A while ago, I read the excellent book by Erik Brynjolfsson and Andrew McCafee, Race Against the Machine. Since then, I’ve thought a lot about automation, whether I’m looking at Google’s driverless cars or checking out at CVS through a self-service kiosk.
It’s a weird time to be alive and, if the common wisdom about robots these days is even half-accurate, this is the beginning. The next areas that most people think will be automated include legal work and healthcare. Shipping and manufacturing will continue to get Robotized, too.
That leaves us with an economy that increasingly relies on Creativity as a commodity. But, with new algorithms in place, even the writers and artists and musicians among us may start getting replaced.
The March of the Machines
Technology is universally claimed to be a force of good. At least by economists.
This is wrong. Technology is not good or evil. Humans are the moralizing force behind technology. And, in most cases, profit is what creates the tangible moral actions of humanity. That makes businesses look into ways to cut labor costs… but it also creates huge, huge pockets of opportunity for people.
Brynjolfsson and McCafee make two key arguments that shed light on technology’s impact on the job market today:
1. Economic productivity keeps on rising but, as opposed to years past, this does not coincide with a rise in available jobs.
2. Today, technology is moving too fast for workers to keep up.
Think about those automated, self-service kiosks you see at airports, grocery stores and convenience stores. Those used to be people. While they may transition to different fields of work in time, a lot of these locations have shut down opportunities practically overnight: you wheel in eight self-service kiosks, plunk them down and suddenly you don’t need eight employees at the cash register. You just need one to make sure everything keeps going smoothly.
But the overriding convenience – both for consumer and corporation – is far too tempting. Heck, I like being able to waltz into CVS without taking off my headphones.
Even the kiosks will be obsolete in ten years. When near-field communications (NFC) becomes the norm on smartphones, we’ll essentially be able to walk into a store and scan our own items and walk out. Retail outlets could very well become self-serve warehouses.
If you’re in the mood for some serious automation education – with a musical soundtrack – check out this five minute video by my band:
Word Smithing
Automation doesn’t just mean robots. It means advanced, learning algorithms. Customer service departments can be entirely outsourced to robots that can mimic human speech and just as efficiently give good answers. Without losing their temper.
These algorithms are even learning how to write books. Phil Parker, a trained economist, has figured out how to create one that can easily write genre books like thrillers and romance, because they all have certain, predictable formulas.
Sports-writing, too, can be automated, because of the limited vocabulary and content in each article.
Parker explains that, by creating algorithms that “mimic the human mind,” he has created over one million books. It’s not quite the horror story that writers have in mind, of course – he points out that the algorithm can follow a formula, which could replace authors who are mostly copying and pasting content. Readers have referred to Parker’s books as “awful and frustrating.”
This kind of thing is in its infancy, but it shows an emerging encroachment of machines onto the knowledge economy, too. That’s when things are going to get weird.
I think that, if we approach this new economy the right way, robots can become tools instead of replacements. Just take a look at the video above. Just the fact that it exists – and can be made and distributed – shows that technology offers some extreme potential. It’s just a matter of keeping up.
Blaise Lucey 
ahem ahem ahem. A few points here Blaise.
Wonderful article, thanks for finding this link to autowriting. Let me explain why I’m very glad of this autowriting, and why. Most artificial intelligence shows most of its proceeds (return) only a little bit into the subject. Consider that the difference to an amateur between the first chess programs and Deep Blue is almost nill (both beat amateurs soundly), and you realize that algorithms reach their effective peaks veerrrry quickly (for reasons too deep to go into here involving code-length and heuristic-types). So if this is a the best autowriting thing anyone can come up with (and so far, I saw no sources of it generating even a genre book that anyone bought and read to completion) that’s good, because it shows that even with billions of dollars poured in you’d get minimal return after this.
Consider a similar case where programming attempts peaked early – conversational bots are still not much beyond ELIZA, one of the first, a machine which uses a very simple strategy of using words you type and then asking questions about them to prolong a conversation. And ELIZA was done twenty-five years ago, I believe.
Additionally, this company is actually legally questionable – if I create 200,000 only mildly structured but mostly random, grammatically correct books and sell them as print-on-demand on Amazon, and for everyone one of those books at least 3 to 4 people will essentially be fooled into buying it (along with medical libraries that don’t distinguish) I’d say someone should sue this guy for essentially conning consumers. Using the guise of “algorithms” to essentially peddle random crap is at best dishonorable and sleazy and at worse should be illegal. He is essentially operating the same way as a chain letter – all you need is 1 bite of a million because there is no investment in the creation.
