A couple of weeks ago, a technical writer I respect posted about AI on LinkedIn. I asked him what he considered made an “AI-literate” technical writer and he came back with a list of criteria, which I found quite interesting. This article is going to cover the first thing he mentioned - knowing what AI is good at and where it falls short.
What are the strengths of AI?
In relation to technical communication, AI is particularly good at:
Revising existing content
Summarising, rewriting, restructuring, changing the tone, the stuff you’d expect from an excellent peer review.Creating first drafts
It’s possible to get AI to create decent, but not final, drafts of content. But what comes out is heavily reliant on what goes in as source material and also the quality of the prompt.Answering questions
If you’re looking for information that’s readily available and widely documented, you’ll most likely get pretty accurate answers. This can be really useful when researching an unfamiliar topic. You can get up to speed on the basics very quickly, and it’s something I’ve used AI for on various projects.Delivering consistency
You can tell AI to follow a style guide, use certain terminology, and format content in a certain way and it’ll do a good job across a huge amount of documents. It’s really strong at recognising patterns and applying the same logic to a large number of documents.
For us technical communicators, AI makes a great assistant for all areas relating to turning source content into something clearer and better. But that’s only one part of our role.
And where does it struggle?
Hallucinations
AI doesn’t know anything. It is all based on training data and the probability of what’s plausible. It will invent things and happily present wrong or out-of-date information as accurate.Judgement calls
AI can take notes of a meeting and summarise them very well. But it can’t interrupt with questions or know what questions to ask. Nor can it judge when an engineer is skimming over information or a product manager is avoiding the awkward truths and focusing only on the positives. The sort of BS that biological technical writers challenge all the time.Context
It relies on what you give it plus its training data. It needs information about your product, customers, processes, pain points, workarounds, and everything else that technical writers absorb every day. And if it doesn’t have all that context, it will likely make it up instead.New information
It’s not good at content about things that are genuinely new. It doesn’t know about new regulations, features, or workflows unless you tell it.Validation
AI can’t really check its own work, as it will use the same sort of pattern-matching to evaluate the content as it used to create it.
Where does that leave technical writers?
Many technical writers predict that we will become more like editors and fact-checkers, taking the information AI produces and making sure it actually works (and is safe). I can see how that could be the first step forward and my current work comes close to that scenario, although I write the first draft myself.
Why technical writers are still needed:
Human-to-human knowledge gathering
Judgement calls on false information or information that’s not detailed enough
Spotting contradictions
Context from institutional knowledge
Understanding what real users need
Deciding what needs to be documented and what can be left out
Spotting errors the AI has made, especially with assumptions about how a product works
Understanding the edge cases and workarounds.
Of course, this article could be out-of-date before I’ve even finished writing it!
