Artificial Intelligence - Intelligent Tutoring Systems.

  

Intelligent tutoring systems are artificial intelligence-based instructional systems that adapt instruction based on a variety of learner variables, such as dynamic measures of students' ongoing knowledge growth, personal interest, motivation to learn, affective states, and aspects of how they self-regulate their learning.

For a variety of problem areas, such as STEM, computer programming, language, and culture, intelligent tutoring systems have been created.

Complex problem-solving activities, collaborative learning activities, inquiry learning or other open-ended learning activities, learning through conversations, game-based learning, and working with simulations or virtual reality environments are among the many types of instructional activities they support.

Intelligent tutoring systems arose from a field of study known as AI in Education (AIED).

MATHia® (previously Cognitive Tutor), SQL-Tutor, ALEKS, and Rea soning Mind's Genie system are among the commercially successful and widely used intelligent tutoring systems.

Intelligent tutoring systems are frequently more successful than conventional kinds of training, according to six comprehensive meta-analyses.

This efficiency might be due to a number of things.

First, intelligent tutoring systems give adaptive help inside issues, allowing classroom instructors to scale one-on-one tutoring beyond what they could do without it.

Second, they allow adaptive problem selection based on the understanding of particular pupils.

Third, cognitive task analysis, cognitive theory, and learning sciences ideas are often used in intelligent tutoring systems.

Fourth, the employment of intelligent tutoring tools in so-called blended classrooms may result in favorable cultural adjustments by allowing teachers to spend more time working one-on-one with pupils.

Fifth, more sophisticated tutoring systems are repeatedly developed using new approaches from the area of educational data mining, based on data.

Finally, Open Learner Models (OLMs), which are visual representations of the system's internal student model, are often used in intelligent tutoring systems.

OLMs have the potential to assist learners in productively reflecting on their current level of learning.

Model-tracing tutors, constraint-based tutors, example-tracing tutors, and ASSISTments are some of the most common intelligent tutoring system paradigms.

These paradigms vary in how they are created, as well as in tutoring behaviors and underlying representations of domain knowledge, student knowledge, and pedagogical knowledge.

For domain reasoning (e.g., producing future steps in a problem given a student's partial answer), assessing student solutions and partial solutions, and student modeling, intelligent tutoring systems use a number of AI approaches (i.e., dynamically estimating and maintaining a range of learner vari ables).

To increase systems' student modeling skills, a range of data mining approaches (including Bayesian models, hidden Markov models, and logistic regression models) are increasingly being applied.

Machine learning approaches, such as reinforcement learning, are utilized to build instructional policies to a lesser extent.

Researchers are looking at concepts for the smart classroom of the future that go beyond the capabilities of present intelligent tutoring technologies.

AI systems, in their visions, typically collaborate with instructors and students to provide excellent learning experiences for all pupils.

Recent research suggests that rather than designing intelligent tutoring systems to handle all aspects of adaptation, such as providing teachers with real-time analytics from an intelligent tutoring system to draw their attention to learners who may need additional support, promising approaches that adaptively share regulation of learning processes across students, teachers, and AI systems—rather than designing intelligent tutoring systems to handle all aspects of adaptation, for example—by providing teachers with real-time analytics from an intelligent tutoring system to draw their attention to learners who may need additional support.

~ Jai Krishna Ponnappan

You may also want to read more about Artificial Intelligence here.

See also: 

Natural Language Processing and Speech Understanding; Workplace Automation.

Further Reading:

Aleven, Vincent, Bruce M. McLaren, Jonathan Sewall, Martin van Velsen, Octav Popescu, Sandra Demi, Michael Ringenberg, and Kenneth R. Koedinger. 2016. “Example-Tracing Tutors: Intelligent Tutor Development for Non-Programmers.” International Journal of Artificial Intelligence in Education 26, no. 1 (March): 224–69.

