Four lessons from ChatGPT: Challenges and opportunities for educators

Contents

1. ChatGPT in the news
2. Lesson 1: ChatGPT does present a challenge to maintaining academic integrity, but this is neither new nor unique
3. Lesson 2: ChatGPT is not just a chatbot but a useful tool for educators whose potential is yet to be fully explored
4. Lesson 3: ChatGPT is starting to be used as a tool for learning and we are still waiting to see how effective it may be
5. Lesson 4: ChatGPT is only one of many tools built on similar underlying technologies in a rapidly changing field. It is too early to fully appreciate what the landscape will look like even one year from now
6. Time to share and learn

ChatGPT in the news

Image generated by text-to-image AI tool Midjourney

The release of ChatGPT by OpenAI in late 2022 has drawn much attention to the increasing power of systems built using the latest developments of machine learning.

While ChatGPT is already the fourth iteration of the GPT technology (the first being announced in 2018), it has captured the attention of the media as well as the education community due to the quality of the text it can generate. This is magnified by its chat interface which makes it much easier to interact with the tool without any technical knowledge.

The last two months have seen a flurry of reactions both optimistic and concerned, as well as explorations of various highly practical use cases. People from diverse backgrounds have pushed the limits and discovered many unexpected uses. The immediate concern (dating back to the release of GPT-2 in 2019) is the use of this technology for cheating. And in fact, we are hearing those concerns from colleagues as well as people outside academia. However, when we survey the published reflections on ChatGPT we see a more sober and even optimistic picture emerge.

In this piece we have tried to summarise the discussion that has taken place over the last two months across numerous blog posts, news articles and pre-prints. We have synthesised the discussion to date into four lessons covering both the impact on summative assessment and other aspects of academic work.

What is clear, even at this early stage, is that ChatGPT offers an initial glimpse at the potential of this type of technology. It will take some time before we have a more concrete picture of the actual impact on teaching and assessment as new products, approaches and ways of working emerge around it.

Lesson 1: ChatGPT does present a challenge to maintaining academic integrity, but this is neither new nor unique

The first question many people have asked is: ‘Can this tool be used to write a student’s essay for them?’. Even a brief engagement with ChatGPT will reveal the answer to be ‘yes’ but with many qualifications. The system is designed to produce cohesive and plausible text, but it is not factual in the sense that we expect from a database search. It will intersperse accurate facts with fabrications that have come to be known as ‘hallucinations’.

Nevertheless, it is not difficult to prompt the system to produce work that will not be easily identified as machine generated and will probably for the time being escape detection by plagiarism checkers such as Turnitin. As such, it is a possible tool for cheating.

However, as the reactions from the educator community indicate, this merely underscores the challenges of maintaining academic integrity rather than completely changing the situation.

In a summary of his recent opinion piece in Times Higher Education, Danny Oppenheimer, Professor of Psychology and Decision Science at Carnegie Melon, claims that much less is new than it may have appeared.

Cotton et al (2023) in a preprint of a paper co-written with the use of ChatGPT agree and provide support for this position.

Cotton at al (2023) offer hope by suggesting that machine generated text may ultimately be easier to detect than papers purchased from professional contractors even if all the tools for detection have not yet been fully developed.

A consensus seems to be emerging that educators can use the new challenge of ChatGPT and other AI text-generation tools as an opportunity to (re)design assessments that require students to demonstrate higher-order learning outcomes that draw on learning across the curriculum in order to achieve higher outcomes. The impetus is on designing assessments that provide robust academic challenge for students across a range of tasks, conditions and timings to support a flexible and inclusive suite of assessments. Within weeks of ChatGPT’s release, Ryan Watkins, Professor of Educational Technology Leadership, and Human-Technology Collaboration, at George Washington University, shared a set of suggestions on how he is planning to modify his syllabus and assessment now that this tool is available. His approach combined being more explicit about academic integrity with rethinking how he would diversify the ways in which students can present what they have learned.

Cynthia Alby, lecturer at Georgia College, also offers suggestions in a similar vein in her set of Short-term solutions, but stresses the importance of seeing this new environment as an opportunity as well as a challenge:

Such sentiment does not appear to be limited to the English-speaking world. This recent article in Le Monde quotes several French lecturers excited by the opportunities presented by this new tool rather than focusing solely on the potential threat to academic integrity.

Lecturer from NYU Abu Dhabi, Nancy Gleason urges educators in Times Higher Education not to overreact but rather to redouble their efforts at diversifying their assessment methods because tools such as ChatGPT will simply become part of the toolset available outside of education.

It is also important to note that submission of student writing for assessment is not the only area where the impact of ChatGPT and other similar tools is being felt. As this article in ‘Nature’ describes, questions of authorship have already arisen in legitimate submissions to academic journals.

In conclusion, there seems to be general agreement with the advice from JISC published on 10 January 2023:

Lesson 2: ChatGPT is not just a chatbot but a useful tool for educators whose potential is yet to be fully explored

On her blog Assessment in Higher Education, Rachel Forsyth, author of ‘Confident Assessment in Higher Education’, asks not only What should educators do next? to protect academic integrity, but also how they can use this is a tool for their own practice and how this will change the view of academic integrity. After testing whether ChatGPT can generate a sensible first draft of a paper, she suggests:

She then asks: ‘I still wrote the abstracts, and did the research which I’m going to write about. Am I cheating?’. The variety of uses to which academics may put a tool such as this to assist in their writing is yet to be determined, but it is hard to imagine that it will remain unused.

