L'apprentissage à l'heure de l'intelligence artificielle

ChatGPT: An Ongoing and Unprecedented Disruption for Education and Training?

ChatGPT-Egg-Chicken_bandeau.jpeg

The advent of Artificial Intelligence (AI) has enabled numerous innovations in different fields, and education is no exception. Educational chatbots, especially AI-powered chatbots such as ChatGPT, are revolutionising the way we learn and interact with information. Since its introduction to the public, ChatGPT has become a world-wide phenomenon and has generated both awe and fear with its sheer computing power. The disruption ChatGPT is causing within the educational sector recalls the lukewarm reactions of many seasoned education practitioners and academics toward the arrival of the internet and the advent of Wikipedia. At the time, many perceived the near-immediate access to information as threatening the very essence of good quality, human-conducted research. Today, similar questions arise: should the use of ChatGPT (and other AI-powered apps) be regulated? If so, how, by whom, and with what underlying pedagogical vision? These questions aside, we are living through an unprecedented disruptive revolution that could lead (or force) us to reinvent how education and training is conceptualised and operationalised.[1] That said, where do we, as educational practitioners, stand? Should we fear the ChatGPT wave or surf on it? How can we anticipate the future to adapt our current practices?


ChatGPT and education: a relationship in the making?
 

ChatGPT is a natural language processing (NLP) system based on artificial neural networks. It was developed by OpenAI, an artificial intelligence research organisation, and is based on the GPT (Generative Pre-trained Transformer) language model. This model is trained on large amounts of text data, which allows it to understand and generate natural language. The version that took the world by storm in late November 2022 was built on top of OpenAI's GPT-3 family of large language models. Its successor, GPT-4, which was released on 14 March 2023, was announced to be “far more powerful and capable than GPT-3"[2]. Among other things, the new version can process about eight times as many words as the previous version (up to 25,000 words) and can respond to images (for instance adding relevant captions)[3]. OpenAI also indicates that GPT-4 provides the basis for more advanced reasoning capabilities and a safer environment for users, claiming that “GPT-4 is 82% less likely to respond to requests for disallowed content and 40% more likely to produce factual responses” than previous models.[4]

ChatGPT may be used in educational contexts in a variety of ways.[5] Here is a non-exhaustive list of examples in which ChatGPT can be incorporated within pedagogical learning designs:

  • Research assistance: ChatGPT can be used to help students find information relevant to their research. Using natural language processing algorithms, ChatGPT can analyse students' search queries and provide relevant results.
  • Content generation: ChatGPT can be used to generate educational content, such as exercises, tests, quizzes, assignments and reports. Using prior learning data, ChatGPT can generate personalised content for each student, which can help students better understand concepts.
  • Text correction: ChatGPT can also be used for text correction. Using a natural language processing algorithm, ChatGPT can identify grammatical and spelling errors in a text and provide suggestions for correcting them.
  • Educational chatbots: ChatGPT can be used to create educational chatbots that can answer students' questions. These chatbots can be used to provide information on a specific topic, answer students' questions or even provide advice and guidance on more general issues, such as how to study better.

In the end, ChatGPT is already being used in educational contexts whether we like it or not. Moreover, ChatGPT is but one out of the many similar AI-based tools that already exist and that aim to automate or accelerate certain existing tasks. It is essential that educational institutions invest time and resources to stay on top of IA-related educational tools to best accompany their instructors in choosing which tools should integrated in pedagogical designs and how.


The impact of ChatGPT on education and training: a few benefits with important shortcomings
 

Some of the benefits of using ChatGPT in education include: 

  • Accessibility: ChatGPT can help make education more accessible by allowing students to access information anytime, anywhere. Educational chatbots can be used at home, at school or even outside the school context to provide continuing education on the go.
  • Personalised learning: Using prior learning data, ChatGPT can generate personalised content for each student based on their needs and preferences, which can help students grasp concepts at their own rhythm.
  • Improved learning efficiency: ChatGPT can help improve learning efficiency by allowing students to ask questions and receive immediate answers. Educational chatbots can provide additional explanations to help students understand concepts more deeply. 
  • Reducing instructor stress: ChatGPT can help reduce instructor stress by answering students' frequently asked questions. Instructors can thereby focus on the most important tasks such as course design, personal coaching and assessment, while ChatGPT can answer common questions.
  • Cost savings: This one is not uncontroversial. ChatGPT may contribute to reduce the cost of education. Educational chatbots can replace some additional tutors, which can help institutions mitigate teaching costs and make education more accessible (provided there are no technological gaps or barriers to start with). 