Aleven, Vincent, Elizabeth A. McLaughlin, R. Amos Glenn, and Kenneth R. Koedinger. 2017. “Instruction Based on Adaptive Learning Technologies.” In Handbook of Research on Learning and Instruction, Second edition, edited by Richard E. Mayer and Patricia Alexander, 522–60. New York: Routledge.

du Boulay, Benedict. 2016. “Recent Meta-Reviews and Meta-Analyses of AIED Systems.” International Journal of Artificial Intelligence in Education 26, no. 1: 536–37.

du Boulay, Benedict. 2019. “Escape from the Skinner Box: The Case for Contemporary Intelligent Learning Environments.” British Journal of Educational Technology, 50, no. 6: 2902–19.

Heffernan, Neil T., and Cristina Lindquist Heffernan. 2014. “The ASSISTments Ecosystem: Building a Platform that Brings Scientists and Teachers Together for Minimally Invasive Research on Human Learning and Teaching.” International Journal of Artificial Intelligence in Education 24, no. 4: 470–97.

Koedinger, Kenneth R., and Albert T. Corbett. 2006. “Cognitive Tutors: Technology Bringing Learning Sciences to the Classroom.” In The Cambridge Handbook of the Learning Sciences, edited by Robert K. Sawyer, 61–78. New York: Cambridge University Press.

Mitrovic, Antonija. 2012. “Fifteen Years of Constraint-Based Tutors: What We Have Achieved and Where We Are Going.” User Modeling and User-Adapted Interaction 22, no. 1–2: 39–72.

Nye, Benjamin D., Arthur C. Graesser, and Xiangen Hu. 2014. “AutoTutor and Family: A Review of 17 Years of Natural Language Tutoring.” International Journal of Artificial Intelligence in Education 24, no. 4: 427–69.

Pane, John F., Beth Ann Griffin, Daniel F. McCaffrey, and Rita Karam. 2014. “Effectiveness of Cognitive Tutor Algebra I at Scale.” Educational Evaluation and Policy Analysis 36, no. 2: 127–44.

Schofield, Janet W., Rebecca Eurich-Fulcer, and Chen L. Britt. 1994. “Teachers, Computer Tutors, and Teaching: The Artificially Intelligent Tutor as an Agent for Classroom Change.” American Educational Research Journal 31, no. 3: 579–607.

VanLehn, Kurt. 2016. “Regulative Loops, Step Loops, and Task Loops.” International Journal of Artificial Intelligence in Education 26, no. 1: 107–12.

Artificial Intelligence - What Are Cognitive Architectures?

 

A cognitive architecture is a customized computer model of the human mind that aims to imitate all elements of human cognition completely.

Cognitive architectures are coherent theories that explain how a set of fixed mental structures and mechanisms may do intelligent work in a range of diverse surroundings.

A cognitive architecture has two main components: a theory of how the human mind works and a computing representation of the theory.

The cognitive theory that underpins a cognitive architecture will attempt to bring together the findings of a wide range of experiments and hypotheses into a single, comprehensive framework capable of describing a wide range of human behavior utilizing a set of evidence-based processes.

The framework established in the theory of cognition is then used to build the computational representation.

Cognitive architectures like ACT-R (Adaptive Control of Thought Rational), Soar, and CLARION (Connectionist Learning with Adaptive Rule Induction On-line) can predict, explain, and model complex human behavior like driving a car, solving a math problem, or recalling when you last saw the hippie in the park by combining modeling behavior and modeling the structure of a cognitive system.

There are four techniques to reaching human-level intelligence inside a cognitive architecture, according to computer scientists Stuart Russell and Peter Norvig: 

(1) constructing systems that think like people, 

(2) constructing systems that think logically, 

(3) constructing systems that behave rationally, and 

(4) constructing systems that act like humans.

The behavior of a system that thinks like a person is produced using recognized human processes.

This is the most common technique in cognitive modeling, as shown by structures such as John Anderson's ACT-R, Allen Newell and Herb Simon's General Problem Solver, and the first applications of the general cognitive architecture known as Soar.

For example, the ACT-R model combines theories of physical movement, visual attention, and cognition.

The model differentiates between declarative and procedural knowledge.

