{"id":157395,"date":"2025-03-07T10:00:00","date_gmt":"2025-03-07T10:00:00","guid":{"rendered":"https:\/\/jerichowriters.com\/townhouse\/?p=157395"},"modified":"2025-03-07T10:00:00","modified_gmt":"2025-03-07T10:00:00","slug":"ai-a-viking-longship-and-the-future-of-writing","status":"publish","type":"post","link":"https:\/\/jerichowriters.com\/ai-a-viking-longship-and-the-future-of-writing\/","title":{"rendered":"AI, a Viking longship and the future of writing"},"content":{"rendered":"\n

For a long time, I\u2019ve held off writing about the impact of AI on what we do. Partly, none of us knows the answer and I have no special expertise in the area. But also, the impact still seemed quite remote. AI has seemed like something that might impact relatively tedious tasks (writing Google-optimised articles about vehicle maintenance, say) but not more obviously artistic \/ complex ones \u2013 like writing a memoir or a novel.<\/p>\n\n\n\n

But \u2013 well, here\u2019s a story.<\/p>\n\n\n\n

Or rather, here WILL be a story, except that I want to put a shout out for Harry Harrison\u2019s book THE WELFARE. I never met Harry, but he was a loyal Son of Jericho \u2013 always kind, always helpful, and a wonderful writer. He died recently, unexpectedly soon and with perfect bravery and grace. His book was unfinished, but his lovely writing group helped complete his book. It\u2019s available now<\/a><\/strong>, in paperback. Harry Harrison was a beautiful man; this book is a lovely memorial.<\/p>\n\n\n\n

OK. The story:<\/p>\n\n\n\n

My kids were given an extended homework task that involved writing a \u2018day in the life of\u2019 story about a Viking.<\/p>\n\n\n\n

My older boy, who is no huge fan of the written word, settled down, rather glumly, to perform the task.<\/p>\n\n\n\n

Sometime later, he showed me his story. It was typed, which is fine: he\u2019s happier with the keyboard than the pen. He had written about 200 words. And the story was really quite good. It had a simple dawn \/ voyage \/ battle \/ rest structure. The prose was simple, but clear and effective (\u201cThe air was cool, the sea was calm.\u201d) There were no typos or punctuation errors, but Tom explained he\u2019d used the spellcheck tools to get rid of them.<\/p>\n\n\n\n

He seemed genuinely proud.<\/p>\n\n\n\n

I was, I have to say, sceptical. I put the text through a plagiarism checker to make sure he hadn\u2019t just lifted it wholesale from somewhere. But he hadn\u2019t. I even checked his internet search history. Honestly, I ended up thinking that he\u2019d done the work and I was proud of him.<\/p>\n\n\n\n

He hadn\u2019t, of course.<\/p>\n\n\n\n

He\u2019d used the copilot tool in Word (which I\u2019ve never used myself or shown him how to use) and just had AI create the story lock, stock and barrel. That story, alas, was better than anything Tom was capable of writing himself.<\/p>\n\n\n\n

Now, we\u2019re talking about a 200-word story \u2018written by\u2019 an 11-year-old. We\u2019re not talking about novels, let alone novels for adults, let alone anything with aspirations to art<\/em>.<\/p>\n\n\n\n

But take a look at the following chunks of text. One was written by AI, one by my (text-averse) son, one by my (text-ophile) daughter. Oh, and just to make it more fun, I\u2019ve included a fourth chunk of text which represents a second excerpt by one of those three writers. The order is random.<\/p>\n\n\n\n

Text A<\/strong><\/p>\n\n\n\n

It was lunchtime, when the cook Eirik was calling me. I climbed down from my platform and went to eat. There was freshly caught fish and sour milk. After a while I went back to my platform.<\/em><\/p>\n\n\n\n

Suddenly, I thought I saw a ship.<\/em><\/p>\n\n\n\n

\u201cI\u2019ve seen a ship,\u201d I shouted.<\/em><\/p>\n\n\n\n

Text B<\/strong><\/p>\n\n\n\n

“Hold the net tight,” his dad told Leif as they rowed out. They spent a long time catching fish in the sun.<\/em><\/p>\n\n\n\n

At lunch time, they came back with lots of fish. Leif helped his dad put salt on some fish to keep for winter. Later, in the middle of the village, an old man got everyone to sit around the fire. He told stories about Thor’s hammer and brave fighters who sailed to far-away places.<\/em><\/p>\n\n\n\n

Text C<\/strong><\/p>\n\n\n\n

I smiled. I had no doubt our boat was queen of the seas. No one doubted it, except for Arne our old, wrinkled cook. He cooked amazingly, though sometimes pieces of hair from his long grey beard swam in the stews he concocted. His beard was so long, and he was so ancient, that people believed many generations of ravens, with feathers as black as charcoal, had roosted within its tangled mass.<\/em><\/p>\n\n\n\n

