As we write this column, it has been several weeks since the release of Chat Generative Pre-trained T (aka “ChatGPT”), by OpenAI (https://openai.com/blog/chatgpt/). ChatGPT is an artificial intelligence (AI) chatbot that can be used for a variety of purposes, such as to generate various genres of writing, including essays and poems, as well as code and titles for one’s writings. Once we had a draft of this article, we asked ChatGPT to suggest some titles for this article and were surprised at the titles it generated. We decided to stick with our own.
Professor & Director, Educational Technology Leadership and Human-Technology Collaboration PHD Programs, George Washington University, 2134 G ST, NW, Washington, DC. Telephone: 202-994-2263.
Professor & Director, Educational Technology Leadership and Human-Technology Collaboration PHD Programs, George Washington University, 2134 G ST, NW, Washington, DC. Telephone: 202-994-2263.
Professor of Educational Technology, George Washington University, 2134 G ST, NW, Washington, DC 20052. Telephone: (202) 994-1884.
Professor of Educational Technology, George Washington University, 2134 G ST, NW, Washington, DC 20052. Telephone: (202) 994-1884.
Although there is a lot of potential, many educators have expressed concerns and are now wondering if “the college essay is dead?” (Marche, 2022, para. 1) and “Why do your homework when a chatbot can do it for you?” (Bowman, 2022, para. 1). We would like to note that throughout technology’s history, there have been concerns regarding its benefits and harms. Orben (2020) argues that our views about technology are largely influenced by “technological determinism” (p. 1146), which is
the idea (a) that the technologies used by a society form basic and fundamental conditions that affect all areas of existence and (b) that when such technologies are innovated, these developments are the single most important driver of changes in said society. (Leonardi, 2012, as cited in Orben, 2020, p. 1146)
Orben also emphasized that this perspective “plays a crucial role in initial reactions to new technologies, and moral panics” (Orben, 2020, p. 1146). Similarly, Morozov (2013) asserts that technosolutionist views, the idea that technology is the solution, even for complex social phenomena like education, have predominantly shaped our understanding of educational technology. However, before we panic, it is important to keep in mind that tools such as ChatGPT are emerging technologies (Veletsianos, 2010, 2016), and as such, it is hard to fully comprehend their implications as they are currently evolving. Moreover, even the definition of AI is unclear, as noted in a prior Ends and Means column by Henderson and Milman (2022). Before the dust settles (if it ever does!), let us delve a little more into what happened and what the implications of this emerging tool means for educators.
Professor of Educational Technology, George Washington University, 2134 G ST, NW, Washington, DC 20052. Telephone: 202-994-9295.
Professor of Educational Technology, George Washington University, 2134 G ST, NW, Washington, DC 20052. Telephone: 202-994-9295.
What Happened?
The first large language models (LLM) started to appear soon after Google released a new tool (called BERT) in 2017. Since then, popular LLMs (such as GPT-3, Megatron-Turing, Pathways, Galactica, and others) have become synonymous for many with “artificial intelligence” or AI. Given that there is no standard accepted definition of AI, it is reasonable that these tools have captured attention under that umbrella since they are allowing computers, for the first time, to generate natural language responses to natural language questions. In other words, they can appear as “intelligent” at times. They can compose poems or essays, write computer code, and achieve many other tasks with remarkable dexterity and impressive results.
That said, they are generative mathematical models (Shanahan, 2022) developed based on very large amounts of content found on the internet, and therefore are routinely inaccurate. Likewise, LLMs do not know the meaning of the words or phrases they string together. They simply know which words or phrases have been stringed together by others in the past. This can result in new passages of writing that sometimes make little sense and other times generate quite articulate and well-grounded, often leading to confidently inaccurate outputs. This should not take away from the impressive technical accomplishments of these innovations, but it can help us understand, admittedly at a very basic level, what these technologies are doing, what their potential strengths are, and what their weaknesses might be.
In December 2022, OpenAI released a public preview of a new model, ChatGPT, whose superior performance over previous models has accelerated the importance of building awareness of these technologies and the implications of their use. While ChatGPT is a major advancement in text-based AI, in the fall of 2022, OpenAI, StabilityAI, and several other groups released image generating models that likewise achieved before-unforeseen visual results.
Implications
The implications of these advancements, especially of LLMs, are just starting to be seen in higher education. In the coming years, we anticipate the implications will likely spread to many aspects of higher education careers (i.e., teaching, research, and service). In teaching, for example, there will be negative uses such as plagiarism (see Figure 1 for some initial thoughts) and also positive opportunities where AI can be leveraged to facilitate greater depth of knowledge. In research, AI may be used negatively to flood journal editors with misinformation article submissions and positively to write complex computer code for analyzing data. Lastly, in service, the examples could include negative uses such as fake letters of recommendation and the positive use to provide automated counseling or academic advising to students.
