Artificial intelligence (AI) text generators are able to spit out thousands of words before a human writer finishes making their coffee, checking their email, and daydreaming about their upcoming holiday trip. Yet it comes with a price.
Together with Erik Totten, as the content managers at Divante, we share a similar interest in AI solutions. Shortly after joining Divante and brainstorming for new marketing opportunities, we decided to do a small-scale research project on the current state of AI content tools.
AI-generated image: "The future of eCommerce"
Introduction: a failed experiment with AI content generation
Our faith in AI is not unfounded, which can be seen with the many machine learning solutions already finding their way to the market. You may already have encountered text checkers, like Grammarly, that improve your writing or even synthesized texts. You can find stock commentaries, sports news, product descriptions, or weather forecasts. Those texts are repetitive to some degree, and you can feed AI with raw data to quickly generate readable and understandable news that will be hard to distinguish from something written by an intern.
Most of those generators are based on GPT-3. According to Wikipedia, “Generative Pre-trained Transformer 3 (GPT-3) is an autoregressive language model that uses deep learning to produce human-like text.” It’s an open-source model widely used by the text generation platforms.
To be honest, I expected that we would be able to leverage AI to create a couple of independent blog posts to fill with generated content. Erik’s research on the generators proved me wrong. His conclusion on tests with Rytr and Neuraltext was that we’re not quite there yet:
“There are severe limitations (...) However, it’s probably wise for everyone to tinker around with these because there are some useful ways to use them. They’ll also only get better over time.”
He concluded that AI isn’t self-sufficient, and we decided to dump the idea of including it in our content marketing as a standalone tool. It’s limited. You need to carefully fact check the text and can’t use the quotes. It has some grammar and punctuation issues and the quality is suspect, not to mention the plagiarism issues. As long as you’re aware of that, it’s a good tool for brainstorming, trying to break writer's block, or generate ideas for PPC ads.
The byproduct of our research: interviews with AI
I personally prefer another aspect of AI text generation. Sometimes the outcome is simply hilarious. That’s why, knowing how underwhelming the results are, I decided to give it a go and task the AI with writing a story on the future of eCommerce. I did it for both amusement and research purposes.
Below, you’ll find four stories written completely by AI along with my short commentary. These texts are basically interviews with the AI. They give you real insight into how AI “understands” the language.
Please note, the following AI stories aren’t edited, and they aren’t chosen from a larger pool, contrary to many examples and blind tests you can find online. They are the first four texts spawned by Rytr.
My only input was a simple question:
“What will be the future of eCommerce?“
“In 2050, there were many different types of eCommerce. The first type was a storefront on a mobile phone. They contained a lot of new features that you wouldn't find at a brick and mortar store. One feature was that you can go into the store and scan items with your phone to see if it's available in your size, color, or both. Another feature is that you can purchase an item from someone else's wish list with just one click.”
A creative angle to approach the question, isn’t it? The paragraph of the tale from 2050 has a proper structure, the threads are linked, and the language is decent. It’s coherent to a certain degree, while it lacks the bigger vision and, in general, a point.
“The second type is called “eGifts”. They are similar to how traditional gifts operate because they are wrapped in paper, but they are delivered through the internet instead of door to door service. You can purchase them for almost any occasion which is convenient because not everyone has the time to stop by the store for every present they need to buy which will run out eventually anyway.
“The third type is called “aCommercial”. It was basically like using Instagram or Snapchat.”
The generator ran out of characters here. That’s too bad because I’d love to learn more on “aCommercial.”
It’s striking how AI-written text almost makes sense. A kid at the beginning of primary school could write an essay like this. A low-scoring SEO writer could spawn this, too. The thing is that you can still see that it does not understand the topic. It’s focused on making it look like an article rather than conveying a message itself. You could say that your AI was so occupied with what can be written that they forgot to ask why it should be written.
“As the youngest CEO in history, I was determined to change the world with technology. My company focused on all things technology related to eCommerce. I wanted to make it easier for people to buy online without needing to leave their house.”
You can see a lot of personality shining through this text. “The youngest CEO in history” must be obviously very young, but it’s credible and consistent in their style of writing.
