Comments on: Can a computer replace J.K. Rowling?

By: StevoR

StevoR — Mon, 01 Aug 2016 04:49:11 +0000

And, of course, other than the random novelty one gets when flipping coins, there is nothing new, no interesting plot elements, new characters, or novel magical tropes in this new text.

Well, there is the rather interesting twist that Harry has somehow changed genders and become a girl or trans : "“What are you doing, Harry?” said Hermione, staring down at her."

By: Douglas C Alder

Douglas C Alder — Sat, 16 Jul 2016 06:58:36 +0000

Oooooh didn’t know she was writing crime fiction under that name – just ordered all 3 – thanks Greg.

By: Julian Frost

Julian Frost — Thu, 14 Jul 2016 17:20:13 +0000

Surprising that no one mentioned this Roald Dahl story.

By: Brainstorms

Brainstorms — Thu, 14 Jul 2016 16:38:43 +0000

You missed the word “alone” in my post… It isn’t just the grammar.

By: Eric Lund

Eric Lund — Thu, 14 Jul 2016 16:24:55 +0000

Seriously, the grammatical issues alone tell us that the software Deutsch used is (by today’s Deep Learning AI standards) substandard.

But it isn’t just the grammar. The phrase, “to answer him the common room perched upon it,” among others, is getting into “Colorless green ideas sleep furiously” territory.

By: Mauro

Mauro — Wed, 13 Jul 2016 21:45:01 +0000

No. At least not yet. At the moment no computer can write so bad stories but gain so good marketing.

By: fherb

fherb — Wed, 13 Jul 2016 21:31:53 +0000

Ok: What Max Deutsch did, was really a simple job: Put some Harry Potters into a fly brain and look, what will happen.

Deep learning is not so extreme simple. If you hope you will get a better result, so you have to analyse a bit more: how could the data go through a human brain in order to bring a good result?

1) You have to divide the brain into separate parts. Which parts and which structure between these parts, this is the most important task!

2) A grammar part network has to sort the learning information into a logical lane.

3) Maybe there is an understanding task. A subtask could be to understand this as a picture story.

4) There have to be a logic level: Find the most valid value for a structure of all objects and all persons.

… and so on.

I think: All single steps of such a task can be solved by deep Neural Networks. But for Learning, you need also such networks to prepare the information as a lerning information. (Learning is one task. Creation of new content is another task.)

The next level, what should be not really complicate, is to connect these single system parts as a learning and content creation system.

Google shows step by step in the monthes of the last 12 months, that it is possible to bring software neural networks in a state of creation of new content. Maybe as a picture based on a set of start pictures or some information as words. Or as the input of a Go play.

What Max did, was too simple. But what Google does, that is fantastic and shocking in the same moment!

By: Brainstorms

Brainstorms — Wed, 13 Jul 2016 18:44:30 +0000

OTOH, some may find the computer’s output more engrossing reading… Certainly is unpredictable and commands more attention. (I’m dying to read the outcome of your excerpt!)

Seriously, the grammatical issues alone tell us that the software Deutsch used is (by today’s Deep Learning AI standards) substandard.

I wouldn’t write off the AI angle quite yet… There’s too much unexpressed potential there. J.K. Rowling may get to keep her job, certainly. But she may soon have some decent competition chasing her at it.