The diagram presents a lightbulb silhouette with its filament section filled with interconnected gears of various sizes. Six different statements are arranged around the lightbulb, three on the left and three on the right. The statements are each marked by a checkmark inside a circle. The statements on the left are as follows. Using Artificial Intelligence for school work does not automatically equate to misconduct is at the top. Artificial intelligence can be used ethically for teaching, learning, and assessment is in the middle. Trying to ban the use of artificial intelligence in school is not only futile, it is irresponsible is at the bottom. The statements on the right are as follows. Human imagination and creativity are not threatened by artificial intelligence is at the top. Assessments must be fit for purpose and should align with the learning outcomes is in the middle. Artificial intelligence is not going anywhere. We must learn to work with new technology, not against it is at the bottom.Sarah’s thoughts: artificial intelligence and academic integrity (Eaton, n.d.)
The diagram presents a lightbulb silhouette with its filament section filled with interconnected gears of various sizes. Six different statements are arranged around the lightbulb, three on the left and three on the right. The statements are each marked by a checkmark inside a circle. The statements on the left are as follows. Using Artificial Intelligence for school work does not automatically equate to misconduct is at the top. Artificial intelligence can be used ethically for teaching, learning, and assessment is in the middle. Trying to ban the use of artificial intelligence in school is not only futile, it is irresponsible is at the bottom. The statements on the right are as follows. Human imagination and creativity are not threatened by artificial intelligence is at the top. Assessments must be fit for purpose and should align with the learning outcomes is in the middle. Artificial intelligence is not going anywhere. We must learn to work with new technology, not against it is at the bottom.Sarah’s thoughts: artificial intelligence and academic integrity (Eaton, n.d.)
How students, faculty, and administrators decide to use LLMs (and other AI-based tools) in the future is far from clear. There are likely thousands of applications that we could not imagine today, and each application will require us to consider the ethical implications along with the potential positive gains.
In Figure 1, Eaton (n.d.) provides thoughts about the intersection of AI tools and academic integrity.
When asked, according to chatGPT (n.d.)
It is possible that AI could make it easier for students to cheat on essay assignments. For example, AI-powered essay generators or plagiarism detection tools could be used by students to produce or identify copied content. However, it is unlikely that AI will completely eliminate the ability for professors to assign essay questions, as effective teaching and assessment still requires human expertise and judgment.
First Do No Harm (Primum Non Nocere)
We teach in a fully online graduate masters educational technology program and believe it is our duty to explore the use of AI technologies, including its potential for harm—and aim to proceed with caution. Stephenson and Harvey (2022) also contend that “it is important that we adopt a critical mindset in response to the triumphalist narrative of AI’s remaking of higher education learning and teaching practice (p. 122).The Latin phrase, “primum non nocere” (first do no harm) is a good guiding principle—but how does one realize this in practice?
We believe exploring these technologies with our students and discussing their ideas and thoughts will help us also understand their implications. However, we also need to be transparent that its characteristics and outcomes could change, as the following statement, which we plan to include in our comprehensive exams, explains,
Until additional guidance on how to cite, reference, and ethically collaborate with AI-based systems (including, but not limited to, language models) is sufficiently debated and guidelines are provided by American Psychological Association (APA; for whom the discipline of education follows their style guide), you should cite and reference any use of resources (e.g., quotes, paraphrases) generated by AI-based systems as you would any other online resource. Doing otherwise will be considered plagiarism within the ETL program.
Note: Generative AI-based systems (i.e., systems that can create content) are new and it is your responsibility as a scholar/professional to monitor best practices and official guidelines as they are developed in the coming months and years.
Recommendations
AI (e.g., LLM) tools are/can/will be used in a number of ways, some that make us concerned and some that offer exciting opportunities to accelerate learning. As a result, there is not a one-size-fits-all approach to how we engage with these technologies (and the technologies that will follow).
Here are a few initial recommendations:
Test drive the tools for yourself. You can best understand the potentials (positive and negative) if you first understand the capabilities of the tools.
Take time to reflect on the ethical implications of using AI tools. Discuss these reflections with colleagues and students.
Consider how AI tools might impact educational policies currently in place or ones that should be developed.
Provide ethics and plagiarism resources to your students to ensure that they are knowledgeable of the context for the expectations you set.
Do not panic and avoid “knee jerk reactions.”
Be careful not to anthropomorphize these technologies (e.g., avoid “chatGPT knows …”, or “the AI thinks …”).
Remain flexible since your policies and practices will likely have to change as AI tools evolve.
Look for opportunities. Though changes (to your course assignments, for instance) will require work, there are going to be opportunities to use AI technologies to achieve teaching/research goals that you could not achieve before.
