Artificial intelligence, especially deep learning techniques, are being used to improve student modeling and personalize education. Researchers have used recursive neural networks for knowledge tracing and found they can model student learning better than previous models. Deep learning models can analyze student responses without human annotations. Classroom engagement systems allow real-time responses from all students, but more advanced applications are still needed for more in-depth response evaluation and feedback. Studies show step-based tutoring systems that provide scaffolding and feedback at different levels can be as effective as human tutors in helping students learn.
Artificial intelligence, especially deep learning techniques, are being used to improve student modeling and personalize education. Researchers have used recursive neural networks for knowledge tracing and found they can model student learning better than previous models. Deep learning models can analyze student responses without human annotations. Classroom engagement systems allow real-time responses from all students, but more advanced applications are still needed for more in-depth response evaluation and feedback. Studies show step-based tutoring systems that provide scaffolding and feedback at different levels can be as effective as human tutors in helping students learn.
Artificial intelligence, especially deep learning techniques, are being used to improve student modeling and personalize education. Researchers have used recursive neural networks for knowledge tracing and found they can model student learning better than previous models. Deep learning models can analyze student responses without human annotations. Classroom engagement systems allow real-time responses from all students, but more advanced applications are still needed for more in-depth response evaluation and feedback. Studies show step-based tutoring systems that provide scaffolding and feedback at different levels can be as effective as human tutors in helping students learn.
Development of education in the era of artificial intelligence
Artificial intelligence (AI) is defined as the ability of computer-controlled devices to do tasks
like human beings. In the evolution of artificial intelligence, amazing developments are made that have played a significant role in human lives. There are real examples of Artificial intelligence applications which are capable of learning the human thinking pattern. Similarly, using AI in the education sector has brought remarkable success in both practice and theory in the new millennium. Deep Learning has appeared as a breakthrough methodology for executing machine learning through neural networks to accomplish extremely complex thinking duties. (Goksel & Bozkurt, 2019, pp. 225–227, 232) Neural networks give evidence of concept that aid in addressing difficult questions like the extent to which visual object identification needs repeated processing. The neural network model comprises linked units, each unit adding a weighted sum of inputs & passing it via a non-linear activation function. Such unit networks can implement randomly complicated functions among inputs (Such as photographs) & outputs (for instance, main object name in every photograph), like compositions of liner or nonlinear sub-functions. Neural network models outshine at pattern identification and pattern generation task, no matter the patterns are dynamic or static. However, they are still far behind the humans like the ability to learn novel concepts from one time experience, to make wide-ranging inferences from the minute details, and generalize successfully to entirely new domains. Similar to biological neural networks, models do not step into the world with the knowledge about everything, about their surroundings or about ways to accomplish activities that will be needed from them. Rather than this, they learn. There are three major types of learning algorithms; supervised, unsupervised, and reinforcement. Currently, deep neural networks are the top not artificial intelligence (AI) and considered to the most influential predictive models, however, discovered for single-neuron activity, large scale cortical tasks and various behavioural aspects. (Storrs & Kriegeskorte, 2019, p. 2-3, 17)
Currently, the educational researcher is implementing deep learning techniques for
improving student modelling. One of the educational researcher utilized Recursive Neural Networks for knowledge tracing problem to get improved predictive capability and to model student learning over the Bayesian knowledge tracing model. It was concluded that deep learning models cannot be neglected in terms of the benefit they offer. With these models, human expert annotations are not needed and the benefit of any student input can be obtained which can be vectorised. (Peterson & Pardos, 2016, 9. p. 322) Classroom Engagement Systems (CESs), a novice technology to enhance student and teacher interaction, permits all students in the classrooms to respond to free-response questions, thereby, all students are engaged at the same time. Pavia et al. introduced Comprehension SEEDING comprising of three goals with the purpose of student engagement; Inquiry generation increased discussion and self-explanation. It was an over- simplified approach. It was concluded that available educational applications are weak in a way or other. Hence, next-generation applications are required that can offer a thorough and finer-grained evaluation of two responses with enhanced feedback. Minimal meaningful Propositions have been introduced lately, performs such finer-grained evaluation of student Responses. Two methods were employed for the detection of entailment relation among MMP reference answer and student response. One method is utilizing pre-trained word embedding as input for DNN (deep neural network), and the second is SVM classifier by employing hand-engineered features. In this, a Modified Version of the dataset was utilized. However, these methods surpassed the frequent class baseline performance by 31% and alternative approaches and past systems by 15 %. (Nielsen & Bulgarov, 2018, p. 5045 -5051) In a study, five instruction types were compared, three types of computer-based tutoring, human tutoring and no tutoring. Out of three types of computer-based tutoring, an answer- based system poses some issues as it requires more reasoning among interactions and has huge granulation of interaction. The second type is a step-based tutoring system, it is designed in a way to allow users to enter the stages which they will cover while solving issues normally, without tutoring. Step based tutoring systems have a comparatively smaller size of granulation than answer based tutoring system. Sub-step based tutoring system can provide scaffolding as well as feedback at such a detail level that is quite finer as compared to steps students will usually enter while resolving a problem. (VanLEHN, 2011, p. 197-198) However, above all step-up tutoring system exhibit direct and indirect advantages. Direct advantage includes scaffolding and feedback that helps students to bridge gaps in the right way aiding them to self-repair or construct their knowledge. An indirect advantage is that students come across many learning opportunities because they continue working on figuring out a solution rather than giving up. Contrarily, Answer based or no tutoring systems will lead students towards guessing and ultimately quitting due to weak scaffolding and feedback. It was concluded that increasing effectiveness from 0.31 (answer based tutoring system) to 0.76 (step-based tutoring system) indicate that currently available Intelligent Tutoring Systems are as effective as the human tutoring system. ( VanLEHN, 2011, p. 203, 213-214)