How to harness AI without being a rocket scientist

AI is defined (in the Oxford dictionary) as "computer systems able to perform tasks normally requiring human intelligence". This evokes a range of interpretations, from the mundane to the apocalyptic. From the extremely poor gameplay of the "AI" in your favourite turn-based PC game in the 90s, through to humanoid cybernetic organisms sent from the future to kill the leader of a human resistance (ironically from even earlier than the 90s), they all get the same label.

In this post, I'm going loosely with the definition: 

"The simulation of human intelligence processes by machines, especially computer systems”.

The current state of the art

Yes, some incredible things are happening in the world of AI at the moment. By now, we've all seen ChatGPT passing law, business, and medical exams from major universities, or artworks produced by Dall-E 2 that are practically indistinguishable from the work of a human artist (to most of us, anyway). Most major automotive brands are working on autonomous vehicles, and of course, it's been years since a human was able to beat a machine at most games.

But these advances are all the work of huge companies investing mind-boggling amounts into progressing the state of the art. I'm guessing your job isn't to teach cars to drive themselves.

AI for the everyday digital marketer

Your job as a digital marketer probably involves wrestling with huge amounts of content in various forms, on any number of different platforms, to get a message out to myriad customers via multiple digital channels. So how can, or should, you be using AI?

Let's loosely rephrase that big definition I gave up there ☝️ before:

"The simulation (by machines) of a human solving complex, repetitive, or time-consuming content problems."

Another way to think of this even more simply is "clever automation". 

Clever automation using existing AI microservices

You don't need to program AI yourself, or figure out how to use these bleeding-edge technologies dominating the news feeds, to change the way you work day to day for the better. 

Rather than trying to figure out how you can use ChatGPT to replace your entire marketing department, the key is to start much, much smaller than that. Below, I'm going to be talking about 'microservices' - designed to do one thing, and one thing only (and do it well). There are hundreds of AI-powered tools that can help digital marketers already out there waiting for you to plug in, today. I've collected a few examples to get your creative juices flowing below!

Sentiment analysis

Have you ever written an email while extremely grumpy, then re-read it the next day? How about a blog post? Or a speech? 😬 Getting someone to proofread your writing is a very good idea. But what if you're in a hurry? What if they're busy? What if you work alone?

Our robot friends have been able to do word counts, spelling, and grammar checks for decades, but did you know they can analyse more than that? AI services can be easily integrated to give you immediate feedback on your writing, from pulling out keywords as the robots see them, to analysing whether you're writing in a positive or negative tone, and even rating its 'readability', all in real-time.

At a recent User Group at Luminary's offices in Melbourne, we saw Brian Soltis (Cloud Solutions Architect from Microsoft in the USA, and a good friend of mine) demonstrate integrating sentiment analysis into his Kontent.ai project. In his example, an automatic webhook would notify Microsoft Azure's Cognitive Services, which would immediately analyse his post. If it was positive, it could be published right away. If it was too negative, it could be put back into a workflow step requiring review by the author. Very nice. 👏

Sentiment analysis may even be baked into your CMS already! But if not, there are services with APIs waiting to be connected.

Smart content recommendations

Have you ever had to maintain 'related articles' for hundreds of blog posts, both recommending them for new posts, but also going back and updating any old posts when a new article comes out that might be a good recommendation for readers of it? No, of course you haven't, because no one does. It's way too time-consuming.

If you scroll to the bottom of this post, you'll see a couple more recommended to you. They're not the same posts that will be recommended to someone else. If you don't believe me, send this article to a friend, and ask what they see 😉. Thank you, robot friend.

For our smart content recommendations, we're using Recombee to make it all happen completely automatically. In fact, I wrote a whole blog post about it here!

Machine translations

Not as good as an actual human, of course. But have you checked out the prices for human translation services recently? 🤯 What if your little robot friend could get it to 95% as good as a human within seconds, and then use the baked-in workflow functionality in your CMS to pass it on for review by said actual human?

Any headless CMS worth its salt that supports multi-lingual content will have excellent functionality for integrations, both outgoing (probably via 'webhooks') and incoming (probably via a 'Content Management API'). This means it's straightforward to set up a Workflow, where you can assign your article to a 'Ready for Translation' step, which automatically fires your article off to a robot for immediate translation. When the translation is complete, a serverless function can easily update the content of the language variants and push the article through to the next step in the workflow for final human review.

There are many machine translation services available now. Language translation is much less intense on computing resources than image generation, so you might find the pricing extremely compelling too. In fact, at the time of writing, Microsoft's Azure Translator API service is free for up to 2 million characters of translation per month. Quick maths suggests that's over ten thousand words of content per day, just within the free plan.

Image tagging

If you've ever reviewed the results of a WCAG accessibility audit, you'll know how often humans faithfully fill out the alternate or 'alt' text for images in a CMS. Some people will point out that AI doesn't always tag images with 100% accuracy. But do you know what's worse for vision-impaired users than having all of your images labelled with 99% accuracy? That's right... having them not labelled at all.

The benefits go much further than providing inclusive and accessible content to your visitors, however. Have you ever wanted to search for an image in your media library but couldn't remember the file name? Wouldn't it be nice if you could just search for what was in the image in your CMS asset library, just like you can with Google Photos?

Many enterprise Digital Asset Management systems have this feature baked in. Alternatively, there is any number of AI-powered image-to-text services, from the big players (Microsoft, Amazon, Google), to small, niche microservices. With how common 'image to text' projects are now, there are even plenty of examples of free, open-source code that could be utilised by the savvy coder.

Image generation

Ah, the bane of my existence... finding high-quality, relevant, and unique images to accompany my blog posts.

Fellow Kontent.ai MVP (and good friend of mine) Brian McKeiver set up an example of this using OpenAI's popular image generation service DALL·E 2, integrated right into this very CMS I'm using now. Here's a screen grab of it in action:

I'd like to thank my little robot friend once again for providing the hero image I ended up using for this article, which I've included again below.

"Albert Einstein designing and building a small rocket ship, blackboard with mathematical calculations in the background"

How do I use all of these services?

You might be lucky enough to have some of this capability baked into your CMS. But unfortunately, there's no way it'll have all of these functions available out of the box.

The good news is, all of these services have one thing in common: a MACH architecture.

MACH stands for Microservices, APIs, Cloud-native, and Headless. They are microservices, in that they exist to perform one function only (e.g. generating images), they have APIs meaning they are easy to integrate with, they are cloud-native, meaning you don't have to download, install, or host them anywhere, they're just sitting there waiting for you.

That's MAC; the only remaining piece is H, for headless CMS.

If you're already using a headless CMS, or indeed a hybrid CMS that can operate in an API-first, headless manner, then you're in luck! Look for AI microservices, that are cloud-native, and API-first. They're literally designed to be used in situations like just yours. You will need to do some work to connect the pieces, but in some cases (as shown in my image generation example above), you can do it with little-to-no code. And if it does require some coding from a (probably human, for now) developer, it's likely to be fairly straightforward, due to their MACH architecture.

If any of these examples sparked your imagination, reach out for a chat!