Over the past few years, we have been dazzled by the power of machine learning (more commonly known as artificial intelligence). OpenAI has demonstrated that their ChatGPT algorithm can perform just as well as humans can in a variety of academic settings, having been able to pass law, business, and medical exams. It’s undoubted that students could use these tools to aid them in writing essays or finishing assignments. There are now even some research papers that list ChatGPT as a co-author. This begs the question: should the use of AI in academia be regulated?
As an experienced machine learning (ML) engineer with over six years in Silicon Valley and who’s well-versed in ML algorithms, I believe the citing of ChatGPT as an author in any scientific paper is inappropriate. A scientific author is responsible for the work they publish and must be able to be held accountable should the work be of insufficient quality. This cannot be said to be true for ChatGPT, an algorithm with a great capacity for generating fictional content.
Most regulations on this topic would be somewhat unenforceable in practice
Whilst ChatGPT has been demonstrated to provide correct proofs of mathematical theorems when prompted, there is absolutely no guarantee that it will always provide correct information. Any information provided by ChatGPT will need independent verification through cross-referencing. ChatGPT was trained on a large dataset of text from the internet, which means it is subject to the same mistakes and potential biases as simply searching something on Google. Citing ChatGPT as an author is in many ways equivalent to citing the Google search bar as an author and certainly should not be allowed.
However, the topic of regulating the use of AI by students is much more complex. The immediate question is how would one even go about this? Currently, the work produced by ML algorithms like GPT and ChatGPT can be indistinguishable from that written by human beings, especially after human editing. Attempting to ban it is rather akin to giving someone an assignment and telling them they can’t use Google. There’s no consistent and reliable way to tell if they’ve followed the rules. Most regulations on this topic would be somewhat unenforceable in practice. It’s possible to tell if a student only used ChatGPT to generate the whole assignment, but difficult to tell if the student merely used it as an aid and edited the result.
There’s also the question of whether we should even regulate the use of ML algorithms by students. In years prior in schools, teachers would tell students about how they had to learn long division because you wouldn’t always have a calculator. Now everyone carries a very powerful calculator in their pockets and nobody barring a school teacher will tell you not to use it. The present situation, in many ways, feels similar. It is undoubted that ML algorithms will play a far larger role in people’s lives in the future so if anything, I believe students should be taught how to use them and how to use them well. ML is not magic. It can’t simply magic up an essay. It is a tool that requires some knowledge and understanding to use, both in machine learning and the field it is applied to.
ChatGPT, like Google, is just another advanced tool in the repertoire of the modern academic
The use of AI has already taken place in the field of software engineering. GitHub Copilot is an advanced ML-driven assistant for writing code developed by OpenAI and is heavily used within the software engineering industry. We can ask it for say, a sort function and it will devise one for us. We need only to edit it slightly. Copilot massively improves the productivity and efficiency of software engineers. Copilot certainly isn’t magic and doesn’t replace the skill of a software engineer. In fact, you have to be a reasonably skilled programmer to use it as you need to know how to interpret and edit the code it writes for you. It also can’t make the necessary design choices required in programming either. All of those are left to the engineer. Why are other fields so different? If anything, in the years to come the use of ML should be a part of the curriculum instead of being shunned.
Ultimately, ChatGPT, like Google, is just another advanced tool in the repertoire of the modern academic. Should it be given some sort of authorship status? The answer from this machine learning engineer is no. But should it, or can it be realistically banned in any sense? The answer is also probably not. Like any advanced tool, academics will need to learn when and how to use it, as well as when not to.