AI will do MUCH more amazing things than that by 2049!

That's 31 computer years. It's counter-intuitive how fast computers progress because it doesn't have real-life parallels but there should be unbelievable change by that time. I'm in my 30s and remember a time when the internet was barely a part of life...now it's hard to imagine the world without it.

 

AI will do MUCH more amazing things than that by 2049!

That's 31 computer years. It's counter-intuitive how fast computers progress because it doesn't have real-life parallels but there should be unbelievable change by that time. I'm in my 30s and remember a time when the internet was barely a part of life...now it's hard to imagine the world without it.

I go further back to working on a dairy farm in the school holidays and waiting for the AI man to arrive to service the cows. This was when AI meant Artificial Insemination

 

And nobody will by read then. Seriously, will anybody even bother by then?

Or all the readers will be AI's scamming to put money into the AI programmer's hands. 

Uncertainty and ingenuity. Gotta love moraless capitalism. Or is that morassless?

Either way, I maintain that AI is limited by profit. And the total cost of the AI + AI Herder and hardware + support contracts must be less than the cost of paying a writer. And you can bet that the AI companies will want a cut of the action, earning cash based on profits, which is the current system. The idea of an AI writer sounds good, because a company gets to keep the IP (presumably, the law is fuzzy there), but you still have to implement a working system, pay the marketers, pay for a cover, pay the support staff, and advertise. The writer actually represents a small percentage of the take.

In addition, the AI may perform no better than a human writer. AI systems have already been invented that have no statistical advantage over humans. Humans do lots of things badly, so AI will matter there. Where humans do things well, AIs don't have any advantage at all.

 

What we need right now is AI that can PREDICT which book will become a bestseller. Publishers have been getting it wrong for years.

 

What we need right now is AI that can PREDICT which book will become a bestseller. Publishers have been getting it wrong for years.

This.

 

I had a conversation about AI with a HarperCollins editor, way back the 90's. (I'm a programmer by trade, I just write fiction for stress relief.) Anyway, she was looking forward to the day when a piece of software could go through the slushpile and throw out all the misses, keeping the hits.

I told her you'd have to train the thing to write a good book before it would recognise one, and then she wouldn't have a job. She changed the subject, and asked if I'd considered write fat Fantasy novels instead of SF comedy, and I said no.

On another related topic, I rarely meet anyone under 25 who actually reads for pleasure these days. It's all android/ipad games. I have 2 adult daughters, and the only thing any of their friends read is ... fan fiction, shipping characters in tv shows. Argh.

 

On another related topic, I rarely meet anyone under 25 who actually reads for pleasure these days. It's all android/ipad games. I have 2 adult daughters, and the only thing any of their friends read is ... fan fiction, shipping characters in tv shows. Argh.

But -- don't forget Harry Potter

 

Harry Po

But -- don't forget Harry Potter

I should have qualified my age range a bit more ;-) Kids under 12 are definitely reading, it's just that when they hit high school they're bombarded with classics and reading becomes yet another school subject.

 

AI, like fusion, is one of those technologies that has been massively overhyped and overpromised by it's proponents for decades. The miracles are always "just another 10 or 20 years away" but somehow we never quite seem to get there. Simple AIs created for specific, well-understood tasks tend to work pretty well (although I'll be a lot more impressed when my AI-driven opponents stop getting stuck against walls and such), but the more generalized you get, the less effective they are. The problem is that computers are stupid, and they only "know" what the programmer tells them.

