What AI means for editors
Photo by Anton Savinov on Unsplash
Peter Riches
Like many people, I’ve been wondering what artificial intelligence (AI) – specifically, large language models or LLMs – will mean for the future of my job.
Making predictions about the evolution of AI is an easy way to embarrass yourself, given how rapidly it’s changing (and changing the world). Even so, LLMs appear to be another tool – albeit a very powerful one – ready to assist me in performing my role rather than replace me. At least, so far.
LLMs use incredibly large amounts of data to find patterns and then predict what word comes next in a sequence. In the right circumstances and with good prompting, this enables them to produce a solid first draft or suggest credible improvements and corrections to an existing piece of writing.
The limitations of LLMs for creating or reviewing text are well documented, but in my work the most critical one is the inability to understand context. This can lead to errors of fact (or ‘hallucinations’), unnecessary repetition (repeating the same concept using different wording) and the inability to recognise biases (something that requires lived experience and self-reflection).
These are all things humans can do pretty well. Professional writers and editors can generally do them very well, because this is part of our special skill set.
That’s not to say that LLMs aren’t great for other things. My experience to date shows that LLMs rarely make spelling errors, and the grammar is usually faultless. They can identify when writing is hackneyed or clichéd, when sentences or paragraphs are very different in tone from the rest of the content, and make helpful suggestions to rephrase a particular idea.
At Red Pony, we already use professional editing tools such as PerfectIt and FRedit to correct basic errors and bring consistency to the text – either before we start work on a document (as an initial clean-up of the text) or at the end (as a final quality assurance check). While these tools don’t use AI, they do rely on extensive data sets to prompt the editor where a correction might be required.
Again, context is critical. It takes human judgement to appreciate whether these tools have identified a genuine error, or just something that doesn’t fit the standard pattern. Errors in fact can be harder to spot, but professional editors are trained to interrogate claims and question their veracity. These are human skills, and they are not easily replicated using mathematical models.
We will continue to assess how we can use new tools, including those using AI, to work faster and smarter. We will also be bringing skills and knowledge developed over many years of working as professional writers and editors to understand the advantages and the limitations of these tools, then apply them accordingly.