Anyways, writing is completely safe from automation in the foreseeable future – I really wouldn’t worry. As someone who keeps up on the AI literature, I can safely say that writing is like super-Go, a game that computers can’t even dream of playing. Consider that each word has to be placed appropriately, and there can be around 100,000 words per book – words are equivalent not to moves but to places. Go is far more difficult because it is an unconstrained game with a 19×19 (361) places, a stone can be set in any one at any time. Chess is easier, because it is 8×8, so 64 places – supercomputers can consider all possible future moves, due to the small board size. Writing would have 100,000 places – (a 1,000 x 1,000 board), making it so much harder than Go.
And that’s just the easiest foreseeable problem – chess and Go have well-defined goals, but writing – what is the goal? How do you rank sentences? If you could rank sentences, you could read. You must also understand what makes a sentence beautiful. This is called the Hard AI Problem in the AI literature and is well-recognized. To do task X implies task Y which implies task W and so on and eventually you recognize that you need a full artificial mind to do these particular tasks.
Anyways, writers will never be automated due to writing being a member of the Hard AI club. Here are some of the AI-Hard problems (solve one, solve them all).
1. Vision and other senses.
2. Understanding natural language (writing and reading).
3. Reacting to unforeseen circumstances and general planning.
4. Rationality and creativity.
Of course, it’s difficult to see how we could call an entity with those traits an automaton… if writing every gets “automated” it’ll be because someone cracked the AI-Hard problem and then it won’t but automation but speciation, and who cares if you are replaced by children you consider worthy?
Yes, I agree that writing is probably safe from automation right now, as I implied in the post, haha. And I don’t think we’re going to see robots writing, anyway… there probably isn’t any money in it.
However, I think you may underestimate the power of big data analytics & intelligent machine learning. You apply something like analytics to the human sentiment on social media (already possible) and to books and you’re not far off.
Machines are already writing sports reports… http://www.nytimes.com/2011/09/11/business/computer-generated-articles-are-gaining-traction.html?pagewanted=all&_r=0
Absolutely – my point was that most advances in machine learning blah blah (my relationship to AI researchers is the relationship of a physicist to an engineer, i.e., dismissive) move immediately to the peak they are ever going to get very quickly after the field is turned to them (saturation). Chess programs “peaked” very quickly, as have stock market programs – so my prediction is that it peaks at around sports/weather/financial data reports and never gets anymore sophisticated (nonlinear returns on investment). Same thing for almost all AI.
Anyways, I don’t underestimate big data or machine learning – I actually do work with machine learning algorithms and artificial neural networks – but all this stuff is just algorithmic tricks. To write a real news article (not just flinch and sort some data) is an AI-complete problem, and there is no indication that Big Data gets anywhere closer to solving AI-complete problems than GOFAI did in the sixties-eighties. All this stuff is just cheap Chinese Room tricks – basically a smart-sorting copy/paste machine. The problem is that people always project – if computers can do X, how soon before they can do Y? But people fail to realize exactly how different X and Y is, and how, by using cheap tricks, you can fake highly-constrained tasks, like reporting on the scores of a football game. If you read the article, you can realize this. First of all, it is not “writing sentences”, it is a MadLib program with a database of probably a couple hundred “stock sentences”, all relating scores to one another, broken up by quarters. It didn’t write those sentences, the programmers did. Secondly, it deals only with the relationship between scores – it plugs in the names of the teams and the scores and the yardage.
Imagine that one of the coaches in the game had taken off their headset and screamed at another coach – this poor dumb thing couldn’t have possibly known that – it doesn’t watch the game, it just takes the simple data of the scores and matlibs it into the preformed sentences. For example, this would never work for soccer, because to report on soccer the lengths of kicks and so on is not recorded – the reporter actually has to watch the game, whereas this madlib program is so far away from that as to be essential a cut & paste decision tree.
Anyways, I just don’t like this kind of AI hype, it’s been going on for sixty years and we have almost nothing resembling real AI, just really fast decision trees which, in hyper-constrained situations, are successful.
Well, we get into the idea of “what is writing” then… yes, Parker’s algorithm has “wrote” a million books, but people just thought they were written badly, not by a machine. So that distinction is already lost.
I work with a lot of big data analytics companies. It’s absolutely possible for algorithms to predict human behavior at this point, because all you do is aggregate data from multiple databases to create patterns of human behavior – is this customer likely to leave our service for a competitor after X happened? Well, what did one million people from this demographic do when X happened to them? Is this coach going to yell after Y happened?
AI hasn’t moved much, because there was nothing for computers to draw on to mimic behavior. Now, there’s the entire Internet, which is a database of human behavior the likes of which has never been seen – or sorted and laid out – before.
Image recognition software is going to easily take care of the “did a soccer player kick the ball” issue, too. Not only that, by monitoring tweets or whatever about the game itself, those can be inserted into the article itself.
As usual, though, this digresses from the main point… no, I don’t think robots are going to replace writers anytime soon, but we need to think about what kind of “writing” we’re talking about.