Andy Hood, VP Emerging Technologies at WPP, explains what generative AI is, what it can do and shares five key things that marketers need to know about it.
If you’re reading this article, you responded to work generated by an AI (artificial intelligence). The prompt entered into OpenAI’s GPT3 Playground was, “Write a fun SEO headline for an article about generative AI and marketing.” Once it had been inputted, the AI took about five seconds to deliver the header of this article. The image (“a conceptual image about generative AI and marketing in the style of a collage” typed into DALL-E 2) took a bit longer – around 20 seconds.
What Is Generative AI?
AI, as a rule, is simulating, in a machine, the same cognitive processes that we see in people. It’s a way of problem solving. And you probably use it in your daily life already without even realising – like when your smartphone enhances a badly-taken photograph into something worthy of the Gram or when you use Google maps to find the fastest route to your destination.
Generative AI is a type of AI that can generate content in response to prompts – in the form of words, images, video and even music. Instead of typing an image into the search bar of an image library and getting a menu of images that already exist, we typed an image description into Dall-E 2 and it generated a menu of images to match what we typed. In other words, it didn’t find images, it made them.
Generative AI is still in its infancy, but it’s poised to transform the marketing industry – although likely not the way you think. Here’s what you need to know.
1/ AI needs to be trained.
In the same way a human can draw a bus out of their head because they've learnt what a bus is and what it's not, AI needs to learn in order to generate a piece of content. It does this by absorbing data – lots of it! 2.3 billion (not a typo) captioned images were used to train the image generator Stable Diffusion.
2/ The problem is, training data can be biased.
AI can only learn from what's out there and what's out there tends to be influenced by human biases. In fact, problems with AI almost always comes down to the training data. We did some research with Lambda Labs’s AI generator StyleGAN 3, which can generate – almost disturbingly – realistic images of people, who don’t exist in the real world. What we found was that the AI was able to age the men beautifully, but women started to look more and more masculine the older they got. When we got into the training data, we discovered that the issue was not with the AI, it was the fact that the images of adult women used to train the AI were vastly skewed to the 20 to 40 age group, meaning the AI just wasn’t seeing enough older women, versus men, to be able to learn what women look like as they age. A big focus of ours at WPP is identifying and eliminating these biases.
3/ Content generation is more complex than it sounds.
Finding the perfect prompts that will get the AI to generate the concept that you're imagining in your brain is even harder than trying to find representative imagery in a stock library. For starters, the AI needs to have been sufficiently trained – if it’s never seen an image labelled “bus”, you’ll have no joy typing “bus” into your prompt. Instead, you’ll need to describe what a bus looks like, using image references the AI has been trained on. Which brings us to the second conundrum – if you’ve ever played a game like 30 Seconds, you’ll know that describing what you have in your mind vividly enough for it to be reproduced is not always as simple as it sounds. Get it right though, and your designers will never have to trawl through a stock image or music library again.
4/ Generative AI creates possibility.
More than anything else, generative AI is useful for executing cool ideas that used to be impossible. In the campaign Remastered Memories, Canadian charity True Patriot Love Foundation used AI to turn soldiers’ letters home into heart-wrenching works of art.
5/ The future is AI. But not like in your nightmares.
Generative AI is not coming for your jobs – nor those of your designers. However, that being said, designers will lose their jobs to other designers who know how to use the tools. That’s because generative AI will become like the Adobe Suite of the near future – a tool to make the creative industry faster, more capable and more efficient.
Originally published on BusinessLIVE.