Amazon Set to Produce AI Generated Novels

[Pizza_Cant_Read]

Was browsing my favorite technology news site and saw this Quartz article. What do writers here think about AI generated books? Assuming Google and major publishers move in this direction, what do writers think about machines "taking away jobs" and if such software is ever released under an open source license would you consider using it to prototype your own stories and to otherwise accelerate your own release cycle?

https://qz.com/1453112/amazon-has-ev...riven-fiction/

[Nate the great]

Your title is incredibly misleading. We have no evidence that Amazon is getting ready to "Produce AI Generated Novels", and that article doesn't even present a good argument that Amazon is capable of doing so.

[Pizza_Cant_Read]

Originally Posted by Nate the great

Your title is incredibly misleading. We have no evidence that Amazon is getting ready to "Produce AI Generated Novels", and that article doesn't even present a good argument that Amazon is capable of doing so.

The idea of Amazon using algorithms to try and lock down some sort of formula for interesting content, based on observations of Netflix and Chinese tech media companies doing similar, is only speculation... you are right. I will also freely admit that it is my personal belief based on reading AI news that such experimentation is inevitable but of course I wouldn't ask for you to speculate on that speculation. I first read this article on Slashdot, where there are already many comments from people debating the technological feasibility of AI generated novels and I didn't come here for more of the same. But MobileRead is an interesting and unique community which straddles both tech and writing which is why I asked:

"Assuming Google and major publishers move in this direction, what do writers think about machines "taking away jobs" and if such software is ever released under an open source license would you consider using it to prototype your own stories and to otherwise accelerate your own release cycle?"

Cheers for first response! I hoped this would create an interesting conversation based on tools like this being used by independent authors as well as giant god-Corps but it is all outside my purview. Heck... I feel insecure posting something of any length, even anonymously online because I know I can't write my way out of a wet paper bag

-PB

[fjtorres]

Originally Posted by Nate the great

Your title is incredibly misleading. We have no evidence that Amazon is getting ready to "Produce AI Generated Novels", and that article doesn't even present a good argument that Amazon is capable of doing so.

Indeed.
Microsoft and a hundred startups have the exact same "capabilities" and the exact same inclination as Amazon. Which is to say zero interest in publishing.

The highly touted "AI" in most media brags is simply a slightly more sophisticated database query technology for big businesses. Amazon's "AI" efforts, like Microsoft's, IBM, Google, et al, are aimed at big companies writing their own line-of-business applications, mostly on the cloud computing services that are displacing the last business-class mainframes and a lot of corporate centralized computing functions.

Here's a good summary of where Microsoft is going, which is similar to what everybody else in the field is doing:

https://www.zdnet.com/article/making...pproach-to-ai/

Note:

Microsoft is working on a number of AI "Accelerators," solution templates and analytics templates to give users a way to build on top of some repeatable patterns and practices around AI. Microsoft announced earlier this week a new solution accelerator for virtual assistantsthat it's making available on GitHub that is aimed at allowing customers and partners to build their own enterprise-grade "conversational assistant" that can be personalized for their customer base. This accelerator uses the Azure Bot Service under the covers.

Microsoft's plan is to try to get developers to write new apps and services using the Azure Bot Service or by calling specific application programming interfaces (APIs) that it offers through its so-called cognitive services. Users have the option of building on top of Microsoft's own vision, text, speech, knowledge and other cognitive services and/or to customize these APIs.

Microsoft has identified some key, repeatable patterns for AI solutions around things like business agents/bots; person/object/activity detection; knowledge mining of documents and video; and autonomous vehicle, networks and other kinds of systems. And it is targeting these around a handful of key vertical markets, like healthcare, insurance, finance, retail/marketing and manufacturing.

Speaking of cognitive services, Microsoft is taking an interesting tack in making some of its cognitive services available for use in containers. The idea is by containerizing these APIs, Microsoft can make them available in cases where users can't use them easily in the public cloud.

As of this week, Microsoft made five AI capabilities available in containers in public preview form: key phrase extraction, language detection, sentiment analysis, face and emoticon detection and OCR/text recognition. More containerized services are planned, including speech and language understanding.

Bolding mine.
Some of the pattern recognition functions *could* be used to assemble narratives from boilerplate but there is little money in that area. Those with the capabilities have better (more profitable) uses for the resources and those with the interest have neither the capabilities nor the resources.

You might see a stunt or demonstration project out of academia but in practical terms it's a dead end. Only actual humans work for as little return as there is in writing fiction.

