Technical writing... by robots?
March 17, 2014 7:52 PM   Subscribe

I saw a somewhat blogspammy news article the other day (regrettably, I cannot remember where, and google is failing me) which listed the top ten or twenty professions that would be eliminated by automation in the next twenty years. Among them were ones I would expect: telemarketing, retail sales, etc. Nothing surprising there. And then they threw technical writing into the mix. My question is, how could software automate technical writing?

This was surprising to me, because every economic health source seems to be saying that software jobs will keep growing indefinitely, and it seems to me that software companies would always need people adept at the technical side of things who are also capable of writing documentation, interactive help, and training materials.

I know there are tools for generating API documentation, but tech writers do more than just Javadoc, right? Was this article just stirring up controversy, or do tech writers really have reason to fear for their jobs in the future? If so, by what means?

More broadly, I'm also curious about what research, if any, has been done on automation-proof jobs. What makes a job particularly susceptible to obsolescence beside the obvious examples of welders and assembly line workers?
posted by deathpanels to Work & Money (12 answers total) 7 users marked this as a favorite
Response by poster: (The example of assembly line workers being obvious because assembly line work is designed to make humans work like machines. Replacing the humans with machines is the final step.)
posted by deathpanels at 8:00 PM on March 17, 2014

Take a look at the company Narrative Science.
posted by dfriedman at 8:04 PM on March 17, 2014

This company automates sports reporting.
posted by oceanjesse at 8:06 PM on March 17, 2014 [2 favorites]

At a certain point they say the translation of sufficiently detailed requirements into code is a mechanical process. From there I could see most of the documentation being an automated process also.

Although the documentation might be better served by being machine parseable if its other programs using our programs. But that's depressingly dystopian.
posted by TheAdamist at 8:11 PM on March 17, 2014

Check out Tyler Cowan's book "Average is Over". He looks at winning and losing jobs and skill sets in the coming years.
If your job can be done by a machine, it will. If you can effectively work with something like Watson (as seen on Jeopardy), producing something better than you can produce on your own and better than the machine intelligence can do on its own, you will be a winner and do well.
posted by PickeringPete at 8:12 PM on March 17, 2014

I'm a Geography (GIS) student at a large state school. Outside of making meaning/context decisions about what a problem is and how to represent it, most of what I work with is wrestling with data formatting, feeding those files into the right software tool, and feeding the results through several others after that. When done properly, this entails the creation of lots of metadata embedded in each file, documenting the origin, transformation and output of the data, using which tools and how.

I can see how somebody might write an expert system that would follow a worklog and yank out lists and instances of all the tools, variables and transformations. Combine these with a format template and a library of specific verbiage attached to each tool's use, then you've got the book report that writes itself.

If I had one of these, I'd be a lot more diligent about generating metadata!
posted by cult_url_bias at 10:58 PM on March 17, 2014

Best answer: Full disclosure: I'm a technical writer.

Was this article just stirring up controversy, or do tech writers really have reason to fear for their jobs in the future? If so, by what means?

To me, it sounds like somebody needed X number of items on their list about The Future and they picked technical writing it sounds vaguely like something that could be streamlined out of the software development process (sort of like test automation).

Personally, I see automation as something that will help human tech writers, not replace them. Content that can be automated is usually reference content -- like info about fields in an API call -- and it can eat up a lot of my time without providing a lot of business value. Having that type of documentation auto-generated would free me up to work on more conceptual and task-oriented content, which is what customers really want.
posted by neushoorn at 1:05 AM on March 18, 2014 [4 favorites]

I've worked as a programmer and I agree with neushoorn. The important, subject parts of technical writing are not so different from other writing as to lend themselves to automation. And the sports journalism thing is a red herring: that's essentially tabular data going into a template with a little bit of natural language fuzziness to mask the fact that it's tabular data going into a template.

Also, the whole "sufficiently detailed spec" thing is a pipe dream that people have been chasing for decades. You know what spec turns out to be sufficiently detailed to autogenerate code from? That would be the code itself. The closest we've come is making it easier to create domain-specific languages, and that's not close at all.
posted by d. z. wang at 5:03 AM on March 18, 2014 [3 favorites]

Response by poster: I'm aware of Quill (Narrative Science's platform for generating news stories), but that's just taking sports statistics, which are already well formatted, and inserting them into sentences. I guess I'm having a hard time imagining how software documentation could be generated, given the general lack of metadata. Unless there's a sudden rise in the development and use of very metadata-rich DSLs?
posted by deathpanels at 5:14 AM on March 18, 2014

Best answer: Unless there's a sudden rise in the development and use of very metadata-rich DSLs?

Even then, it's hard for me to imagine any auto-generated content that does more than simply describe the software, and I believe a technical writer's value lies in being able to do more than that. Metadata can probably answer the question "What does it do?", but users actually want to know "How do I use it to do X?". Sometimes X might be easy to explain with an automated answer, but usually, X is a complex business question that requires a human explanation.
posted by neushoorn at 6:21 AM on March 18, 2014 [1 favorite]

Check out or The Atlantic.
posted by PickeringPete at 3:04 PM on March 18, 2014

So I answered this the other night, but today I noticed an article about an automated system generating an LA Times article on an earthquake, and there was a pretty good reddit thread listing some prior art.
posted by cult_url_bias at 1:03 PM on March 19, 2014

« Older Requesting guidance re: stalkerish ex-friend   |   Wordpress install problems Newer »
This thread is closed to new comments.