How useful is AI for generating words?
Picture by Ideogram
As it happens, I first addressed this question in 2012, and the first part of this essay, down to the part about robots, is taken from the notes I made then. My opinion hasn’t changed, in spite of the enormous strides in AI in the last twelve years.
The term “automatic writing” is usually associated with a particular psychic phenomenon. However, software now exists that can take data, such as sports results, and generate reports from it.
What surprised me is that the reports are readable, and not obviously computer-generated. Take, for example, this extract of a report of a football game:
“The Badgers scored 20 points in the first quarter on a Russell Wilson touchdown pass, a Montee Ball touchdown run and a James White touchdown run.”
I have no idea what it all means, but it’s more reader-friendly than a table of results. That’s from a company called Narrative Science, as reported in an article called Computer-generated journalism: A new kind of automatic writing.
Here’s another example, this time from a project called StatsMonkey, as reported in Automated Sports Reporters Coming This Summer:
“An outstanding effort by Willie Argo carried the Illini to an 11-5 victory over the Nittany Lions on Saturday at Medlar Field.”
Note that the report is not, as you might expect, as dry as dust. I know the inclusion of the word “outstanding” would not qualify it for the Pulitzer Prize, but you have to congratulate the programmers for making the inclusion of a non-objective word possible.
So should sports journalists be quaking in their boots and scanning the job centre postings? I think not. This automated reporting is fine for crunching large amounts of data and making it palatable, but to make a report come alive still requires a human being.
I cannot, for instance, imagine a machine producing something like this:
“[George] Best had come in along the goal line from the corner-flag in a blur of intricate deception. Having briskly embarrassed three or four challengers, he drove the ball high into the net with a fierce simplicity that made spectators wonder if the acuteness of the angle had been an optical illusion.
‘What was the time of that goal?’ asked a young reporter in the Manchester United press box. ‘Never mind the time, son’, said an older voice beside him. ‘Just write down the date.’”
Aside from being nice prose, that is a wonderful example of the old writerly adage, show don’t tell. How come? Because what the writer was telling us is that this was an historic moment, a moment that would, in other words, go down in football history. Who cares about the time? The important piece of information from an historical point of view is the date.
Here's another piece from the same writer:
“It was the kind of wind that seemed to peel the flesh off your bones and come back for the marrow.”
I feel cold just reading that – and I am writing this on one of the rare sunny days here in London!
Both of those quotes are from the pen of Hugh McIlvanney, as reported in Hugh McIlvanney remains the matchless Master.
The writer of that article, Norman Giller, tells us:
“To watch McIlvanney at work is not a pretty sight. He carves slowly like Leonardo, chiselling out every word with care and consideration.”
Somehow, I can’t see anyone describing a robot reporter in such terms!
So what can AI be used for, if anything, as far as the writer is concerned?
Summarising documents
Apparently, Claude 3, the new bot on the block, can summarise documents hundreds of pages long in virtually no time at all. Something like this would be handy for me at the moment, because I’m working my way through the transcript of a conference (on AI as it happens!). Unfortunately, you have to pay in order to be able to enjoy this facility.
I can also see this feature being useful for trawling through author contracts, although personally I wouldn’t rely on AI for something that important. Not at the moment at least.
Generating ideas and outlines
I’ve tried using AI such as ChatGPT, Claude and Perplexity to generate course outlines and book outlines. The results were pretty close to what I’d come up with myself. The big difference was that the AI took ten seconds whereas I took at least thirty minutes. Moreover, not only did each bot come up with one or two things I hadn’t thought of, in the case of Perplexity I was provided with references to websites where I could download relevant articles explaining the item in more depth.
Doing the grunt work
I’ve heard of cases where authors have used AI to manage their online shop. I haven’t tried this myself, but here’s one suggestion. Around twenty years ago I was selling my books directly from my website, and ebooks through an online company that made it all seamless. However, when it came to analysing my sales figures it took ages, even though I was a wizard at Excel. I came across an AI application today that can do that work and produce detailed breakdowns for less than a $50 one-off price. I haven’t linked to it because I have not yet tried it myself, but I’m sure you can see the attraction.
Conclusion
I think writers who refuse to have anything to do with AI are in danger of throwing the baby out with the bath water. We should regard this as an updated version of the thing writers have always done, if they could afford it. That is, to focus on the writing and hire other people to do the research and manage the accounts. If you’re not yet able to do that, then thank goodness we are able to use an efficient and costless or almost costless assistant to help us.
For articles on life, literature and writing, have a look at Eclecticism by clicking the above graphic