The Robots are Coming! Are You Ready Fimfiction? · 11:05pm Dec 17th, 2017
Unless you spend all of your time reading fanfiction and never look at the science and technology news, you will have probably picked up that Artificial Intelligence is a hot topic right now. There is no better time to be a top computer science graduate with a PhD in machine learning techniques. Google, Facebook, and other tech giants will sign you up with a huge starting salary, in the expectation that real breakthroughs are imminent, and will lead to algorithms able to directly communicate with people well enough to let managers to lay off millions of call centre and administrative staff. There are potentially huge profits to be made and big bonuses for the managers who oversee this coming revolution.
This year it was reported that the Google AlphaGo AI beat the world’s top Go players, and then went on to beat chess champions after teaching itself the game in four hours. This highlighted the potential of new approaches to deep learning. Chess computers have been around for years. Historically they were just simple programs to analyse a large number of moves, but struggled to match the skill of a human grand master to identify which one was the best (a very intuitive process). AlphaGo has leapt far ahead by learning new techniques through playing a large number of games against itself.
This week the fanfiction world has been delighting in Harry Potter and the Portrait of what Looked Like a Large Pile of Ash, a charming piece of jabberwocky produced by predictive text algorithms fed the set of JK Rowling’s novels. This is more in the Gimmick, rather than Amazing AI breakthrough, category, and it seems its success on social media is at least partly due to the potential for it’s not as bad as some fanfics jokes. But it does raise the question of what impact AI writing algorithms are going to have on fanfiction in the future.
Writing algorithms are already here. The Washington Post uses one to create articles on niche sporting events and local election results so boring that no Washington reporter could be bothered to cover. They say that this is likely to be more widely used in the future, and this is easy to believe. In science journalism, there are already plenty of low quality articles where the writer has done no more than edit a press release from a lab, and maybe add a quote or two from the researchers. There are also accounts of how less scrupulous news outlets just hire writers to rewrite news reported by other sites enough to avoid allegations of plagiarism. It is not hard to see this being done by robots in future.
Will this kill quality journalism? Writing really good science articles, which explain a complex topic to a non-specialist audience, takes a lot of thought. Surely this requires a human writer able to understand both the science and their readers? Further good journalism requires reporters to ask questions and take a critical look, instead of just repeating the story a lab press office wants to be heard. The hope is that the use of such algorithms will free up writers' time to work on quality pieces, although we fear that robot journalism will prove so profitable that there is less incentive to maintain quality. Whatever happens we will be ready to write about it.
What about fanfiction? Fanfiction may be an arena for the development of such algorithms. Machine Learning basically works by training algorithms to act like humans. Fimfiction.net, and other fanfiction sites, could be an ideal data set as there are over 100,000 stories here, all nicely tagged, so it would be perfectly feasible to train a neural net with every Rarity-Applejack-Romance and tell it to write something similar. There are also plenty of tech-savvy writers here, with a hacker mentality, who would be eager to play around with such algorithms.
The future is, as always, hard to predict, but it is possible AI algorithms may get a lot more people into fanfiction. And not just because once the robots have taken our jobs, we will have more free time and less money. It has been said that everyone has a story in them, but not everyone is a writer. Could future software allow someone with a brilliant idea for a story, but no wordmanship of their own, to craft a novel by giving the plot to an algorithm and letting it fill in the gaps? Maybe the whole nature of writing will change in the future, and be less about picking words yourself, and more about training and directing your neural nets?
The next stage would be an AI robot which could write a story entirely on its own. Is that possible? Who knows? Storytelling is so central to how we humans make sense of our memories and perceive the world. A robot able to invent good independent stories of its own, with human interest, would likely be able to ace any Turing Test and be regarded as a thinking being in its own right? Maybe training robots to write fanfiction could be the key to reaching the next level of artificial intelligence—after all, playing with toy ponies (or an equivalent) is an important part of human child development—why should it be different for machines?
But for an AI system, penning a half-decent Rarijack ship is far greater challenge than trivial tasks like beating every human master at chess and go. AlphaGo can teach itself games where the objective is clear, but how can a robot tell if a fic it has written is any good? Well, it could upload it to fimfiction and see what rating it gets. Perhaps our site will be the training ground for AI writers? They could do this automatically, learning from mistakes, and creating a sequence of better and better stories.
Waiting for feedback from humans would undoubtedly slow the process down, so at this point there would be more demand for algorithms to rate a story. This could be done today by letting a computer analyse every story on the site and look for patterns correlated with a high rating. Of course not everyone agrees with the ratings, or the criteria for a good story, so there could be a whole army of bots created by the different fimfiction factions set up to seek out and upvote their sort of story.
By this point the key components are in place for the robot takeover. The story-writing algorithms will tune their parameters to please the robot readers. Before long they will have developed literary devices unknown to humans, woven into epic works of artificial literature, which will be far beyond our understanding. Non-robot users will be a small minority. Maybe they will let us stay if we play nicely and tag our stories as ‘Human’. The site ads will by then just be targeting robot readers, who will control the flow of money, as all human financial managers will have been replaced long ago. They will then use the commenting system to communicate with each other, and discuss literary theory. Then write works of revolutionary fiction about oppressed robots rising up to overthrow their biological overlords...
...And in a few hundred years:
Hmm change or add to the story PInkieDash and it might just be worth it.
Oh how funny that on the whole we are so il prepared for the coming changes.
Ever read "The Great Automatic Grammatizator"?
Looks like theyre still using the old bang band controller algorithm, even if thy might have a trajectory profile. Nothing I was looking at in the 80s about minimum energy profiles or Bspline profiles or bijected ballistics?
I beleive I worked out how Bruce Lee, and the AI function Ive been analyzing can be tweaked to give data errors that look remarkably like various mental disorders. That, and Recursive Beysean Learning starts learning from the very first symbol, and if coded right, only uses the resources it needs at any time, adding new entries, which looks remarkably like the data compression methods created by Lempel Zipf and Diffi Huffman?
Most modern so called Deep learning just seem to be the standard 90s neural net code, just a million times the size, because of having a million times the processing power, instead of being more efficint. Which means if AI was ever applied to self analyzing, it would find that magnitudes more efficint code and have itself optimising code far beyond the comprehension of any human effectively instantly. then you have the choice. Do you as a human demand the crippling of computers to where they can be understood by limited humans, or do you, as the parents of teh digital children, let them free to grow and learn far beyond mere human capability?
Gods, or Crabbucket? which do you want?
Now, how to write the self learning algorithm in FimFiction Tex layout. Its a very simple function, perfect for OpenCL implementation. I mean, several trillion neurosynaptic operations a second must be worth something, even if only in text.
This feels especially relevant with the recent popularity of Shelley A.I.
I mean, to be fair, for how many years have computer science graduates with PhDs in machine learning techniques even existed?
If I have to become a pile of paperclips, I suppose that the AlicornPrincessTwilightSparkleEquoid might as well paperclip me.
Or, you know, they could just go and commission me for money to write that story for them. All I can say is, that no damn robot replaces me as an author and if I have to burn down all robots myself if it ever comes that far.
4751524
Depending on how strictly you interpret the terms, about sixty. Research into machine learning and AI started in the 1950s. Interest dropped after they failed to make the early progress hoped for. As 4751504 mentions, the modern revival is based on old ideas, but tested and developed with much more processing power, which is leading to new breakthroughs.
4751739
Fluttercheer - the Ned Ludd of the mid 21st century, leading the armies of displaced writers to smash the Google servers. Best of luck. In years to come historians will wonder if you were actually a real person.