However, to limit the discussion to how ChatGPT and other similar tools can be used to generate essays would be to ignore its other possibilities that can be used today. We are already hearing informally from colleagues that they have found creative uses for this tool that do not raise questions of integrity.

Some very practical uses have been shared by educators in a recent EducationWeek piece: 19 Ways to Use ChatGPT in Your Classroom (Opinion). Although these examples come from secondary classrooms, they can be easily modified for many higher education contexts.

ChatGPT can not only generate text, but also snippets of computer code in various languages, tables, lists, and even Excel formulas. All of these require editing and checking, but they can be a great time-saving device. This has led the economist Arnold Kling to suggest:  

However, he warns that this will not be as natural as having a chat with a friend.

In fact, a new term has emerged in the last year since other similar tools have appeared called ‘prompt engineering’ that describes the skill of formulating prompts in such a way that it produces useful results.

Given that this type of tool is brand new, and its possibilities not fully explored, it is not yet clear what all those skills are. What is certain is that those who will take the time to learn about the underlying principles of this tool will be best placed to take advantage of it.

Lesson 3: ChatGPT is starting to be used as a tool for learning and we are still waiting to see how effective it may be

In all the concern about students using ChatGPT or other AI tools to cheat, it may be easy to forget that they can also use it to support their learning. It is too early to say with any confidence what the overall impact of this will be, but not too early to imagine real uses.

Given that ChatGPT can be used for almost any language, language learning is one such use case. Within a week of ChatGPT’s release, translator Tom Gally started imagining the different ways language students can use it to help them learn in ChatGPT and Language Education.

These include generating their own vocabulary quizzes, suggesting comprehension questions for a text in another language, comparing translations, etc. The results produced by ChatGPT are far from perfect but already show great promise.

This detailed description of a 5-day course at Johns Hopkins University by Professor of Strategic Study Thomas Rid outlines the uses of ChatGPT along with student testimonials:

Rid describes both the strengths and limitations of using this tool, but concludes in a very optimistic tone urging educators to explore the question of ‘how can the most creative, the most ambitious, and the most brilliant students achieve even better results faster?’.

This underscores the need echoed in many of the resources linked throughout this piece that we need to not only focus on preventing students from using this tool in inappropriate contexts, but also support them in using it in ways that we believe would be beneficial to their learning. At the institutional level, Oxford's new Digital Education Strategy for 2023-27 articulates a similar commitment:

Lesson 4: ChatGPT is only one of many tools built on similar underlying technologies in a rapidly changing field. It is too early to fully appreciate what the landscape will look like even one year from now

Despite its high profile, ChatGPT is neither completely new, nor the last development in the use of machine learning models (often referred to as Large Language Models or more generally foundation models). ChatGPT’s direct predecessor GPT-3 was released in 2021, and writing coach at Copenhagen Business School Thomas Basbøll was soon able to demonstrate its power to produce credible essays with good prompts and later discuss what challenge it may pose to the ‘integrity of the essay’.

GPT-3 and its competitors have started making an impact in the product space. For example, Notion, the popular note-taking tool recently released an alpha version of an AI assistant that helps with brainstorming, summarising or even drafting. A number of copy-writing services have emerged even earlier (often focused on marketing and website development) based on the much less powerful predecessors. Some others such as Scholarcy, Consensus or Audemic have also started making inroads into the academic space improving discovery and summarisation. Both established and emerging services have been reviewed in this discussion of the impact of AI on Scholarly Publishing.

Nor is OpenAI (with its heavy investment from Microsoft), the only company creating similar tools. ChatGPT is developed using relatively well-understood machine learning techniques that are accessible to those with relevant expertise, vast computational resources, and access to data.

Despite its undisputed power and utility, ChatGPT is not a product but a ‘research preview’ released by OpenAI to ‘to get users’ feedback and learn about its strengths and weaknesses’. The costs of running this ‘public’ service are estimated to be significant and OpenAI has already announced plans for a commercial API.

Finally, a very similar technology has already been productised in image generation, computer code, voice generation, etc. And we should also not forget that similar developments are slowly being incorporated into existing products by Microsoft and Google. Latest developments in dictation, text to speech or text prediction available in Microsoft products take advantage of earlier generations of similar technology and can be expected to continue to improve.

It is not clear yet what exact shape this new ecosystem of tools will take, and neither is it certain what future improvements we can expect. But it is without doubt that this technology has real uses and will continue to increase its presence in both our professional and personal lives.

Time to share and learn

We are very much in the early stages of getting to grips with what the new technologies represented by ChatGPT will mean for our work or even identity as academics and educators.

We would very much appreciate finding out more about how ChatGPT and similar tools are being used by both academics and students. If you would like to share a particular use case of ChatGPT, please let us know via this short form.

More information about some of the use cases and tools we have explored is available on Edutools.fyi. You can also find an up-to-date list of tools based on similar AI technologies along with a stream of news articles on FutureTools.io.

Image credit 

The cover image of this article was generated by Midjourney, a text-to-image AI-based tool and is licensed under Creative Commons Non-Commercial License.