Why has nobody coded up an AI “article title generator” yet?
This seems to me to be one of the easiest ways to become fabulously rich. Here’s the idea.If someone with access to the right data wanted to, they could tune an artificial neural network (AI) to generate near perfect article titles.
How We Get Fabulously Rich
Modern media companies make their money based on clicks, which are farmed from social media. Classic model media organizations do still exist, but the definition of “classic” basically means “about to get rusty and fall apart at any moment.” The future of media is in the clicks. Most of the traffic comes from secondary sharing. Possibly, nearly all of it. And people share titles.
A staggering 59% of article shares happen without the sharer even reading the article. I literally just did it. That link? It’s to a Forbes article I’ve never read, and here I am linking it on Medium. I could have linked the Washington Post article on the same topic, but the Forbes one had a slightly better headline on my Google Search.
But now I’ve linked them both. Never read either.
Oh damn. Now we’re seeing how this works, in real time.
But that spills through. Watch the math. Let’s pretend we have a social network, such as Facebook, where each user has a couple hundred connections. Let’s say for the sake of argument that ten of those connections are tight enough to the user, and aligned enough with the ideology of the article, that they might be interested in sharing it. Six of the ten share it without even reading it. Then thirty six of the sixty users in layer two share it without reading it. Then two hundred sixteen….
…you get the idea. It’s a cascade.
Given that the title is responsible for over 50% of the shares, the title is more important than the body of the article. More important than the content. More important than the truth.
There’s a lot of actual money, when you do the analysis, tied up in having the perfect article title. We should be able to tune an AI to generate near perfect ones. For a refresher on how these things work, click here and skim to the second section:
You’d need some esoteric data though. You’d need a database of every article title you could find, as well as its time series traffic history. You’d need a language heuristics package to parse phrases into data models of phrases. You’d probably need more mysterious programming mammer jammer I haven’t thought of. And then you feed the titles into the input layer and tune it to click histories in the output layer. Once tuned, it becomes a traffic prediction engine, where you feed in a new title and it tells you the traffic that title probably gets you. And then, because “computer,” you feed it a kabillion randomly generated article titles, sort the kabillion titles by score, and filter out what makes any sort of sense inside the top 1%. Then write the article around the title. Because who cares, nobody reads the article anyway. (Are you reading this? HELLO! Help, I’m trapped inside a fortune cookie factory.)
Does Popular Science have one of these things already?
Our first problem: Big Media companies are the only entities with access to [article title vs traffic] data in enough depth to tune the model. So how do we get rich? Here’s our plan: We convince them to let us tune our model at AI Article Title Generator Incorporated (stock ticker: AIATG) towards their data. A tool like this could be worth whole percentage points in gross revenue to them. They’d pay out the nose. If Big Media doesn’t already have this, then it’s probably only because they don’t realize they need it. In particular, they need to get it before the other Big Media company down the street gets it.
I really don’t think it would be that hard to do, technically. We cook up AIATG Incorporated in Silicon Valley or Raleigh Durham, pitch to NBC or similar, and knock the project out in a year for a few million dollars, which is nothing compared to the potential payout. Big payout. Whole percentage points on NBC’s gross revenue payout.
But then the next big media corporation comes knocking on AIATG’s door, because they are getting killed in the (secondary sharing from title only) marginal traffic marketplace. Obviously we sell them one too, tuned to their data, for the same price as the first sale. The development for the second one is fantastically easy, because we’re just re-tuning the thing we’ve already built to different data. That’s almost a push-button exercise. We make bank there, and another client comes along, and another. And we all become fabulously rich.
But it doesn’t end there.
After AIATG Inc. has eight of these AI installations in place, we will also have eight times more actual data than any of the major media companies had on their own. So AIATG Media, the newly founded media wing of the company, can develop its own article title generator orders of magnitude more capable than all the other media outlets. AIATG Media begins eating NBC’s lunch by keeping all the best article titles for ourselves. AIATG not only has the potential to be a major media player, but to quite possibly destroy all legacy media.
It’s Bezos big. Way bigger than Musk.
And what are the other Big Media companies going to do? Come after us? Accuse us of manipulating the media? They’re all complicit in the scheme, and even if they weren’t, we have a weaponized article title generator far more powerful than they do, so we destroy them. In the media.
It’s not at all a hard thing to envision, if you look at the comps.
China is probably already got something like this running, related to their creepy social credit score. They’d do it like this. The higher your score is, the more loyal you are to the Chinese government. That’s almost literally how that score works. They’d map their processed article title database against social credit score fluctuations. Figure out what article titles make the people the most patriotic, or the most anarchic. All buried in the ANN, and predictive. They could control an entire population that way. And if there’s one thing the Chinese have got, it’s population. No more Tiananmen Square incidents if everyone’s loyal.
I wonder if this sort of project isn’t exactly why the Chinese invented the social credit score in the first place. Maybe they already have the AIATG.
But holy smokes, if that’s true then they could weaponize it against us. What would a weaponized version look like? What if they tuned it to anger and hatred, measured by some publicly available metric, or just some complicated balance of daily proxies for measures of patriotism or something. Tune it to that. But then they’d need a media network to deploy it, do disseminate article titles that destabilize a populace. They don’t really have such a media network in the English speaking countries.
But Russia does. Maybe Russia’s been doing this crap to us for a while. I’m not sure why they would have developed it over there. Were they thinking about this sort of idea back into the cold war? Computers were garbage back then.
I don’t think Russia has a system like this, because if they did then the stuff that came out of their social media botfarms in 2016 wouldn’t have been so laughably bad. But maybe they’re working on it.
And the last thought I have on the matter, is to wonder how highly an article entitled “AI Article Title Generator” would score in the AIATG predictive engine.
I should probably write an article like that, to find out.