With image generators, like DALL-E and Midjourney, able to generate images in distinct styles, from Picasso to Modernism, and artificial intelligent (AI) chatbots, like GPT4, able to produce competently written works, it’s easy to see how revolutionary generative AI could be for marketers’ productivity. At the same time, these significant advancements in automation have many creators and artists fearful as they wonder how their careers could be impacted.

According to a recent Goldman Sachs study, as many as 300 million jobs could be partially disrupted by generative AI across the US and Europe. Now, that doesn’t mean that all these jobs could be replaced by AI; it’s quite the opposite. The report estimates that “Some 7% of US jobs could be replaced by AI, with 63% being complimented, and 30% being unaffected by it.”

Just like the many technological advancements we’ve seen change the way we live and work, such as the introduction to the internet and smartphones, generative AI could have a similar impact. How marketers work and connect with customers is going to change, but to what degree? In this article, we are going to examine how AI is currently affecting marketers’ roles and predict whether it has the potential to completely automate creative and strategic tasks. Let's begin by analyzing the accuracy of generative AI platforms.  

Generative AI: How Does It Work?

To determine the accuracy of generative AI platforms, we must understand how these platforms work. Their foundation is built on search engines. Just like how Google crawls hundreds of millions of pages using their own web crawlers to index the internet, ChatGPT crawls websites similarly in a phase known as pre-training. In pre-training, ChatGPT analyzes lines of text and breaks these words and phrases into billions of tokens. These tokens are then clustered, enabling the platform to learn patterns and relationships between different texts. This gives it the ability to predict what response should come next in any given sentence, giving the user the illusion that it has the capability for human-like thoughts.  

To ensure ChatGPT can be as accurate as possible when predicting responses, OpenAI is constantly putting the large language model through pre-training. Pre-training can either be supervised or unsupervised, with the difference being dictated by the corresponding outputs.

In supervised pre-training, ChatGPT is trained on a labeled dataset with desired outputs. Engineers then test it, seeing if ChatGPT is able to give the desired predicted response based on the dataset it analyzed. If it doesn’t, OpenAI’s engineers continue to optimize the platform and finetune the algorithms that make up the platform.  

With unsupervised pre-training, the opposite occurs. ChatGPT crawls the internet, but there’s no specific output being dictated. The platform is making its best prediction on the information you want based on the billions of data points it feeds into its deep learning neural network, or in simpler terms, its large database.

In both scenarios, ChatGPT still attempts to cluster related data and detect anomalies, but with no human intervention, it’s at risk of clustering information that shouldn’t be grouped together. And as a result, it sometimes makes the wrong assumptions. In summary, ChatGPT can provide you with answers that are not fully reliable, but still speaks authoritatively like they are factually accurate.

Generative Content: How Accurate Is It Now?

In our last article on AI, we recommended that you only ask AI chatbots for content and answers that are within your area of expertise. These AI chatbot developers have a long road ahead of them to ensure accuracy. Take this example: Harry McCracken, the global editor of Fast Company, wrote that he asked ChatGPT the simple question, “What was the first TV cartoon?” He recounted that the tool gave him a different answer every time.

I’ve had similar experiences. For context, I am a huge fan of video game history. I even took multiple electives on it during my college years. To test ChatGPT, I asked it, “What was the first video game ever published?” Of the five times I asked it to generate an answer, it told me it was Tennis for Two four of the five times. For context, Tennis for Two is known as one of the first video games ever created, designed by William Higinbotham, that went on to be displayed at the Brookhaven National Laboratory’s annual public exhibition. You see, this game was never published as it never was sold commercially. The truth and the answer I was seeking was Computer Space, which is the first arcade game commercially available to consumers in 1971.

Even with these basic questions seeking one piece of information, ChatGPT often provides inaccurate answers. And it seems this will continue to be an issue for many years down the line because generative AI is not sentient. It’s limited by its programmed logic and the data it analyzes from the internet. It cannot think beyond that. So, the engineers who optimize ChatGPT are going to have to wrestle with what types of information it prioritizes as important and factual. That ruleset is dictated manually by those engineers.

Image Generation: How Will It Affect Designers?

In their current form, image generators, like DALL-E and Midjourney can craft amazing imagery, leading many to wonder how it could impact design positions. For a while, these positions seemed safe as these programs struggled to generate photorealistic images. They couldn’t realistically replicate human features. That is quickly changing with recent updates. Just last month, Midjourney released version 5 of their platform, which can now more accurately generate the number of fingers on people’s hands, which was a major limitation in version 4. Now, it’s becoming more difficult to point out the differences between AI-generated images and real photography.  

In fact, generative AI is becoming so good at generating photorealistic images that the global clothing brand, Levi’s, is starting to utilize computer-generated models to show off their newest threads.  

While image generation has already begun to disrupt many industries, marketers shouldn’t worry yet. AI, in its current form, is not able to make all the strategic decisions you need to make surrounding your brand and campaigns. Synchronizing all the different elements that make up your brand is too complicated for image AI programs to consider, and still will be in the near future. 

From a design standpoint, these programs cannot provide working files that break down the unique design elements in a generated piece. So, if you are trying to get a specific image into your brand colors and style, it isn’t going to come close to providing the unique touch your designers will need to incorporate into their designs. Generating AI images can be useful for producing ideas and speeding up the layout and composition process, but you will still need designers who are able to comprehend what makes your brand unique and create and modify designs that communicate the message you want to be conveyed to your audience. 

Let’s Answer the Question: Could Generative AI Completely Automate Creative and Strategic Tasks? 

Here at Lev, currently, we don't think so. We believe generative AI has a long way to go before it could replace jobs, particularly in marketing. 

Why? To summarize, generative AI platforms: 

1. Are currently not capable of human problem-solving and decision-making. They can only refer to previous data found on the internet or users provide them. Marketers still will need to customize the works generative AI provides to meet their customers’ expectations. 

2. Are still prone to providing false information. So, if you want to use generative AI, we recommend you ask it to perform tasks that you have strong expertise in. 

3. Can’t produce imagery that takes all your brand standards into account. And with image generation tools, such as Midjourney and DALL-E, not able to provide working files, you’ll still need designers to incorporate precise edits to assets.  

For these reasons, marketers and content creators are going to be needed for a long time unless there are dramatic advances in these technologies within the next few years. But that doesn’t mean AI can’t be useful currently. Generative AI could help you be more productive and personalize customer communications, but when you do utilize these exciting new tools, ensure you are using them responsibly! 

Looking for more ways you can utilize innovative technologies to level up your marketing and customer experience? Reach out to us today to discover how we can help you bring your strategy, data, and technology together in a way that breaks down informational silos that are preventing you from crafting dynamic, personalized experiences.