In order to program a computer to do something, you have to thoroughly understand what you're telling the computer to do, and be able to reduce that to code that's simple enough to actually work without being a godawful, bug-ridden mess. Getting a computer to produce gobbledygook poetry isn't difficult. But think of all the things that go into a novel, and remember that almost every single bit of it is based on emotion. The writer is trying to evoke an emotional response from the reader, and do it in a way that is reasonable based on the story and that doesn't overwhelm the reader. How do you program a computer to create a character that's sympathetic to the reader? How do you program realistic human responses for characters that can cover every conceivable situation into a computer? How do you program complex human relationships into a computer? How do you program a computer so that it knows how a mother who has recently lost a child is going to react to a comment overheard during a business dinner made by one of her husband's coworkers? How do you explain that first fluttery moment of amazement when you realize you're falling in love? How do you explain the wild joy of standing out on the beach at night during a storm, or the warm, cozy feeling of walking into a kitchen to the scent of rolls baking in the oven that reminds you of your grandmother's house when you were a child?

We understand these things on a gut level because we are human and we've been soaking in the whole universe around us with all five senses every second since birth. How do you get all of that into a computer? Who understands characters and plot and scenes and dialog and all the other things that go into making a good novel well enough to be able to reduce that to a set of code specs that a team of programmers can get into a computer? Most of the time we can barely communicate those things effectively to each other, and even then people have wildly differing opinions on the best way to approach them. At best you'd end up with something like the very worst of the carbon-copy mass-produced fiction or more likely TV shows that would be banal and lifeless with no voice and no real connection at all with the reader. Because to program a computer to write a novel you'd essentially have to program it to be human and then infuse it with a lifetime's worth of experiences. Not only can't we do that now, we haven't the slightest concept of how to begin to approach the problem. Even if we did, it's the kind of project that would take decades even if someone started on the code right this moment.

30 years? No way. 130, maybe.

 

AI, like fusion, is one of those technologies that has been massively overhyped and overpromised by it's proponents for decades. The miracles are always "just another 10 or 20 years away" but somehow we never quite seem to get there. Simple AIs created for specific, well-understood tasks tend to work pretty well (although I'll be a lot more impressed when my AI-driven opponents stop getting stuck against walls and such), but the more generalized you get, the less effective they are. The problem is that computers are stupid, and they only "know" what the programmer tells them.

In order to program a computer to do something, you have to thoroughly understand what you're telling the computer to do, and be able to reduce that to code that's simple enough to actually work without being a godawful, bug-ridden mess. Getting a computer to produce gobbledygook poetry isn't difficult. But think of all the things that go into a novel, and remember that almost every single bit of it is based on emotion. The writer is trying to evoke an emotional response from the reader, and do it in a way that is reasonable based on the story and that doesn't overwhelm the reader. How do you program a computer to create a character that's sympathetic to the reader? How do you program realistic human responses for characters that can cover every conceivable situation into a computer? How do you program complex human relationships into a computer? How do you program a computer so that it knows how a mother who has recently lost a child is going to react to a comment overheard during a business dinner made by one of her husband's coworkers? How do you explain that first fluttery moment of amazement when you realize you're falling in love? How do you explain the wild joy of standing out on the beach at night during a storm, or the warm, cozy feeling of walking into a kitchen to the scent of rolls baking in the oven that reminds you of your grandmother's house when you were a child?

We understand these things on a gut level because we are human and we've been soaking in the whole universe around us with all five senses every second since birth. How do you get all of that into a computer? Who understands characters and plot and scenes and dialog and all the other things that go into making a good novel well enough to be able to reduce that to a set of code specs that a team of programmers can get into a computer? Most of the time we can barely communicate those things effectively to each other, and even then people have wildly differing opinions on the best way to approach them. At best you'd end up with something like the very worst of the carbon-copy mass-produced fiction or more likely TV shows that would be banal and lifeless with no voice and no real connection at all with the reader. Because to program a computer to write a novel you'd essentially have to program it to be human and then infuse it with a lifetime's worth of experiences. Not only can't we do that now, we haven't the slightest concept of how to begin to approach the problem. Even if we did, it's the kind of project that would take decades even if someone started on the code right this moment.

30 years? No way. 130, maybe.

That's actually hilariously false.

