Abstract
Recent advancements in artificial intelligence (AI) have brought relatively esoteric topics, such as language models and diffusion, into the mainstream conversation. Particularly because of ChatGPT, a human preference-aligned chat bot from Microsoft-backed OpenAI, most people have now had some interactions with generative AI models. This tool, suddenly released to the world, opens up a plethora of possibilities and risks. As a response, both the United States and European Union have begun working on AI regulations, a marker of the technology's potential impact.
However, the largest impact of new AI technologies may be felt in education. Many of the goals of educational researchers, such as providing personalized and high-quality education to students around the world, are now much closer to being tangible. A large body of current research is investigating automatic generation of educational content, such as test questions and virtual tutor dialogs, often using ChatGPT or similar models. Some of these systems are even being used by real students. For example, Khan Academy's new chat bot, Khanmigo, is powered by a recent version of ChatGPT, and WolframAlpha also has a ChatGPT plugin to explain solutions in eloquent language. These tools, along with the many more that will be released in the coming years, promise the ability to improve the quality of self-learning, make teachers’ jobs easier, and make education more accessible. However, for all the potential benefits, the risks could cause students more harm than good. These AI models have been known to make factual errors, often referred to as “hallucinations,” that could result in students receiving false or misleading information. They have also been known to exhibit cultural and gender biases, as well as occasionally generating explicit or harmful content. Beyond such current limitations, there are larger concerns, such as students using the models to cheat on homework without detection, and the models eventually replacing human teaching assistants, tutors, and even classroom instructors. The immediate appearance of these technologies calls for an immediate response. If they are used correctly, education could benefit greatly, improving student outcomes and providing quality education to those who wouldn’t have it otherwise. However, if they are unregulated and the risks are ignored, both students and teachers could suffer greatly, and education as we know it could be existentially threatened. This special issue of JETS presents a glimpse into the future of AI-augmented education, along with an important measure of public perspective on the use of generative AI, particularly ChatGPT, in educational settings.
This special issue begins with a paper in which the authors investigate ChatGPT's ability to automate various tasks performed by educators, most importantly generating assessment problems. The authors find that ChatGPT can effectively generate a breadth of on-topic questions in a variety of styles, although somewhat below the quality of expert human instructors.
In the second paper, the authors have students use ChatGPT in a collaborative way to write course essays. The results reveal the potential uses, shortcomings, and risks of using ChatGPT to aid students in their assignments.
In the third paper, the authors investigate the sentiment of university students towards AI systems. The authors uncover under which population demographics sentiments are more similar or different. In the next paper, the authors investigate a variety of beliefs and perspectives towards ChatGPT from university students and faculty. They uncover that people are concerned about ChatGPT being used for cheating, but are also hopeful for its potential benefit to higher education.
The fifth paper investigates broad research trends in educational AI through a bibliometric study. They present the exponential increase of publications in recent years, as well as changing global trends and the need for increased cooperation between researchers.
The final paper is not actually about AI, although it is a topic that AI will likely be applied to in the near future. It describes using two different computer-assisted reading programs over the summer with third graders who are at-risk for reading failure. The results are encouraging, with conclusions that could readily inform an automated tutoring system.