Machines that tell stories. What potential do they hold — both commercially and otherwise? How might they affect the professions of journalism and technical communication?
Will “robots” soon write stories and then read them to us? (Image source: Matanya Horowitz)
I came upon a fascinating article this week, titled New technologies and their stories. The article’s contents were curated by design researcher Hanna Zoon at the Fontys University of Applied Sciences in the Netherlands.
I greatly enjoyed the article, despite a couple of hindrances in reading it. First, Zoon often refers to herself in the third person. Second, the article is in German. (My rusty high-school German, buttressed by Google Translate, rode to the rescue.)
Zoon starts by saying “Computers can do different things than people.” Then she describes some of those things.
Robot journalism, or software that automatically disseminates reports about, for example, sports scores or seismic activity. This has been around for a while (I wrote about it five years ago) and I daresay it’s already commonplace. In software technical communication we have Javadoc, which generates documentation from comments in the source code, and tools like DrExplain, which take some of the drudgery out of documenting user interfaces. But both of those are only as good as the strings in the comments and in the UI.
Narrative Science’s Quill, which claims to generate narratives — or stories — based on raw data. This one is new to me, and I’d like to hear your opinions about whether it really works.
Infographics: Although Zoon cites the work of Edward Tufte — who stands head and shoulders above everyone in conveying useful information through pictures — she’s not at all upbeat about the future of infographics. I agree with her, if only because there’s just one Edward Tufte and nobody else comes close to matching his skill level.
The Quantified Self, an attempt to describe a person’s life in terms of data, for example the amount of food consumed. While the Internet of Things will make it very easy to amass this kind of data, for me the whole thing has a whiff of the Harper’s Index: the output is more whimsical than useful.
IBM Watson, the Jeopardy-winning computer that now, apparently, also writes stories. Zoon doesn’t have much to say about Watson — except to call it “artificial intelligence at its best.” From what I’ve seen, Watson is adept at making inferences from the data it crunches. But even the great and powerful Watson struggles to make the kind of meaningful inferences that humans make every day. For at least the time being, people can do different things than computers.
Zoon ends by posing a tantalizing question: what ethical issues might arise from mechanized storytelling? I’m not sure. Despite what I wrote in 2010, I don’t see the robots putting journalists or technical writers out of work — at least not for a while. But could copyright issues arise when stories are written by software rather than by people? Could a company be held liable if its software robot dispensed medical advice that proved to be faulty?
I’d love to hear what you think about all of this, in the Comments section.
Leading Technical Communication 
Larry, this is a fascinating subject. I think it is important not only to look at the big storytelling machines, but also to look at the small storytelling algorithms that are already all around us.
A very large part of technical communication is essentially narrated data. Technical writers take in that data and execute an algorithm for narrating it. In many cases, the rules of that algorithm are worked out in detail and specified in a style guide. In this scenario, the style guide is a program expressing the algorithm and the writer is the processor that executes that algorithm.
That is not all of tech comm, by any means, but it is a lot of it.
Lots of jobs work the same way (“typewriter” and “computer” used to be job titles). But since electronic computers are better at executing algorithms than humans, over time we stop employing people to execute algorithms and start hiring people to write algorithms.(Lots of tech writers now write scripts to automate parts of their task, including the task of narrating data.)
Generic tools like Quill are not going to replace tech writers because the algorithms that tech writers execute, and the data they execute them on, are specific to individual businesses. If tech writers are replaced it will be my content engineers who write the specific algorithms required to narrate date in specific cases. Structured writing it the foundation of this switch.
Thanks, Mark. You make a very interesting analogy: that the process of tech comm is largely an algorithm that we execute to produce a story. I’m not sure I completely agree: while it’s true that we’re bound by style guides and “best practices” — and for that matter by DTDs and schemas — I think there’s still plenty of room for the tech writer to add value through good audience analysis, presentation, and so forth. I’ll grant you, though, that the line between “algorithm” and “value add” is blurry and is probably shifting. Great points. Thanks again for your comments.
Larry, I would not equate executing an algorithm with not adding value. In fact, all useful algorithms add value.
Doing good audience analysis and defining the presentation is an essential part of writing the algorithm. Good audience analysis tells us what information is needed and how it should be presented. This defines the inputs and the outputs of the algorithm. The task is then to create an algorithm that finds those inputs and turns them into those outputs. What tech writers have traditionally done is to define that algorithm and then execute it themsleves.
The main value add that the tech writer is performing, therefore, is defining the algorithm. Something has to execute the algorithm once it is defined, and that has traditionally been the writer themselves or sometimes the junior writers that they direct.
Today, it is feasible to delegate some or all of the task of executing the algorithm to machines. This may mean extracting data and creating content entirely from the extracted data. It may also mean structuring and directing content creation done by writers — making the writers more consistent and more efficient. Over an entire doc set, it can mean a combination of these things.
It can also mean doing things that are desirable but that human being cannot execute quickly enough to make them economical. Automated document assembly and linking can produce useful documents with highly desirable design features that no human author could have produced by manual effort with the time and resources available.
Indeed, conventional documentation designs have been substantially influenced by what algorithms humans can execute efficiently, often resulting in significant compromises with usability and quality compared to an ideal design.
The ability of the writer to add value in these cases comes from their ability to program machine algorithms that execute content creating algorithms that humans can devise but cannot execute efficiently, thus opening up new design possibilities and raising the value of what we produce.
It takes the mind of a skilled communicator to devise such algorithms. People with these communication skills are not going to be replaced, therefore (not at least by the current generation of technology), but they are going to have to learn some new skills for putting those communication skills into effect.
Pingback: New technologies and their stories | TechCommGeekMom
Pingback: Living and Learning | Leading Technical Communication