AI has had mind-blowing breakthroughs one after another. I remember my father just a couple of years proudly stating that cars will never drive without humans behind the wheel. We are already finding that car AI is testing safer than any human and will soon be everywhere. Computers already dominate most defined contests like strenght, speed, and mind games like chess or the super hard Chinese game "GO".

It's usually the older generation who misevaluate new tech. and that's for good reason...they know very little about it compared to people who grew up with it or use it everyday.

 

That's actually hilariously false.

AI has had mind-blowing breakthroughs one after another. I remember my father just a couple of years proudly stating that cars will never drive without humans behind the wheel. We are already finding that car AI is testing safer than any human and will soon be everywhere. Computers already dominate most defined contests like strenght, speed, and mind games like chess or the super hard Chinese game "GO".

It's usually the older generation who misevaluate new tech. and that's for good reason...they know very little about it compared to people who grew up with it or use it everyday.

Um. I spent 25 years as a software developer and did my master's thesis on pathfinding AI. Try again. Games are trivial because they're designed around a very limited set of rules and moves and the computer can brute-force its way through them. Of course the more complex the game, the more poorly the computer AI performs. We've all seen that in games we play. The developers typically offset this by giving the computer an edge (faster production, stronger units, etc.). Car AI is still in its infancy and is in no way a proven concept in general use. Under controlled, limited conditions it works most of the time - until the computer runs into something that the developers didn't think of, or conditions aren't what it was programmed to expect. When that happens... Well, I don't want to be riding in it when it does. And it's still a trivial problem compared to writing a novel that anyone would want to read.

 

Um. I spent 25 years as a software developer and did my master's thesis on pathfinding AI. Try again. Games are trivial because they're designed around a very limited set of rules and moves and the computer can brute-force its way through them. Of course the more complex the game, the more poorly the computer AI performs. We've all seen that in games we play. The developers typically offset this by giving the computer an edge (faster production, stronger units, etc.). Car AI is still in its infancy and is in no way a proven concept in general use. Under controlled, limited conditions it works most of the time - until the computer runs into something that the developers didn't think of, or conditions aren't what it was programmed to expect. When that happens... Well, I don't want to be riding in it when it does. And it's still a trivial problem compared to writing a novel that anyone would want to read.

You are saying things that are blatantly false.

Chess and Go aren't "limited moves" there are estimated 10 to the 120th power of lower bound game tree complexity in chess there are only an estimated 10 to the 81st power of ATOMS in the UNIVERSE.

So no, you are wrong.

 

You are saying things that are blatantly false.

Chess and Go aren't "limited moves" there are estimated 10 to the 120th power of lower bound game tree complexity in chess there are only an estimated 10 to the 81st power of ATOMS in the UNIVERSE.

So no, you are wrong.

That's completely irrelevant. The problem domain is very limited in scope - small board, limited number of pieces that can only behave in certain well-defined ways. How do you think those games work? You make a move. The computer sorts through all the possible next moves then does some analysis to pick the best one. In the more complex systems it will have lists of "best possible" moves already saved for that particular situation, and some will run the computations several moves ahead. It's why they need supercomputers to "out-think" human grand masters. It takes a lot of processing power to crunch through all of that. The basic programming isn't that difficult; it's the number crunching that requires the effort. Because of the very limited nature of the game it's possible to create software that will play a "perfect" game.

But when you start developing more complex games with a much larger problem domain, the AIs don't perform nearly as well. Because they don't think, they just do what they're programmed to do. And the more complex the behavior required by the game, the worse they perform. Most of us have seen enemies get "stuck" or wander off in some bizarre fashion because their pathfinding routines had a breakdown. Again, the developers compensate for the fact that AI players aren't as good by boosting their base capabilities and making them stronger/shoot more accurately/produce faster...