Text D<\/strong><\/p>\n\n\n\n

The new ship was getting nearer by the minute. It was here. I swung across to the other ship.<\/em><\/p>\n\n\n\n

I started by killing the weak and feeble, then moved on to the hulks and the better fighters. I so nearly got killed, but Halfdan saved me. Phew, that was close.<\/em><\/p>\n\n\n\n

Just take a moment to sort through who you think has authored what.<\/p>\n\n\n\n

OK.<\/p>\n\n\n\n

I think it\u2019s not hard to determine that Text D<\/strong> is my son\u2019s work. That just feels eleven years old, right? A boy wants a battle scene but has only the very vaguest notion of how to choreograph it, and Text D is the alarmingly hotch-potch result.<\/p>\n\n\n\n

Text C<\/strong> clearly belongs to my daughter. It\u2019s just too bananas, too off-piste, to have been generated by a machine. That\u2019s true of the whole raven \/ beard image. But it\u2019s also true of the details \u2013 \u2018pieces of hair\u2019 rather than \u2018tiny hairs\u2019, for example.<\/p>\n\n\n\n

Then Text A <\/strong>versus Text B<\/strong>? Well, I\u2019m not sure there are many tells here \u2013 except that Text A more obviously joins to Text D, so we can figure out that Text B belongs to a machine, A and D to a rather small human.<\/p>\n\n\n\n

And what does all this tell us?<\/p>\n\n\n\n

Well, I think it tells us that the current, still immature, generation of models is weirdly powerful. No news there.<\/p>\n\n\n\n

I also think it reminds us that AM \u2013 Artificial Morality \u2013 is not even in its infancy. It\u2019s unborn and barely thought of. From what I understand, my son basically asked a machine to help him cheat and the machine did so without a moment\u2019s pause. The machine did not say, as any vaguely sensible adult would have done, \u2018Look, are you sure? Wouldn\u2019t it be more helpful for your education if you actually did this work yourself<\/em>? Maybe you could do it and I could nudge you when you get stuck?\u2019<\/p>\n\n\n\n

AI without AM seems like a dangerous path to me. That\u2019s also not exactly a novel observation.<\/p>\n\n\n\n

But I also think this whole episode tells us that, for now, what those models are good at is generating the kind of text you expect to see because it\u2019s the kind of text you\u2019ve seen before<\/em>. Because the internet isn\u2019t full of people like Tom writing breezily about killing the weak and the feeble before moving on to the hulks (!), the models don\u2019t pop that kind of sentence out.<\/p>\n\n\n\n

When Tom is writing (Text A) in the way that he\u2019s expected to write for this assignment, the machines (Text B) keep almost perfect pace. In fact, from a pure prose perspective, the machine is writing just that little bit better, albeit still in the range of 11-year-old vocab and sentence structure.<\/p>\n\n\n\n

But text C? With its generations of ravens and pieces of hair? In the end, what AI models are doing is stunning, but the heavy lifting is still, in the end, a kind of creative statistical analysis of huge volumes of text. Almost inevitably, it tends towards the median, the average \u2013 the expected.<\/p>\n\n\n\n

Clearly, as models get better, they\u2019ll get more capable and the range of uses will become more expansive. Suppose, for example, you wanted to create a primer on German-English grammar along with some vocabulary lists suitable for early learners. I think you could probably create a very good first draft of that book in about a day, relying on AI to do the heavy lifting for you.<\/p>\n\n\n\n

That says to me that already, at the most mechanical end of the education market, AI is capable of (very largely) replacing the work now done by (underpaid) authors.<\/p>\n\n\n\n

But what about next year? Or in 5 years\u2019 time?<\/p>\n\n\n\n

I don\u2019t know. But:<\/p>\n\n\n\n

    \n
  1. The more distinctive your voice \u2013 the further away from that median line you tread \u2013 the longer it\u2019ll take for a machine to catch you up, and perhaps it never will.<\/li>\n\n\n\n
  2. The stronger your relationship with your actual readers, the more impossible it is that any machine could ever replace you. That relationship needs to be founded on delivery of value of course (great writing), but it\u2019s also supported by just being a nice human in regular communication \u2013 we\u2019re talking about mailing lists, here, or at least an active Facebook page.<\/li>\n<\/ol>\n\n\n\n

    And all that syncs with everything I say anyway. Write well. Write distinctively. Ditch generic ways of expressing yourself in favour of ways that are loaded with character and enriched by layers of subtext.<\/p>\n\n\n\n

    Build that mailing list.<\/p>\n\n\n\n

    Be you. Be human.<\/p>\n\n\n\n

    ***<\/p>\n\n\n\n

    FEEDBACK FRIDAY<\/strong><\/p>\n\n\n\n

    A different sort of task for Feedback Friday this week: <\/p>\n\n\n\n