Ma, W. & Adesope, O. (2014). Intelligent Tutoring Systems and Learning Outcomes: A Meta-Analysis. Journal of Educational Psychology, 106(4), 901-918. doi: 10.1037/a0037123

Nwana, Hyacinth. (1990). Intelligent Tutoring Systems: An Overview. Artificial Intelligence Review, 4, 251-277. Retrieved June 10th, 2011 from https://svn.v2.nl/andres/.../Intelligent%20Tutoring%20Systems.pdf

Olney, A. M., & Cade, W. L. (2015). Authoring Intelligent Tutoring Systems Using Human Computation: Designing for intrinsic motivation. In International Conference on Augmented Cognition, 628-639. Retrieved from: https://link.springer.com/chapter/10.1007/978-3-319-20816-9_60

Ong, J. & Ramachandran, S. (2000). Intelligent Tutoring Systems: The What and The How. Retrieved February 14, 2007, from http://www.learningcircuits.org/2000/feb2000/ong.htm.

Piramuthu, S. (2005). Knowledge-Base Web-Enabled Agents and Intelligent Tutoring Systems. Retrieved January 24, 2008, from Academic Search Premiere

Ross, S. & Casey, J. (1994). Using Interactive Software to Develop Students' Problem Solving Skills: Evaluation of the "Intelligent Physics Tutor." ERIC Document Access Code ED373754.

Ritter, Steve. (2010). The Research Behind the Carnegie Learning Math Series. Retrieved June 25th, 2011 from http://www.carnegielearning.com/

Steenbergen-Hu, S., & Cooper, H. (2013). A meta-analysis of the effectiveness of intelligent tutoring systems on K–12 students’ mathematical learning. Journal of Educational Psychology, 105(4), 970-987. doi: 10.1037/a0032447

Thomas, Eric. (n.d.). Intelligent Tutoring Systems (ITS). Retrieved February 14, 2007, from http://coe.sdsu.edu/eet/Articles/tutoringsystem/start.htm

Urban - Lurain, M. Intelligent Tutoring Systems: An Historic Review in the Context of the Development of Artificial Intelligence and Educational Psychology. Retrieved February 12, 2007, from http://www.cse.msu.edu/rgroups/cse101/ITS/its.htm.

Xhakaj, F., Aleven, V., & McLaren, B. M. (2017, June). Effects of a Dashboard for an Intelligent Tutoring System on Teacher Knowledge, Lesson Plans and Class Sessions. In International Conference on Artificial Intelligence in Education (pp. 582-585). Springer, Cham. Retrieved from https://pdfs.semanticscholar.org/ecbc/d7edad0aa94e6ea33756c8ed41129e111f0d.pdf

External Links