AI driven cars aren't new. Pole Position came out in 1982. The problem domain isn't particularly difficult - basic pathfinding and Newtonian physics. The problem was getting computers to the point where you could build a reliable one capable of holding the necessary data which was small enough to work in a car, generate the necessary maps, and tie that in with the modern GPS system. The older GPS systems weren't accurate or reliable enough to use for this type of application. Computers powerful enough to do the necessary work weren't reliable enough to function for extended periods inside a car - reliance on mechanical rather than solid state systems meant every bump in the road had the potential to cause a system failure. That's engineering, not AI.

 

That's completely irrelevant. The problem domain is very limited in scope - small board, limited number of pieces that can only behave in certain well-defined ways. How do you think those games work? You make a move. The computer sorts through all the possible next moves then does some analysis to pick the best one. In the more complex systems it will have lists of "best possible" moves already saved for that particular situation, and some will run the computations several moves ahead. It's why they need supercomputers to "out-think" human grand masters. It takes a lot of processing power to crunch through all of that. The basic programming isn't that difficult; it's the number crunching that requires the effort. Because of the very limited nature of the game it's possible to create software that will play a "perfect" game.

But when you start developing more complex games with a much larger problem domain, the AIs don't perform nearly as well. Because they don't think, they just do what they're programmed to do. And the more complex the behavior required by the game, the worse they perform. Most of us have seen enemies get "stuck" or wander off in some bizarre fashion because their pathfinding routines had a breakdown. Again, the developers compensate for the fact that AI players aren't as good by boosting their base capabilities and making them stronger/shoot more accurately/produce faster...

AI driven cars aren't new. Pole Position came out in 1982. The problem domain isn't particularly difficult - basic pathfinding and Newtonian physics. The problem was getting computers to the point where you could build a reliable one capable of holding the necessary data which was small enough to work in a car, generate the necessary maps, and tie that in with the modern GPS system. The older GPS systems weren't accurate or reliable enough to use for this type of application. Computers powerful enough to do the necessary work weren't reliable enough to function for extended periods inside a car - reliance on mechanical rather than solid state systems meant every bump in the road had the potential to cause a system failure. That's engineering, not AI.

Actually, that's just not true.

Alpha Go wasn't programmed with any strategy to play Go (unlike many former game-playing engines). It was given 72 hours to play against itself and is currently the strongest force in the world and even the top humans are given a four dan (move) head start to keep it even remotely close.

https://deepmind.com/blog/alphago-zero-learning-scratch/

AI has changed and grown a lot since you were familiar with it(from the statements you've made). I've ridden in an AI controlled vehicle and it operated better than any human I've ever seen. Avoiding obstacles that appear out of no where in a fraction of a second.

Things are a changin...

 

I'm still waiting for a thousand monkeys at a thousand typewriters to finish the complete works of William Shakespeare. Once that's finished then I think AI will have its turn.

 

Actually, that's just not true.

Alpha Go wasn't programmed with any strategy to play Go (unlike many former game-playing engines). It was given 72 hours to play against itself and is currently the strongest force in the world and even the top humans are given a four dan (move) head start to keep it even remotely close.

https://deepmind.com/blog/alphago-zero-learning-scratch/

Again, completely irrelevant. The problem domain is simple and well-understood and easy to write code for. The fact that they used a combination of neural nets and a self-populating expert system to determine the best strategies by having it learn good/bad moves by playing against itself rather than pre-filling the data doesn't make the slightest bit of difference. It's still a simple problem domain that can be broken down to simple instructions which can be easily coded.

As you expand the problem domain, AI performance gets worse. The more complex the game, the worse the performance. This is a big reason why people prefer to play against human rather than computer opponents in complex games - the AI tends to be too predictable and easy to beat, even with modern computers that are orders of magnitude more powerful than were available even a few years ago. For simple, easily-defined tasks they work great. When you start asking them to do more complex things they aren't nearly as good.

And what you're describing with the car has a lot more to do with engineering than AI. It's fairly trivial to program what you're describing. It's a much, much, much more difficult task to engineer the physical systems which function reliably enough to make the car do what the computer is telling it to do well enough so that the passenger arrives in one piece. You're conflating the physical engineering with the AI running the systems. That engineering is orders of magnitude harder.

