Unless you’ve been living under a rock, AI-generated content has taken a massive step forward in the last few months alone. There’s already quite a big controversy around AI-generated images infringing on the rights of artists and illustrators, but in this author’s humble opinion, the real concern is how it applies to the content-marketing-industrial-complex.
If you’ve run any kind of business, you will know that content is king, and that content must be shareable, and quickly make it’s way through social networks to generate engagement and clicks. This online activity in turn influences search-engine result page ranking, which in turn results in more traffic.
For businesses that generate any kind of revenue online (be it through ads, affiliates or direct sales), traffic equals revenue.
The thing is, in order for companies to be competitive, they need a content marketing team, and that team will need writers, editors and graphic designers to make that content pop. That’s a minimum headcount of three people that you need to have on the books on top of your more traditional marketing team.
Content marketing is an expensive and tiring proposition – making any form of content is hard, blogging is probably the easiest and cheapest, followed by audio and then at the very top is original video content.
The thing is, a great deal of online content is in fact complete garbage, tossed together by an underpaid intern and then farmed out to an army of bots to simulate social activity and generate reach. Now I’m not talking about highly technical writing like what we do over at Tyk – where we need to have strong opinions about technical areas which inevitably get nitpicked on by an army of equivalent API-nerds. No, I’m talking about commodity content for your everyday stuff: generic consumer goods, drop-shipping outfits, e-book farms, and AliExpress resellers of niche tat.
To demonstrate: we can generate almost near-perfect generic corporate stock photos for use in transitory content, article fodder to provide more visual impact and imply a certain professionalism:
Images ideal for a story about inflation, or the stock market… or perhaps something to accompany a story about a companies great hiring process. This is free, non-copyrighted, and – because this style of content is so transitory – never going to be put under any kind of real scrutiny.
Or maybe we need to get scarier, an push some good-old Fear, Uncertainty, and Doubt, what if there’s been a data breach? Or we are fear-mongering about how kids playing Call-of-Duty are being groomed to become furries (they are not, this is a joke based on this ridiculous misinformation that made it into real news outlets):
So we have the images, now let’s generate some copy to accompany it. Copy-generating NLMs are becoming better and better – so much so that Amazon is thinking of replacing it’s first-stage recruitment process with AI altogether.
I mean, check out this intro I generated in a few minutes using contentbot.ai:
Call of Duty is a popular video game series that has been around for over a decade. In recent years, the series has seen a resurgence in popularity, thanks in part to its inclusion of a number of furry characters. These characters have become popular among the furry community and have helped to introduce new members to the fandom. The furry community is a group of people who enjoy dressing up as and interacting with anthropomorphic animals. Furries often congregate at conventions, where they can socialize and show off their costumes. The community has its own culture and lingo, and is often seen as being welcoming and inclusive. Some members of the furry community have criticized Call of Duty for its depiction of furries, arguing that it
Complete and utter tosh, thankfully ContentBot has content-topic policies that restrict certain prompts, for example – if I use the word ‘pedophile’ in the prompt I got a warning. So that’s something at least.
Unfortunately the genie is out of the bottle, a technically-savvy misanthrope could just run one of the open source models out there, and use that to generate all kinds of awful content. We’ve seen this happen before with Microsoft’s “Tay” and more recently with Meta’s “Galactica”. Tay could tweet some pretty hateful stuff, and Galactica could be used to make the worst racist commentary sound official.
The world is already at a stage where there is a massive degree of anti-intellectualism and distrust of expertise (something Carl Sagan warned us about all the way back in the early 90’s, and Isaac Asimov before that), with the advent of language models like GPT-3 and image models like Stable Diffusion, building out bot-farms that automatically spew propaganda will be a breeze for any state actor with a grudge – and best of all, they won’t need a troll-army anymore, they can almost completely automate the disinformation process.
Given that there is also an increasing trend of lower critical thinking in the western world, it makes it so much easier for people to be manipulated at the click of a button. No more need for a conspiracy, just a decent bot farm and some well-trained AI language models.
Sorry – my paranoia is peaking, back to content marketing…
For commercial purposes, generating mediocre click-bait and studding it with AI-generated photos all of a sudden makes content generation an automatic process that only requires minor editorial control, leaving firms (and agencies) free to literally swamp the already swampy internet with completely auto-generated sewage.
Oddly enough, the root-cause of this potentially coming to pass all comes down to another form of deep-learning: search algorithms, search ranking and social media recommendation algorithms.
If content discovery wasn’t so insanely valuable for businesses, and search didn’t promote the use of timely and increasingly throw-away content, but instead found a way to enforce some kind of editorial control over what it is recommending, then we may be able to bypass the festering cesspool we’re slowly turning the internet into.
Of course, someone will come along and create an AI to enable editorial control – something that it looks like Twitter was actively developing before the Musk takeover.
I mean, ultimately a machine (directed by a human intellect) will be telling swathes of people what to think, buy, and believe, and we’ll have real problems as individuals judging what is trustworthy, because so much of it isn’t.