[ad_1]
OpenAI last year released ChatGPT-4, the latest version of its powerful artificial intelligence (AI) text generator. The tool can produce believable human-like answers to almost any question asked by users. It can write limericks, tell jokes and outline novels. It can produce believable answers to almost any question a high school teacher or university lecturer might ask students.
Previous iterations often generated text full of strange and obvious mistakes. The responses generated by ChatGPT are able to pass tests in a number of subjects. It is tempting to think that we will always be able to distinguish between the work of AI and the work of humans, especially when it comes to typical human tasks such as creative writing, careful reasoning, and making new connections between different types of information.
Unfortunately, this optimism is misplaced. The prose and poetry produced by AI can be beautiful. And with some clever innuendo, AI tools can generate acceptable argumentative essays in philosophy and bioethics. This raises a serious concern for universities that students will be able to pass assessments without having written a single word – or without necessarily understanding the material they are to be tested on. This is not just a future concern; students have already begun submitting work produced by AI.
Some institutions consider the use of AI text generators to be cheating. Several schools and universities have banned the use of ChatGPT, but such bans would be difficult to enforce. Compared to traditional forms of plagiarism, students’ use of AI-generated text is harder to detect—and even harder to prove, partly because the new ChatGPT generates new responses every time a user inputs the same prompt.
For its part, OpenAI is developing tools for AI-assisted fraud detection – though such tools are prone to making mistakes, and can currently be prevented by asking ChatGPT to write in a style its detector can’t pick up.
Generative AI tools like ChatGPT are set to make far-reaching changes to our approach to writing tasks. Among other things, they will make some of the tedious and difficult parts of the writing process easier. OpenAI CEO Sam Altman has compared the release of ChatGPT to the advent of the calculator. Calculators brought a lot of benefits; Altman claims ChatGPT will do the same. Schools have adopted calculators, changing the way they test and teach math; now we need to do the same for ChatGPT. Rather than comforting us, the analogy with calculators should alert us to the seriousness of the task we face.
We see two main dangers from tools like ChatGPT. The first is that they will produce content that is superficially credible but completely false. Thus AI outputs can give us a very inaccurate picture of the world.
Contrary to appearances, ChatGPT is not trying (but, often, failing) to state facts about the world. Instead, it is (successfully) performing a different task – generating superficially plausible or believable responses to a prompt.
The second concern is that reliance on these tools will result in the erosion of critical skills. For example, essay writing is valuable partly because the act of writing can help us think through difficult concepts and generate new ideas.
In these early stages of the introduction of generative AI, educators may feel overwhelmed by the rapidly changing technological environment, but students are also joining us on this journey.
We suggest four approaches. ChatGPT can be a useful tool. For example, it can help generate ideas and get words on the page. Concerns about misinformation are serious. But these can best be addressed by teaching students how to use these tools, how to understand their limitations and how to check their output. Fortunately, the core skills developed by a good education provide a strong foundation for this project. Teaching students to read critically, evaluate or confirm evidence and distinguish good arguments from bad are things universities should already be doing.
One approach could be to develop specific assessment tasks where students generate, analyse and critique AI outputs. While such tasks may have some role, we would caution against placing generative AI at the centre of education.
We should remind ourselves that for most people the choice to attend higher education comes from a genuine interest in a subject. This fact may go some way towards reducing the temptation to outsource one’s learning to AI, especially if the value of completing the task is clear to students. By designing assessments that are relevant to students’ future careers and making the purpose of the tasks clear about their development, we can encourage learners to engage with assessments in the way we intend.
Assessment that connects with and capitalizes on students’ interests can motivate learners to stay engaged, so that they don’t see any value in handing over their pursuit of knowledge to AI.
One of the main concerns about AI text generation is that students may not understand what they are looking at, given the work they submit. This concern can be overcome by balancing written work with other forms of assessment. In particular, individual oral presentations cannot be taken over by any algorithm, and so could be an ideal alternative (provided, of course, that any increase in the workload of teaching staff is supported by the institution).
Supplementing traditional essays with other assessments should not come at the expense of good assessment design. On the contrary, there are good pedagogical reasons to differentiate written work with these other types of assessment; oral communication skills are highly valued in a wide range of professions.
Another strategy involves creating assignments where students have to demonstrate their understanding. This strategy may have a role, but it comes at a cost. We are moving away from pen-and-paper exams and toward authentic assessments—that is, assessments that evaluate skills students can use in real-world situations.
Very few workplaces require employees to write detailed discussions of difficult questions by hand, in private, and without ubiquitous modern conveniences such as internet connections and word processors. One option is to combine written essays with a presentation and discussion of this work during class time, possibly modeled after the format of a viva presentation or thesis defense (though this could be made lighter and shorter depending on the group being taught).
In our own experiments, we found that ChatGPT can give reliable answers about major works in our respective disciplines. However, when asked about the cutting edge of scholarly debate, it performs very poorly, as the task it was trained on has little discussion of this work. When asked to reference its claims, it tends to confuse sources that do not exist.
Dystopian visions in which AI teachers give students tasks and then hand them over to AI seem all too plausible. The immediate challenge for educators is to determine what an AI-literate skill set looks like, and how to assess whether students have these skills, especially when many of us are new to these skills ourselves.
The serious challenge posed by the ‘threat’ of artificial intelligence is to imagine what education would be like if the tools at our disposal freed us from the need to use these crucial skills.
This article is written by Julian Coplin, Lecturer in Philosophy, Robert Sparrow, Senior Teaching Fellow Nicola Rivers, and PhD Candidate Joshua Hatherley at Monash University in Australia.
[ad_2]