“I had a great thought of an app that would help you find your perfect outfit for work or any occasion. The app would tell you what looks good on you and what does not. It would also recommend other items that might be similar to the shirt or pants you are looking at.”
Great idea for a startup. Maybe not unique, but still catchy. Stitch Fits does it well.
“I partnered with some engineers who created my app for me. They called it Personal Fashion Advisor and it gives you the best recommendation for any piece of clothing, shoes, or accessories that is not available online.”
It does the naming, too. That’s a proper name for an app.
“Ever since I made this huge decision 18 months ago, many big companies have tried to copy our idea but I'm confident they won't be able to create something as great as our Personal Fashion Advisor App!”
Now, this is an excellent final shout-out from, so the AI says, the youngest CEO in history. It’s probably too blunt to make it into any case study, not to mention that it’s fake, but all in all, it makes sense. It’s consistent in its style, and everything adds up.
“The notion of what will be the future of eCommerce is put in motion on a day when a customer walks into a brick and mortar store to purchase their favorite brand of snack.”
What an excellent abstract storytelling opening. I love it.
“They have been on their phone looking at reviews and prices and comparing to other brands. When they head to the product, they use a high-tech augmented reality that identifies the product then sets up its own price comparison. A new, different pricing table pops up for them, where they see the price of the product in their preferred store is less than 5% of the price on Amazon. In addition, they get a coupon from the local store with 20% off to make their purchase even more appealing.”
This text deteriorates a bit because it doesn’t really make sense to read reviews on your favorite brand of snacks. The pace of the story increases, and it all gets increasingly obscure. What’s more, selling stuff by blasting 20% off coupons might be a controversial strategy. However, the story still holds together.
“The next day this customer walks into another brick and mortar store and sees augmented reality screens displaying prices for competitor products. They make another purchase from this store because it's 30% less than what they saw in competitor stores and 10% less than in Amazon.“
What? You lost me there. This story feels a bit weird because it doesn’t lead us to any conclusion. If this is the future of eCommerce, it’s quite confusing. There’s no climax and no reason to it all, not to mention how cutthroat it is when it comes to pricing policy. It’s got that postapocalyptic nihilist flavor, you know?
“Life has changed so much over the centuries. From primitive tools to high-tech devices, there are many advancements to improve the quality of our lives. But one thing has remained constant: eCommerce.”
This absurd statement delivered with total confidence just cracks me up. That’s an AI comedy level straight from another synthetic AI Harry Potter and the Portrait of what Looked Like a Large Pile of Ash short story.
“What are the benefits of eCommerce? You have access to an infinite number of products, shop 24/7, and buy without even leaving your house. What are the disadvantages? You can never touch or feel the product before you buy it, shipping rates are expensive, and it's too easy to impulse buy items that you might not need or can afford.”
Makes sense. Even if it’s not relevant to the question.
“Will eCommerce still be a dominate form of shopping in the future?
As far as we know, this was just a thought exercise.”
Whoa, that’s a surprisingly evasive answer! It arouses suspicion. What are you hiding, AI? What terrible future have you foreseen? Why can’t we learn the dark truth? We’ll never learn the answers because we ran out of credits for generating text.
Summary: what to expect from modern AI text generators?
The current text generators have severe limitations, yet they can still create content that will be difficult to distinguish from the one written by a mediocre human writer. There are moments where it falls apart, and you can see that the text lacks a backbone based on sense and understanding. With a proper editing process, you could polish those spots.
This test was a hard one because we expected the platform to write a story. That was particularly difficult for AI. If you expect it to spawn eCommerce product descriptions, that’s an entirely different problem, and it’s much easier to tackle for the AI.
Still, we all have faith in the progress that will be made. With IT solutions, even if something isn’t excellent yet, you can count on it to get better over time. You need to keep an eye on what brings mediocre results for now but has a big potential in the future. It’s still worth it to come back to the AI generators every now and then to see how they develop. At some point, we’ll finally see acceptable quality, and the technology will be ripe for content creating purposes. It already is for some particular tasks.
As far as we know, this was just a thought exercise.
Published October 25, 2021