[DiapDealer]

Originally Posted by Pizza_Cant_Read

But MobileRead is an interesting and unique community which straddles both tech and writing which is why I asked:

While there's some writer-centric sections, I've always been under the impression that Mobileread (not Mobilewrite) was a community primarily focused on reading-tech and reading.

[Pizza_Cant_Read]

Originally Posted by DiapDealer

While there's some writer-centric sections, I've always been under the impression that Mobileread (not Mobilewrite) was a community primarily focused on reading-tech and reading.

you got me there hahahahaha

[DiapDealer]

Well, it wasn't that much of a stretch, truth be told. There's still large portions of society that believe all avid readers want to write a novel.

[Pizza_Cant_Read]

No, you got me I guess the writing passionate people are just the core crowd I tend to see outside the tech forums on MR. And maybe I am one of those you reference. Once while traveling in India I thought I would write a story about someone who went through Mother Theresa's orphanages. It was going to be a story with a message and I really wanted well developed characters based on the children I observed there who were actually not well taken care of. Many continued on to work there in an understaffed position with low wages because where else were they going to to go? Visiting the orphanage and seeing the bad things was perspective changing, especially after seeing the pilgrims who put her on a pedestal... but writing was much more difficult and I haven't done much to maintain core skills since leaving school. I like to think that AI can help the common writer develop something of a boiler plate or at least maintain consistency, think through the stories they wish to create and hopefully give them room to expand in weak areas.

Like I said, I hoped that juxtaposed against the threat of mega corporations simply hiring writers to 'adjust' AI generated stories would create an interesting conversation. And really, I don't think Amazon would be afraid of using real people to pose as authors, with strict NDAs if they found success. Especially in more formulaic genres. I have never subscribed to Kindle Unlimited but I hear much of the content is low grade anyhow. How would we know if someone truly wrote something rather than a bot? Well I guess because such a team to QA the product would cost more than the poorly paid author mentioned earlier :P

It is all about the Turing Test. And now that it has been passed why not make novels with human QA? Better and more profitable would be world generation for video games and other things requiring that level of detailing. But back to the real authors- it is disappointing to see my favorite authors leave certain things less developed or contradictive when those things interest me in the story. And when I see something wrong it shatters the world until I choose to move on. Probably not important in the grand scheme but maybe parts of a story could be generated leaving the essential parts which make the story real and original to the author. That could be big. Maybe. I don't know. But if I relate this to personal assistants then localizing these technologies to empower individuals rather than feeding our soul into the cloud could make a huge difference in letting us make that choice and shape our own development.

[fjtorres]

The problem is "AI" isn't anywhere near being able to actually create.
Most AI Hype is just hype.
What so-called AI can do today is pattern recognition and decision-tree navigation. Any story it "created" would be a "choose your own adventure" splicing of boilerplate. AI can't even write a single paragraph from scratch.
Turing test is just a test of formatting not a real cognition test; how to splice canned answers in a conversational mode to fake out humans.

Here's what honest techies in the AI business are saying:

https://www.cnet.com/news/ai-is-very...red-to-humans/

https://www.iflscience.com/technolog...usly-terrible/

Someday somebody might assemble a database of text snippets and writing rules and get a computer to write a readable story. But it won't be soon and the computer will be named MULTIVAC.

[Pizza_Cant_Read]

"Cognition" is an ambiguous metric next to "sentience". Nothing is written from scratch, really. Everyone builds on what they observed or read in the past. Self thinking? Learning? I would award you karma just for the savvy of your argument but I don't have any left.

Are you a writer? If so would you use these tools if you were enabled to do so? If you are a writer then how would you feel about them being used by companies with these enormous resources?

I don't know if I believe in AI, and I absolutely don't know what defines such. Even with my measure of intelligence I didn't know whether to use a hyphen with "AI Generated". True AI is defined by sentience, no current applications or even perceivable applications are capable of that. All of this I think is a different argument and maybe that is your point, that I mis-titled my post.

Originally Posted by fjtorres

The problem is "AI" isn't anywhere near being able to actually create.
Most AI Hype is just hype.
What so-called AI can do today is pattern recognition and decision-tree navigation. Any story it "created" would be a "choose your own adventure" splicing of boilerplate. AI can't even write a single paragraph from scratch.
Turing test is just a test of formatting not a real cognition test; how to splice canned answers in a conversational mode to fake out humans.

Here's what honest techies in the AI business are saying:

https://www.cnet.com/news/ai-is-very...red-to-humans/

https://www.iflscience.com/technolog...usly-terrible/

Someday somebody might assemble a database of text snippets and writing rules and get a computer to write a readable story. But it won't be soon and the computer will be named MULTIVAC.

