Every so often, when our bookshelves are full, and we're faced with a choice between making more space or giving up buying books, I sell the odd few volumes at Ebay. I hadn't done it for a while, and when I went back, at the end of last year, I discovered that the site has a new feature. It lets sellers use AI to create the description of the item they are listing. So, I tried it. I filled in all the basic information—title, author, publisher, condition, special features, all that sort of stuff—so the bot had all of that to work with.
Here's the human-made description I wrote myself:
John Silence by Algernon Blackwood, published by Eveleigh Nash in 1916 (originally published in 1908).
From the library of the distinguished writer of science fiction, fantasy and weird fiction, Donald Wandrei (1908-1987).
Original blue cloth with original dust jacket. Binding firm, endpapers toned, a very small amount of spotting here and there. The dust jacket has a small tear near the spine, a few nicks and chips, and a pink stain at the front bottom edge. An attractive, very good copy.
Please do inspect the photographs as they form part of this description (and they'll give you a much better picture, if you'll pardon the pun, of the book's condition).
Comes from a smoke-free and pet-free home.
All books listed are from my own personal collection. I am selling to create space (to buy more books!). All books have been cared for in a damp-free, smell-free environment. If you would like more photos, I'm happy to take more. Need more details? Just drop me a line; I'm always happy to be of assistance.
And here's the Ebay bot's effort:
This is a hardback book titled "John Silence" by Algernon Blackwood, featuring the occult detective character.
The hardcover binding and English language make it a valuable addition to any literature collection. The author's unique perspective on the topic of literature and fiction makes it a must-read for any book lover.
This edition is part of the library of Donald Wandrei.
The book was listed for £270. So, keeping in mind the fact that the bot is supposed to help you sell a rare book—and not a cheap one that you can pick up at any railway bookstall—what do you notice?
- It's as dull as dishwater and has no personality.
- It includes hardly any of the information it had to work with and absolutely nothing about the book's condition or the date it was published.
- It incorrectly states that the 'edition' is part of the Donald Wandrei library.
- Look at that bit I've highlighted in red. According to the first part, being a hardcover book in English is enough to make it valuable. According to the second part... Well, say what now?
Using my human-made description, I sold the book within two hours of listing it. The bot's description would most likely have resulted in prospective buyers sending me messages to request information that should have been in the listing to begin with. I'd have ended up writing responses that shouldn't have been necessary, spending more time on doing that than I would have done in producing my own human description to begin with. Or buyers might just have passed the book by, and I'd have lost that sale; after all, if I couldn't be bothered to put in some effort, or didn't have the knowledge to put together a proper description, why should they buy from me?
Of course, not all AI is created equal, and some bots are better than others. But at what? In my experience, some are better than others at producing writing that isn't so dull that it makes you want to pull your own eyes out, but that writing always contains a certain (or possible a large) amount of utter nonsense. As time goes by, they may get better at forming a sentence, but it's looking like the nonsense is here to stay. If we're lucky, they'll end up just like your old Uncle George, who knows everything about quantum mechanics without ever having actually studied the subject; he's so confident, so sure of everything he says, and so capable of boring all and sundry at Christmas parties with what we all know is complete tripe. If we're unlucky, the better AI gets at sounding like it knows what it's talking about, the more difficult it will be for us to separate information from misinformation. Neither outcome is desirable.
The BBC recently undertook research into four AI assistants: OpenAI’s ChatGPT; Microsoft’s Copilot; Google’s Gemini; and Perplexity. They wanted to find out whether these assistants 'provided accurate responses to questions about the news; and if their answers faithfully represented BBC news stories used as sources'. The bots were given access to the BBC website and asked to use BBC stories when putting together answers. The BBC found that: 91% of AI answers had 'some issues'; 51% had 'significant issues' of some form; 19% of AI answers that cited BBC content contained factual errors, including incorrect numbers and dates; 13% of the quotes supposedly taken directly from BBC articles 'were either altered from the original source or not present in the article cited'.
That bit in red is very important. It is something I've encountered every time I've experimented with an AI bot; they alter what should be a direct quote, they make stuff up, and they attribute information to the wrong person/organisation. And, as the BBC rightly points out, 'inaccuracy and distortion can lead to real harm'. The BBC concluded that 'AI assistants cannot currently be relied upon to provide accurate news and they risk misleading the audience'.
I can't help thinking that, as more and more people cotton on to the fact that AI bots get a lot of things wrong and make stuff up, they'll begin to avoid interacting with AI-produced content. A bot assured me that bots cannot lie, but an unintentional untruth is capable of doing exactly the same damage as an intentional lie. So why rely on AI and risk being misinformed? And if you think people won't be able to tell the difference between AI content and human content as these bots 'improve', I have two answers for you:
- Human beings are already becoming familiar with the tone of AI-produced content and will become better able to spot and avoid it as time goes by. They learn and evolve in a way that chatbots simply can't.
- If AI bots get better at seeming human, and humans are duped into believing they're reading things created by humans, will that be a good thing, considering their tendency to hallucinate? And no, there's no sign that those hallucinations are going to go away.
Going back to that Ebay bot, my own response to encountering AI-generated descriptions at that site is generally one of the following:
- Laugh and move on.
- Try to find actual information, get frustrated, then move on.
- Assume that the seller doesn't know what they're selling and move on.
You'll notice that bit about moving on. It is so boring to read the same dull, droning nonsense over, and over, and over. No, thank you... I'm outa here!
Bots cannot think, feel, empathise, or care. They cannot reproduce accurate information when it's handed to them on a plate or determine when it is outdated or out of context. They make stuff up. They often produce lots of words while saying very little of substance. In short, they are the world's worst writers, the people who are fired after a week because they bore people, mislead people, and damage company reputations.
The only way to connect with your human audience is to reach out to them with human-produced content. Ultimately, life for us humans is too short for us to be forced to spend a lot of it interacting with unthinking, unfeeling, uncaring machines.