Production rules, which are statements expressed as condition action pairs, are used to express procedural knowledge.

A statement written in the form of IF THEN is an example.

Declarative knowledge is grounded on reality.

It refers to data that is considered static, such as characteristics, events, or objects.

This sort of architecture will provide behavior that contains both right and incorrect behavior.

Instead, a system that thinks rationally would employ logic, computational reasoning, and rules of mind to create consistent and correct behaviors and outputs.

A rational system will employ intrinsic ideas and knowledge to attain objectives via a more broad logic and movement from premises to consequences, which is more flexible to circumstances when complete information is not available.

The rational agent method is another name for acting rationally.

Finally, the Turing Test technique may be viewed of as creating a system that behaves like a person.

To attain humanlike behavior, this technique requires the development of a system capable of natural language processing, knowledge representation, automated reasoning, and machine learning in its most stringent form.

This method does not require every system to achieve all of these requirements; instead, it focuses on the standards that are most important to the job at hand.

Apart from those four methods, cognitive architectures are divided into three categories based on how they process information: symbolic (or cognitivist), emergent (or connectionist), and hybrid.

Symbolic systems are controlled at a high level from the top down and analyze data using a series of IF-THEN statements known as production rules.

EPIC (Executive-Process/Interactive Control) and Soar are two examples of symbolic information processing cognitive systems.

Emergent systems, like neural networks, are constructed by a bottom-up flow of information propagating from input nodes into the remainder of the system.

Emergent systems like Leabra and BECCA (Brain-Emulating Cognition and Control Architecture) employ a self-organizing, distributed network of nodes that may function in parallel.

ACT-R and CAPS (Collaborative, Activation-based, Production System) are examples of hybrid architectures that integrate elements from both forms of information processing.

A hybrid cognitive architecture geared at visual perception and understanding, for example, may utilize symbolic processing for labels and text but an emergent method for visual feature and object recognition.

As particular subtasks become more known, this kind of mixed-methods approach to developing cognitive architectures is becoming increasingly widespread.

This may lead to some confusion when categorizing designs, but it also leads to architectural improvement since the best solutions for each subtask can be included.

In both academic and industry settings, a number of cognitive architectures have been created.

Because of its sturdy and speedy software, Soar is one of the most well-known cognitive architectures that has gone out into industrial applications.

Digital Equipment Corporation (DEC) utilized a proprietary Soar program named R1-Soar to help with the complicated ordering process for the VAX computer system in 1985.

Previously, each component of the system (from software to cable connections) would have to be planned out according to a complicated set of rules and eventualities.

The R1-Soar system used the Soar cognitive architecture to automate this operation, saving an estimated $25 million each year.

Soar Technology, Inc. maintains Soar's industrial implementation, and the business continues to engage on military and government projects.

John Anderson's ACT-R is one of the most well-known and researched cognitive architectures in academia.

ACT-R builds on and expands on previous architectures such as HAM (Human Associative Memory).

Much of the research on ACT-R has focused on expanding memory modeling to include a wider range of memory types and cognitive processes.

ACT-R, on the other hand, has been used in natural language processing, brain area activation prediction, and the construction of smart tutors that can model student behavior and tailor the learning curriculum to their unique requirements.

~ Jai Krishna Ponnappan

You may also want to read more about Artificial Intelligence here.

See also: 

Intelligent Tutoring Systems; Interaction for Cognitive Agents; Newell, Allen.

Further Reading

Anderson, John R. 2007. How Can the Human Mind Occur in the Physical Universe? Oxford, UK: Oxford University Press.

Kotseruba, Iuliia, and John K. Tsotsos. 2020. “40 Years of Cognitive Architectures: Core Cognitive Abilities and Practical Applications.” Artificial Intelligence Review 53, no. 1 (January): 17–94.

Ritter, Frank E., Farnaz Tehranchi, and Jacob D. Oury. 2018. “ACT-R: A Cognitive Architecture for Modeling Cognition.” Wiley Interdisciplinary Reviews: Cognitive Science 10, no. 4: 1–19.