I'll repeat - in order to program a computer to do something, you have to completely understand the problem domain involved and you have to be able to reduce that problem domain into something that your programmers can actually work with. That's not hard to do with games like chess and go which have extremely limited rules, or with something that's fundamentally pathfinding and basic physics like moving a car from point A to point B. A single (decent) programmer who knows the fundamentals can put together a basic simulation for either of those in a couple of days. Won't be complete by any means, but good enough for a basic game or proof of concept.

There's a rather big difference between those things and doing something as nebulous and incredibly complex as writing a novel which has to tell a good story and convey human emotions in a way that will satisfy readers. Computers are good at moving things from point A to point B. But they don't have imagination and they can't do anything that they're not programmed to do. How do you program a computer to understand nuances of dialog? How are you going to program a computer that it needs to insert a bit of humor into the story at a certain point because by alleviating the tension just a bit there it will make the Big Problem about to hit in the next chapter appear even more appalling? Or know when not to do that because it will absolutely ruin the pacing? And that's just mechanics - it doesn't even get into how you're going to tell the stupid thing how to choose the best characters to fit the story, or how to come up with a story idea in the first place. Computers don't have imagination. They cannot do anything that they're not programmed to do. If the computer doesn't know what a beach is, it can't start a story with a woman walking down the beach remembering that last day spent with her husband on the beach before he died in a tragic elevator accident. (The computer would also have to understand what a woman is, what memories are and how they work, what emotions are, what a day is, the concept of time, what a husband is, what death is and why it matters, what an elevator is, and what an accident is. All for just that one concept for one isolated scene that isn't even a story, but just a fragment.

That's the kind of problem domain involved here. It isn't an 8x8 grid with 2 players pushing around 32 game pieces with only 6 piece types. The first fully-automated chess program was written in 1958. It took FORTY YEARS before a computer chess program good enough to beat the very best human players came along, despite all of the major advances in computer technology and AI that occurred in the intervening time. And that's just for a fairly simple game you can teach a child to play in a few hours. Writing a novel is infinitely more complex, and not trivially capable of being reduced to a simple set of rules that you can program into a computer. The problem domain is for all practical intents and purposes infinite.

 

Again, completely irrelevant. The problem domain is simple and well-understood and easy to write code for. The fact that they used a combination of neural nets and a self-populating expert system to determine the best strategies by having it learn good/bad moves by playing against itself rather than pre-filling the data doesn't make the slightest bit of difference. It's still a simple problem domain that can be broken down to simple instructions which can be easily coded.

As you expand the problem domain, AI performance gets worse. The more complex the game, the worse the performance. This is a big reason why people prefer to play against human rather than computer opponents in complex games - the AI tends to be too predictable and easy to beat, even with modern computers that are orders of magnitude more powerful than were available even a few years ago. For simple, easily-defined tasks they work great. When you start asking them to do more complex things they aren't nearly as good.

And what you're describing with the car has a lot more to do with engineering than AI. It's fairly trivial to program what you're describing. It's a much, much, much more difficult task to engineer the physical systems which function reliably enough to make the car do what the computer is telling it to do well enough so that the passenger arrives in one piece. You're conflating the physical engineering with the AI running the systems. That engineering is orders of magnitude harder.

I'll repeat - in order to program a computer to do something, you have to completely understand the problem domain involved and you have to be able to reduce that problem domain into something that your programmers can actually work with. That's not hard to do with games like chess and go which have extremely limited rules, or with something that's fundamentally pathfinding and basic physics like moving a car from point A to point B. A single (decent) programmer who knows the fundamentals can put together a basic simulation for either of those in a couple of days. Won't be complete by any means, but good enough for a basic game or proof of concept.

