Although I make mistakes when I write, it is an activity that comes easily to me. Sometimes it is difficult: finding the best title, opening paragraph, closing paragraph, how to express a thought with the right nuance; all these things can be hard. If I know what I want to say though, the words come naturally and I have never felt the need for AI assistance.<\/p>\n\n\n\n
Nevertheless, AI can generate content, lots of it, and on the surface it is pretty good. Give it a task like, for example, writing a letter to a local politician expressing concern about a dangerous pedestrian crossing, it may well do the job quicker and better than the average person – better, that is, in terms of grammatical correctness, smooth writing style, and staying on topic.<\/p>\n\n\n\n
AI can even write a novel; but publishers are wary. The topic is in the news because, apparently<\/a>, a hot new novel entitled Call me, I’ll hide the body<\/em>, sold according to “what sources said was a $2.4 million deal for the novel and a follow-up,” has been withdrawn by a foreign rights agent because of suspicion that AI was used “as a resource in the writing or editorial process.” <\/p>\n\n\n\n The use of AI here is not so much a problem of quality, but rather of copyright. Much of what we call artificial intelligence is based on searching and synthesising content harvested from multiple sources and an AI-generated novel might include text taken from existing works.<\/p>\n\n\n\n That said, if this novel was in part AI-generated but was nevertheless considered to have worldwide sales potential to merit a multi-million dollar advance, that is an impressive win for AI, even if the deal has now turned sour. <\/p>\n\n\n\n If I felt so inclined, I could instruct Chat GPT or similar to write a 500 word article on software development every morning and post it to this blog. I have just tried it; I will not bore you with the full results but in a few seconds I got a piece entitled AI coding moves beyond autocomplete: why GitHub’s latest enterprise push matters<\/em>.<\/p>\n\n\n\n The AI-generated article looks OK on the surface but an editor would soon find defects. The fact that AI coding has moved beyond autocomplete is ancient history. The piece summarises some recent GitHub announcements but without any critique of how well the company is keeping up with the competition, or whether there are gaps between what it claims and what it has achieved, or what opinions practitioners and analysts have of GitHub’s latest AI features.<\/p>\n\n\n\n One must acknowledge though that regurgitating company announcements as news pieces is not only done by AI. Journalists do this as well sometimes; and if you are a journalist tasked with posting, for example, five pieces a day, this is hard to avoid, there being no time for thoughtful and well-researched analysis.<\/p>\n\n\n\n What if I worked on the Chat GPT output, submitted further prompts to ask for refinements, followed up the sources, added the results of my own research? Could it then merit publication?<\/p>\n\n\n\n It is not a path I want to travel, but perhaps it could.<\/p>\n\n\n\n Like many of us, I often consult reviews of restaurants and hotels before making a booking, and read online reviews of products before making a purchase. It is easy to tell which reviews are fake, being either disguised advertising or meaningless repetition of the official description; and which are based on actual experience of visiting the premises or using the product.<\/p>\n\n\n\n AI can easily create a fake review, but can never write a real one. The best it can do is to summarise and copy the reviews of others. <\/p>\n\n\n\n It is not easy for professional writers. First Google uses AI to borrow our content and show it in search results so that users do not visit the sites which monetise our work. And now AI creates complete works that seem almost like human writing. The key word though is “seem”. Much of the content is illusory and does not bear examination. It does not describe experiences or analyse facts or convey emotion. <\/p>\n\n\n\n For these reasons I consider that human writers have a reasonable moat to defend against AI replacement.<\/p>\n","protected":false},"excerpt":{"rendered":" Although I make mistakes when I write, it is an activity that comes easily to me. Sometimes it is difficult: finding the best title, opening paragraph, closing paragraph, how to express a thought with the right nuance; all these things can be hard. If I know what I want to say though, the words come … Continue reading Can AI write as capably as humans?<\/span>