In 2016, an AI powered teaching assistant, Jill Watson, was introduced by a professor of computer science and human centered computing at Georgia Institute of Technology. Built on IBM’s Watson platform, Jill was introduced in the professor’s online class, Knowledge-based Artificial Intelligence, and answered students’ questions alongside human teaching assistants. When Jill’s identity was eventually revealed at the end of the semester, none of his students had suspected that they had been communicating with an AI all semester. Today, Jill Watson has been run in about 16 classes and is used to offload mundane and routine work.
Over the past decade, artificial intelligence has been embraced by many industries as a way to streamline and automate mundane tasks and operations. For education, this could not have come at a better time. The aftermath of the Covid-19 pandemic has left schools and education in general in a worse state than ever before. According to a recent survey, teachers are working an average of 54 hours per week and teacher dissatisfaction is at an all time high. In addition to their regular duties, teachers are now having to balance pivoting to online learning, covering other classes, and readjusting to new academic standards, all in the midst of a staffing shortage.
The addition of Language AI to classes in the form of voice assistants, teaching assistants, and other models is becoming more common as a way of lessening the workload of teachers and educators. And teaching assistants, whether robot or human, are meant to provide support to both the teaching staff and children in the classroom. AI powered assistants support teachers by introducing advanced learning technologies like virtual laboratories, and workshops as well as routine work like reading books out loud, serving as a tutor, and serving as a never ending source of information for students. Children with disabilities are also supported in moving around and playing.
A Survey of Language AI Tech Stack In the Classroom
AI teaching assistants are a specialized kind of intelligent assistant, enabled by the core technologies of all such systems: voice recognition, conversational AI to handle dialog flow, and, importantly, especially in the context of education: information retrieval.
In the case of the Jill Watson example cited above, the main goal of the first version was to automatically answer questions that were frequently asked relating to the course syllabus information. To build this version, a collection of question and answer pairs were organized into different question categories in a preliminary grouping. A new ontology of the class syllabi was then developed for use by the Jill Watson system. Since the most recent version has greater question answering capabilities in, a hybrid classification approach is now used to separate and answer questions using statistical machine learning methods to categorize the incoming questions and label their underlying intents. After this, a knowledge based classifier parses the question and structures an appropriate response from an underlying knowledge base of relevant course related information.
In order to build more Jill Watson systems quickly, an interactive machine teaching environment called Agent Smith was intended with the aim of making it easy for experts to generate custom Jill Watson systems. The Agent Smith system works by operating on the inputs of a Knowledge Base that contains concepts and relations that form the knowledge that a Jill Watson system would explain, and a set of question templates that capture the structural form of questions asked by potential users in the domain. These knowledge bases are then combined by Agent Smith as part of a generation process to create large sets of questions and their associated intents. These pairs are then used to train the machine learning classifier in a Jill Watson system.
The work done at Georgia Institute of Technology on the Jill Watson project is, obviously, just one of the many projects at the intersection of AI and education, and it highlights the unique challenges faced by technologists working in this area.
Making Education Accessible to Every Body
One of the best use cases of artificial intelligence in education is in the support of neurodivergent children. For many students on the autism spectrum, the major problem with the current educational system is the absence of attention and motivation. The introduction of graphic elements, virtual assistants, and the use of VR has had more success than traditional methods of teaching. This has also extended to using artificial intelligence to teach communication skills, sensitivity to physical contact, and to provide support in decision making for children with autism. A similar level of success has been recorded with students with learning disabilities and difficulties, communication disorders, developmental disabilities, and other special educational needs.
Outside of the specific support for children with disabilities, there is something to be said about the access that AI-generated TA’s can offer. In a world that had to quickly adapt to online models, these TA’s can help even the playing field and ensure a fair and inclusive access to education while providing access to high quality learning and making the teacher’s duties easier. According to studies on how the introduction of artificial intelligence affects students learning behavior, high school students demonstrated higher motivations in the learning process compared with traditional learning methods and in universities, students showed significantly improved understanding during classes.
What’s Next at the Intersection of AI and Education?
The success of artificial intelligence in education does not take away from the concerns that are already being raised about the potential impact that this path might have on both students and teachers. One of them is the likelihood of reproducing inequality in classrooms. Because these models are trained on data that may be biased, there is a possibility that AI-powered Teaching Assistants may reinforce stereotypes and inequalities already present in society. This is particularly harmful for students who might not yet have fully formed belief systems making them susceptible to prejudices and discriminatory behavior. Apart from the risk of perpetuating inequities, there are also some apprehensions about the social consequences of the use of artificial intelligence in education. While there is little chance that AI-powered assistants will be able to fully take over the job of teachers in the near future, there are still concerns about the effects they might have on children as they interact with these technologies. Some of these include whether children would eventually be able to differentiate between artificial intelligence and humans, and what that could mean for their wellbeing.
We now know that children can distinguish between AI and other humans and act accordingly. They engage in more physical contact with humans, feel a greater sense of friendship, and show more fairness towards humans. We also know that AI-powered assistants are able to adequately support the learning and development of children with the potential to do much more. As the technology improves and datasets are diversified and finetuned, more and more AI-powered assistants will appear in classrooms, and this might not necessarily be bad.