There's a rather big difference between those things and doing something as nebulous and incredibly complex as writing a novel which has to tell a good story and convey human emotions in a way that will satisfy readers. Computers are good at moving things from point A to point B. But they don't have imagination and they can't do anything that they're not programmed to do. How do you program a computer to understand nuances of dialog? How are you going to program a computer that it needs to insert a bit of humor into the story at a certain point because by alleviating the tension just a bit there it will make the Big Problem about to hit in the next chapter appear even more appalling? Or know when not to do that because it will absolutely ruin the pacing? And that's just mechanics - it doesn't even get into how you're going to tell the stupid thing how to choose the best characters to fit the story, or how to come up with a story idea in the first place. Computers don't have imagination. They cannot do anything that they're not programmed to do. If the computer doesn't know what a beach is, it can't start a story with a woman walking down the beach remembering that last day spent with her husband on the beach before he died in a tragic elevator accident. (The computer would also have to understand what a woman is, what memories are and how they work, what emotions are, what a day is, the concept of time, what a husband is, what death is and why it matters, what an elevator is, and what an accident is. All for just that one concept for one isolated scene that isn't even a story, but just a fragment.

That's the kind of problem domain involved here. It isn't an 8x8 grid with 2 players pushing around 32 game pieces with only 6 piece types. The first fully-automated chess program was written in 1958. It took FORTY YEARS before a computer chess program good enough to beat the very best human players came along, despite all of the major advances in computer technology and AI that occurred in the intervening time. And that's just for a fairly simple game you can teach a child to play in a few hours. Writing a novel is infinitely more complex, and not trivially capable of being reduced to a simple set of rules that you can program into a computer. The problem domain is for all practical intents and purposes infinite.

You show a complete lack of understanding for how these systems operate.

There is nothing "simple" about the programing of alpha go. In fact, what it's doing now was so unexpected we had top scientists saying an AI wouldn't defeat the top humans for at least 5-10 more years, if ever.

And NO computers don't play games worse when things get more complex they usually crush humans in those scenarios as they are calculating so many moves per second.

Your comment that you can teach the game to a child is cringy. We aren't talking about teaching rules. We are talking about making world champion level players. You can not teach a kid to play chess well with strategy in a couple of hours.

You should read up on CURRENT AI not stuff from five years ago.

 

Even if it's true, which I doubt, I'll probably be dead long before then anyway, so I couldn't care less.

 

You show a complete lack of understanding for how these systems operate.

There is nothing "simple" about the programing of alpha go. In fact, what it's doing now was so unexpected we had top scientists saying an AI wouldn't defeat the top humans for at least 5-10 more years, if ever.

And NO computers don't play games worse when things get more complex they usually crush humans in those scenarios as they are calculating so many moves per second.

Your comment that you can teach the game to a child is cringy. We aren't talking about teaching rules. We are talking about making world champion level players. You can not teach a kid to play chess well with strategy in a couple of hours.

You should read up on CURRENT AI not stuff from five years ago.

Computer programmer here who bangs his head against his desk whenever the AI topic comes up.

What Kelli is saying is spot on.

Don't confuse computational power for intelligence. The things you're describing are not AI.

It takes more than a very fast computer to write a book. It takes actual intelligence.

 

There are a limited number of things a human body can say or do and a limited number of emotions those actions can evoke. Once AI catalogues all these things and analyzes the relationships it stands to reason a manuscript could be written by AI based on those relationships, probably a lot sooner than human writers like us want to believe. AI may one day be better at evoking the desired human emotions than humans are, though not actually feeling them, of course, only mimicking them. Thanks for an interesting thread.

 

Wasn't there a thread like this already where some people analyzed all the famous novels and had the AI produce a sample story? It's certainly possible if you break down all known written works and analyze the heck out of each fragment so an AI could assemble bits and pieces into a framework novel.

