Why would they do this? In essence, it saves time and effort, and ensures that (expensive) degrees can be successfully obtained, albeit in an arguably inauthentic manner. Up until today, students have generated their own work which they then review and edit themselves, or pass on to professional proofreading/editing services for refinement, improvement, and guidance. The process of editing is an extremely important one; hence, the emergence of new AI models raises the question as to whether they, too, can successfully perform such a process.
I argue that, no matter how sophisticated such models become, the job of editing is ultimately one that can only be performed by humans. There are several reasons for taking such a stance. Firstly, the inherent subtlety of language makes it almost impossible for AI to make decisions as to which word is more impactful, effective, or indeed appropriate in a given sentence. For example, ‘slender’, ‘lean’, and ‘skinny’ may have broadly the same meaning, but they have markedly different connotations, a nuance AI cannot capture as it lacks social awareness. Secondly, the use of words such as ‘very’, ‘a lot’, ‘nice’, and ‘good’ are not in themselves incorrect, but they are over-used and somewhat bland. More dynamic alternatives would be ‘extremely’, ‘substantially’, ‘likeable’, and ‘outstanding’, respectively. These are issues AI will not naturally consider, leading to a somewhat bland piece of work. This is especially problematic in a more creative piece of work, such as a poem or novel, or a work aimed at persuading its readership to accept a particular argument. Thirdly, some rules of punctuation can be a little obtuse, with no clear consensus as to when and how they should applied, notable examples being the use of commas, semi-colons, dashes, and parenthesis. It is therefore unclear how AI, programmed by humans, will handle such issues.
Fourthly, there is the issue of how a piece of work is constructed. At the sentence level, there are myriad ways in which a sentence can be formulated. For instance, it could be written in the active voice, as in ‘I conducted an experiment to…’, or the passive voice, as in ‘an experiment was conducted’. Both are acceptable, but the choice as to which to use is often dependent on the rules of a particular discipline, institution, or journal, decisions AI cannot make. Expanding such considerations to the level of paragraphs, the coherence and flow of a given paragraph is a matter of subjective judgement on the part of the editor; they have to make a judgement as to what sounds right, whether each sentence leads logically on to the next, whether there is unnecessary repetition, and whether the paragraph has an appropriate endpoint. With regard to the work as a whole, the editor needs to assess the integrity of the overall argument. They must consider whether the author is telling a coherent and logically ordered story with a clear beginning, middle, and end, and whether each section or paragraph flows neatly on to the next, often through the use of well-chosen linking sentences. This, again, is as much a question of aesthetic judgement as it is of technical exactitude, an aspect that will be beyond the capabilities of AI.
Fifthly, AI will not be able to judge the appropriateness of certain expressions as it lacks an understanding of social/moral conventions which will often vary between cultures. A common example is the use of gender neutral language ‒ how will AI handle the difference between a reference to an actual ‘he’ and the use of ‘he’ as a generic (but unacceptable) term that denotes an unspecified individual. Moreover, whereas an editor may highlight certain arguments as being morally unacceptable, an extreme example being ‘holocaust denial’, AI does not have the capacity to make such judgments, which require an understanding of more abstruse concepts such as attitudes, beliefs, and values.
Sixthly, AI will not have the capacity to judge what constitutes valid evidence for a particular argument. This requires an intricate familiarity with a particular discipline, which AI can potentially learn, but the selection of one piece of evidence over another is often a matter of subjective judgement on the part of the author, the appropriateness of which is best assessed by an expert editor in the field. Moreover, AI may struggle to identify apparent inconsistencies in the use of dates and names, something an experienced editor will be able to handle.
Finally, the editor will form an impression of the work that enables them to make suggestions as to how to improve certain aspects, express particular elements in different ways, and reorganise the material to enhance its impact. They are also able to highlight any elements that are unclear. AI is unlikely to be able to make assessments that rely on an understanding of the inherent meaning of the work, and indeed the wider context in which it is situated, be this at the level of the course, discipline, or society at large.
Clearly, as AI advances, its capabilities will expand to an unprecedented level. In the final analysis, however, editing is an intensely human process that relies on inherently human capacities and traits, and for that reason will never be surpassed by AI.
David K. is a former university lecturer and a currently a senior editor with the Ultimate Proofreader, a leading UK specialist provider of academic and proofreading services.