While ChatGPT may constitute a valuable tool in education and training, it still has some shortcomings that should be considered, including:

  • As good as the data it was trained on: One of the biggest limitations of ChatGPT is that it is a machine learning algorithm that operates based on pre-defined data and algorithms. This means that it may not always be able to provide accurate or complete responses to complex questions or situations that require human judgment and decision-making. Additionally, ChatGPT's responses may be limited by the data it has been trained on, which can lead to biased or incomplete answers. For example, the version that used to run on GPT-3 until mid-March 2023 came with a disclaimer that ChatGPT was trained on data dating from 2021 and would therefore have limited knowledge of post-2021 events. 
  • Inaccurate or completelyprovide fabricated citations: Some tests have revealed that ChatGPT is not reliable when it comes to generating adequate citations and bibliographic resources. Worse, users have raised the issue that ChatGPT tends to generate completely fabricated citations (which nonetheless look very real). This issue has been referred to “hallucination” and might be significantly improved remedied in the future.[6] Still, and rather disappointingly, OpenAI continues to list the issue of “hallucination” among the limitations that have yet to be solved, even with the recently released GPT-4.[7]
  • Motivation, creativity, cheating and plagiarism concerns: There has been widespread worry among educational providers that ChatGPT (and other similar tools) would rob students of their motivation, creativity and critical thinking; as well as fearing that it would create a culture of widespread cheating and plagiarism.[8] Interestingly, the advent of ChatGPT is affecting what some perceived as a new paradigm of relative trust between students and teachers, which had been established as a result of the COVID pandemic. In effect, while the forced move to full distance learning had led to widespread acceptance about the need for, and the interest of distance learning (including distance or e-assessment), the advent of ChatGPT is already calling this paradigm shift into question.
  • Limited personalised learning experience: ChatGPT may not always be able to provide a personalised learning experience that meets the individual needs and learning styles of every student. While ChatGPT can provide personalised feedback based on a student's inputs and performance, it may not always be able to adapt to a student's unique needs or preferences in the same way a human teacher can. Additionally, some students may struggle to interact with ChatGPT, either because they do not feel comfortable using technology or because they prefer a more traditional learning environment. 
  • May lead to further student isolation: While ChatGPT can provide feedback and guidance on academic performance, it may not always be able to provide the emotional support and encouragement that students need to overcome challenges or stay motivated. Additionally, ChatGPT may not be able to provide the same level of social interaction and community-building that students can experience in a traditional classroom setting. 
  • Ethical and privacy concerns: Finally, it is important to recognise that the use of ChatGPT in education and training can raise ethical concerns around privacy and data protection. As an AI-powered chatbot, ChatGPT collects and stores data on user interactions and performance, which can raise questions about who has access to this data and how it is being used. Additionally, the use of ChatGPT can raise concerns around the ownership and control of educational content, as well as the potential for corporate interests to shape the direction and focus of educational programs. 


Recommendations on using ChatGPT in educational contexts
 

It is crucial for institutions to stay on top of trends to understand the limitations of ChatGPT and similar tools. The University of Geneva’s Centre for Continuing and Distance Education conducts a permanent watch[9] on how these tools can be best used in conjunction with other educational resources. As of March/April 2023, our recommendations on using ChatGPT in education contexts include:

  1. First and foremost, we recognize that ChatGPT and other similar AI-powered tools are already here and here to stay. We therefore recommend that education practitioners work towards trying to integrate and work alongside those tools rather than attempting to ban them. 
  2. Second, we believe that while ChatGPT can provide personalised and accessible learning, it cannot replace the value of human interaction and personal attention that a teacher can provide. This reality should therefore by embodied by instructors, who should be confident in their added value in terms of student support and coaching in the classroom.
  3. Third, educational institutions and practitioners should be aware of the possible ethical concerns around privacy, data protection, and corporate influence that come with AI-based tools, as well as major shortcomings such as issues related to inaccurate or fabricated bibliographic citations.
  4. Fourth, to address the key concern of cheating and plagiarism, course pedagogical designs (including assessment means) should focus on skill development, which will be crucial as they enter the world of work, post-graduation. This includes hard skills such as data analysis, coding, project management, research or graphic design; but also soft skills, such as communication & public speaking, teamwork, empathy, leadership and emotional agility.
  5. Fifth, and operationally, this involves conceptualizing assignments that are meaningful in that they strive to lead to skill development. This can be done by incorporating:
    1. experiential learning, such as working on case-studies, taking part in role-plays, interacting with a given field’s professionals, undertaking internships within an institution, etc.
    2. multimodal evaluations, consisting ofsupplementing writing assignments with other productions such as podcasts, videos, or visuals such as mind maps and infographics.[10] For instance, a classic take-home assessment initially consisting of answering an essay question might involve asking the student to also provide a mind-map demonstrating his/her research process and/or a short audio commentary to introduce the paper and its key take-aways.

Ultimately, the AI revolution is here to stay and will likely continue to surprise us in ways we cannot even imagine. This realisation has generated fears across the board, leading key figures in the tech world to urge a halt in AI training, warning of risks to humanity.[11] In fact, it is now opening interesting perspectives in terms of human-AI collaboration and future jobs such as “AI whisperer” or “AI prompt engineer”[12] that, just a few years ago, would have seemed taken out an Isaac Asimov novel.[13] By integrating ChatGPT and other similar tools rather than trying to ban them, educational institutions will be more able to optimise their use, mitigate their shortcomings and hopefully reap the benefits of this unprecedented disruptive technological revolution.

This article was first published in the NewSpecial Magazine (April 2023 edition).