Getting it to do original stuff that appeals to humans would be hard. Really hard. Someone mentioned the cost... people would do it as the ultimate challenge. Sort of like trying to write an AI that could pass the Turing test. We are fascinated by AI and keep testing the limits of what we can do with it.

 

Even when I WAS a voracious reader in my early 20s, I was an anomaly in my peers. Like at a dinner out, I was the only one with a paperback.

I don't think the reading for pleasure population has ever been a majority.....

Also, so far none of the other ways to enjoy a story (movies, video games, TV shows) have eradicated the book. We'll be okay.

 

I had a conversation about AI with a HarperCollins editor, way back the 90's. (I'm a programmer by trade, I just write fiction for stress relief.) Anyway, she was looking forward to the day when a piece of software could go through the slushpile and throw out all the misses, keeping the hits.

<snip>

What follows is in general and not directed at you Simon ... :

I too was a programmer for 40 years. Retired now and write for fun like you, except it is not as fun as I liked. Too many books to compete against. But it is fun to write anyway if you have a plan. Last plan did not work, so it is exciting to have a new plan ...

Anyway, the part of your post that is close at hand, if not here, is ... an algo could detect a well written book I think. But don't let it be the judge. It should spit out the elements the editor thinks is worth noting, how many good elements does the book have etc etc and condense the process of deciding for the editor, - a yes, no, or maybe output and list those elements it saw the editor needs to see and consider. Now the editor's process is made easier and a writer may get a better break. But an author could be getting a rejection letter from a SW bot ... creepy.

The editor can tell in the query letter or even the first six hundred words of the book if it is to be rejected or given more consideration. They just have so many to go through is the problem. And books probably get rejected 100:1 or worse. It is a culling process. What, in what they think, is going to work and make money.

AI in general I think is a misnomer. No machine can be intelligent like us. Just because we can map out and see patterns does not mean we think that way. We don't know how we think WRT intelligence. We do think emotionally. And our intelligence is rooted in that emotional aspect I think. A machine only mimics what we tell it to do, even figure some things out. That is not intelligence like us. I laugh when AI is touted as the next big aha. It is fun to write a story about an aspect of such things. But what we have in reality is a long way away from being real useful. Sorry Alexa ...

A machine will always be a machine in my book. Androids - really ... you really want a lap-top following you around? But it is practically that way now with kids and smart phones. Games arrrgh! But adults do the same thing now too.

I have a family gathering, and they all sit around and touch their smart phones or tablets. No one talks to each other or about anything important - it is funny to watch.

 

AI as we know it today always seeks the quickest, easiest solution to any problem or conflict. If you tell a robot with an equipped AI to go get you a milkshake from the McDonalds next door, it will find the quickest route to McDonalds...and crash through the window trying to get there.

Any author worth their salt knows that conflict or adversity cannot just simply be overcome. It must be a long/drawn-out process, usually getting worse before it gets better and finally ends. An AI with current AI mentality will write a story like this: "Boy meets girl. Boy likes girl. Boy gets girl. Love conquers all. Resistance is futile." It's over in one page.

Hilarious? Yes. Good story? Not really.

 

Anyway, the part of your post that is close at hand, if not here, is ... an algo could detect a well written book I think. But don't let it be the judge. It should spit out the elements the editor thinks is worth noting, how many good elements does the book have etc etc and condense the process of deciding for the editor, - a yes, no, or maybe output and list those elements it saw the editor needs to see and consider. Now the editor's process is made easier and a writer may get a better break. But an author could be getting a rejection letter from a SW bot ... creepy.

I agree this could work. But if editors got lazy and asked for a score out of 100 instead of a detailed report, that would be a problem.

Editors are human (for now), and they all have different tastes. You might submit your novel to 20 publishers and get 20 rejections, until one editor falls in love. Now imagine they're all using Microsoft Rejecto, with the same parameters. That outlier novel will never sell.

Plus think of the brits and aussies, who get a score of 5/100 because they can't spell aluminum or color.