[fjtorres]

Originally Posted by Pizza_Cant_Read

"Cognition" is an ambiguous metric next to "sentience". Nothing is written from scratch, really. Everyone builds on what they observed or read in the past. Self thinking? Learning? I would award you karma just for the savvy of your argument but I don't have any left.

Are you a writer? If so would you use these tools if you were enabled to do so? If you are a writer then how would you feel about them being used by companies with these enormous resources?

I don't know if I believe in AI, and I absolutely don't know what defines such. Even with my measure of intelligence I didn't know whether to use a hyphen with "AI Generated". True AI is defined by sentience, no current applications or even perceivable applications are capable of that. All of this I think is a different argument and maybe that is your point, that I mis-titled my post.

Not a writer.
(Other than technical reports or articles.)

The best presentation I've seen on AI was in BioWare's MASS EFFECT trilogy, where they clearly distinguished between true cybernetic sentience, true self aware Artificial Intelligence with natural self-preservation and replication instincts, and what they termed Virtual Intelligence, conversational database interfaces that seem to be Intelligent until they reach the limits of their programming. They never pretend to be sentient because in the MASS EFFECT universe true AIs are forbidden tech because have a history of going rogue and hostile to organics.

What we have developed so far is a very crude breed of what they called VI's: software that can pass as intelligences under some limited conditions but it's just an illusion, a facade hiding purely deterministic procedural algorithms.

Where the debate of the dangers of "AI" comes from isn't about the software developing sentience but that the software has a certain amount of runtime self modification so that the actual execution code isn't programmed/vetted by humans. Humans write software that tunes itself and determines its execution parameters so debugging and quality assurance is pretty much impossible. instead of being hard coded, bugs are variably emergent. It's like a machine that has an intermitent glitch that comes and goes at random.

So the thought of many is that the much hyped AI tech is inherently unstable and untrustworthy. (You ever hear of the microsoft Tay conversational "AI" chatbot? Went off the rails in a few hours. https://www.techspot.com/news/64222-...-bot-less.html)

"AI" isn't much of a threat if it's used to adjust the lighting in a room but gets iffy if it's in charge of driving a sixteen wheeler down a freeway and could be positively deadly if it's controlling a war machine, like the robot tanks the russians claim to have developed.

(Personally I hope it's as much bluster as some of their space tech claims because it takes a spectacularly stupid person to teach a machine to kill humans and then set it loose without human supervision.)

Bottom line, most AI claims are just marketing hype so far. The tech has its uses in very narrow situations but it's, as the Google guy said, very, very stupid software.

[rcentros]

Originally Posted by Pizza_Cant_Read

"Assuming Google and major publishers move in this direction, what do writers think about machines "taking away jobs" and if such software is ever released under an open source license would you consider using it to prototype your own stories and to otherwise accelerate your own release cycle?"

I don't think "paint-by-number" writing is ever going to work. They've been trying this with "fill-in-the-blanks" writing software for years now. The problem is that writing is not a mechanical process. There's more involved then producing something that "looks" like a book, or a screenplay.

[DNSB]

Originally Posted by rcentros

I don't think "paint-by-number" writing is ever going to work. They've been trying this with "fill-in-the-blanks" writing software for years now. The problem is that writing is not a mechanical process. There's more involved then producing something that "looks" like a book, or a screenplay.

Over the years, there's been more than one author whose output made me wonder about computer generated writing. The type of book sometimes referred to as a 10 finger exercise -- the deadline is 3 days away and I need another 30,000 words!

[Pulpmeister]

There's a funny novel about AI-generated literature called "The Silver Eggheads" (1961) by Fritz Leiber. At least, that's the title on my paperback copy. The equipment that writes the books are "wordmills" and the product is known as "wordwooze." All the author does is log on, enter a few key points, the wordmill does the rest.

[fjtorres]

Originally Posted by rcentros

I don't think "paint-by-number" writing is ever going to work. They've been trying this with "fill-in-the-blanks" writing software for years now. The problem is that writing is not a mechanical process. There's more involved then producing something that "looks" like a book, or a screenplay.

Most authors I've run into despise grammar checkers.
There's a reason the term "grammar nazis" exist; people want to express their ideas in their own voice, with their own cadences, constructions, and expressions, regardless of what the rules may demand.

"Rules are meant for breaking" is the mantra of most good writers. The trick is knowing when and how. A "writing assistant AI" would go insane trying to reconcile the various styles of different writers, to say nothing of the speech patterns of characters.

Grammar checkers are useful for business or technical writing, where precision and clarity are paramount but for fiction, where mood, tone, and metaphor matter, they are counter-indicated. The more the software tries to do, the less useful it becomes.