The Evolving State of AI-Supplemented Computer-Assisted Instruction

The traditional CAI Computer-Assisted Instruction) system depends on the instructors who provide the course material and decide the criteria of evaluation for the students. The advanced versions we have these days have a ‘reactive learning environment' – where students are actively engaged in their online learning programs. The latter systems employ AI (Artificial Intelligence) tools and techniques to take students' interests and performance factors into account and proceed with tutorial dialogues accordingly. Hence, they are known as AICAI (Artificial Intelligence Computer-Assisted Instruction ) System), or simply ICAI for intelligent CAI.

Such AICAIs include a domain expert component (which knows all about the topic that is being taught), a student model that can analyze the responses of the learners and decode their knowledge levels as well as misconceptions, and a component which contains information on appropriate teaching strategies in different scenarios.

Ideally, the course material and user interface of the AICAI systems should be conversational in nature and sound natural.

Some of the leading intelligent online tutoring systems are:

• Scholar, which was designed to teach South American geography. It contains simple facts on the topic which can be used to answer students’ questions, generate factual questions, and evaluate students’ answers.
• Sophie is another example that teaches electronic troubleshooting. It teaches students to develop, test, and debug hypotheses to solve a given problem (or to repair a malfunctioning piece of equipment).

AI Tutors become Smarter with Deep Learning Neural Networks.

A neural network is a machine learning model based on the neurons of the human brains. The ‘machine learning' is the study of giving the computers the autodidactic ability to learn without human intervention. The autodidactic learning processes and outcomes are achieved by allowing the artificial neurons to transfer knowledge to other artificial neurons and to make them smarter through perceptions and past experiences.

Deep Learning allows us to create:

• the pre-trained networks that do not need to be supervised,
• the convolutional neural networks that can be used to analyze visual imagery,
• the recurrent neural networks that can use their memory to process a sequence of inputs such as handwriting recognition or speech recognition, and
• the recursive neural networks which can make structured predictions, which can be quite useful in natural language processing.

Naturally, these deep-learning neural networks can make our AI tutors smarter with time as they continue to self-teach. In computer-assisted instruction systems, virtual tutors and mentors with deep learning capabilities can adapt to the needs of different students and offer them personalized instruction.

Instant Academic Support is Now Possible

Ashok Goel, a professor in the School of Interactive Computing at Georgia Tech, surprised his students when he shared that it was his teaching assistant Jill Watson that had been answering their questions online. Jill was a virtual teaching assistant built on the IBM's Watson platform.

It was observed that the use of virtual teaching assistants increased the student engagement during classroom sessions, and encouraged students to build their own chatbots based on the AI techniques they learned in the class.

Virtual Mentor version 2.0 is another example of an Artificial Intelligence talking entity (or chatbot) that answers entrepreneurship-related questions based on the advice of real mentors. There are AI services like Siri and Cortana that can be of great help to students in the near future, offering customized instruction suitable for their learning style.

Online learning platforms use virtual mentors all the time to provide instant homework help to students (by interpreting the question through natural language processing methods and processing them to extract relevant information from the resources they have access to.)