Using AI for Docs


I started a channel (#ai) for this on Slack, but think the forum would be a better place for discussion.

I am wondering if anyone in the docs community is or is interested in using any form of AI in their workflow. I haven’t come across any other discussions on the topic as it relates to documentation, so hopefully some people will chime in with their ideas on how this might be accomplished, and whether or not it is even a good idea.

Currently, I have been eyeing [IBM’s Watson technology][1] as a promising platform. However, it seems that there is quite a lot of manual labor involved at this point and I’m not sure it would be very useful in the context of documentation. But I’m sure they are hard at work on enhancements, and wouldn’t it be cool to have the full-blown Jeopardy Champion, trained on your technology, at your disposal!

Another interesting development is that Apple’s Siri is being opened up to 3rd party developers and will be part of the upcoming macOS release this fall.


That’s a really interesting idea. It’s something I haven’t thought about at all. What sort of usage did you have in mind for improving workflows?


Well, as AI improves, the sky will be the limit. I’m sure there will be applications that we can’t even dream of right now. Currently, it is not a good out-of-the-box solution. It’s being employed more and more with chatbots, but there’s a lot of manual labor involved. Also, the docs still have to be there, and be well formatted. It’s almost like a poor substitute for search right now. My vision (with no idea of how to implement it) would be for the AI to be completely context aware and thoroughly “knowledgeable” about the subject. It would somehow be able to dynamically generate documentation based on what the user needs at that moment. So, for instance, it would be able to string together bits and pieces of information from different locations and weave it into a useful piece of documentation, such as an example, a short tutorial, illustration, etc. And this would be without the user having to leave their current environment. As the body of information for a subject increases exponentially, I think this will become imperative because it will be exceedingly hard to organize such a vast amount of data.


According to a recent article in the MIT Technology review, the day of the truly intelligent bot may be further off than we may think:

My own thought on this is that the problem may lie in the fact that language is far from being mechanically descriptive of the subject matter, but really operates by triangulating references to stories.

Whether you can teach machines to understand stories is a question well above my pay grade, but I take some comfort in the notion that the storyteller may have a role to play for a while yet.

As to the question of whether an AI might be used to produce docs, I wonder if that is not looking at it from the wrong end. Wouldn’t a better use of the AI be to replace the user in performing the task, thus obviating the need for docs, rather than replacing the writer as a producer of documentation.

Either way, we are out of a job, of course.



Interesting perspectives & thanks for the links!

I’m in no hurry at all for an AI to pass the Turing Test. I agree with Elon Musk, Stephen Hawking and others that it is way too much of a wildcard to have truly intelligent AI without appropriate safeguards (can’t imagine how that would be possible).

The scenario I drew is very far off (I think) as far as automating it totally. In the meantime, I think it is feasible but would still require a lot of human work from both programmers and writers.

Of course, there are surely other applications for AI, I just can’t conceive of what they would be at this time in regards to documentation.


Ah! This looks promising: