Skip to main content

Quartz Uses AI-Written Article to Show Problems With AI-Generated Fiction

There’s a lot of hype surrounding AI right now, and not much in the way of substance. In many cases a purported AI system is little more than a concatenation of if-else statements (or even worse, low-wage workers hired to fake AI functionality) but the hype is so loud right now that media outlets continue to speculate about what AI could do next.

On Friday Quartz decided to show us just how little we had to fear from AI actually taking writing jobs away from human beings.Under the title "Amazon has everything it needs to make massively popular algorithm-driven fiction", Quartz published a 1,283 word piece that was clearly written and published without any human intervention whatsoever.

Yes, just to show us what AI can really do, Quartz has handed control of the content on their site to an AI. This is clearly the case because no human would publish such a long piece without making sure it included an argument or at least a topic sentence. This piece has neither, and while this piece was coherently written, its lack of key details gives away its AI-generated source.

Kudos, Quartz, for showing us the limitations of AI.

O O O

Since my sense of humor is sometimes obtuse, I want you to know that I am actually mocking Quartz here.  That piece is so badly written that it really doesn’t have an argument for me to counter. It is a rambling recital of facts that never approaches a point or reaches a conclusion.

It is so bad that one could honestly use it as an example of human-generated content setting the bar so low that an AI could easily replace a writer.

Edit: This piece is so badly written that someone had to explain to me that it was really about garden variety analytics that is commonly used by news sites, and not algorithms at all.

But could AI replace book authors?

Sure – I don’t expect it to happen, but it is technically possible. Computers have been writing books for over 30 years, just not at scale. At this point the concept is more an idea than a reality, and I don’t see there being enough money in AI-generated books to motivate anyone to create an AI that could credibly write books.

The thing about the book market is that it is neither high-volume nor high-profit compared to many other markets. The ebook market really isn’t that big (not compared to games, or finance software, or just about anything you can name). There are any number of B2B uses for AI that will generate a much better return for Amazon than AI-generated books (as I am sure Amazon has already concluded) which is why I have to wonder why Amazon would waste their time with this idea.

Furthermore, there are so many examples of human-powered fake AI (even for things like receipt digitization, a feat much simpler than writing a whole book) that I would put good money on the probability that any commercial publisher claiming to produce AI-written books will actually be lying to us, and will actually be sourcing their books from ghost writers.

It is easier to fake AI than to make AI, after all. (And when it comes to books, it is also probably cheaper.)

Next!

image by peyri via Flickr

Similar Articles


Comments


Mike Cane November 20, 2018 um 9:57 am

>The thing about the book market is that it is neither high-volume nor high-profit compared to many other markets.

Unless you’re James Patterson. Who would be the first to use AI for his work. If he doesn’t already…

Nate Hoffelder November 20, 2018 um 10:56 am

His ghost writers are cheaper, I think.


Tom November 20, 2018 um 11:19 am

The article isn’t about using AI to create fiction. It’s about using AI to analyze consumer behaviors, and then incorporating those findings into the fiction itself. The linked article to the Netflix director has a quote from him that captures the issue:

"Like Beasts, Maniac will stream on Netflix, which has its own surreal development process. "Because Netflix is a data company, they know exactly how their viewers watch things," Fukunaga says. "So they can look at something you’re writing and say, We know based on our data that if you do this, we will lose this many viewers. So it’s a different kind of note-giving. It’s not like, Let’s discuss this and maybe I’m gonna win. The algorithm’s argument is gonna win at the end of the day. So the question is do we want to make a creative decision at the risk of losing people."

https://www.gq.com/story/cary-fukunaga-netflix-maniac

"So the question is do we want to make a creative decision at the risk of losing people." – A creative decision based on an algorithm. That same question could be applied to books, which is the gist of the somewhat scattered article.

A similar argument in favor of using algorithms is made in the book 'The Bestseller Code'.

https://www.goodreads.com/book/show/28220900-the-bestseller-code

Nate Hoffelder November 20, 2018 um 12:48 pm

Oh, is that what they meant?

Thank you for telling me. I would be embarrassed about missing the point but this piece was so scattered that it was really hard to follow.

Mike Cane November 21, 2018 um 7:30 am

And so eventually all of Netflix programming will become blanded-out with scenes and beats everyone has already seen. Algorithmically predictable with no human surprises.


Will Entrekin November 20, 2018 um 1:25 pm

"All that power comes with great data, which Amazon’s publishing arm is well positioned to exploit in the interest of making books tailored exactly to what people want—down to which page characters should meet on or how many lines of dialogue they should exchange. Though Amazon declined to comment specifically on whether it uses data to shape or determine the content of its own books, the company acknowledged that authors are recruited for their past sales (as is common in traditional publishing)."

Echoing above, I’d argue this paragraph is the article’s thesis, and it’s a powerful idea. I’ve long held that I’d be more interested in KDP Select if Amazon didn’t tell me just how many pages were read but which ones, too. Not saying it would change anything, but if I saw data that indicated that most readers gave up midway through a given chapter, e.g., I’d certainly consider revisiting that chapter.

Nate Hoffelder November 20, 2018 um 2:01 pm

dammit

On the other hand, Kobo has been talking about exactly this for years and years, and JellyBook is already selling this kind of info to publishers.

Nate Hoffelder November 20, 2018 um 2:06 pm

And this is analytics, not algorithm-driven. No wonder I couldn’t figure it out.


Tom November 20, 2018 um 2:22 pm

The writers of 'The Bestseller Code' offer their analytic services to publishers too:

http://www.archerjockers.com/

Nate Hoffelder November 20, 2018 um 2:27 pm

Sounds like what Booklamp used to do.


Lupe dy Cazaril November 21, 2018 um 2:25 am

Sorry if I sound like an AI, but I feel a bit more ad hominem in your comment than an AI should deserve.


Write a Comment