The presentations at the Applied AI Conference held at the Northern Star Scouting headquarters in St. Paul, Minnesota had, appropriately, a Scouting type of “Be Prepared” theme.
Of course, the motto begs the question “Be prepared for what?”
At the November 10, 2023 conference, the there were two answers. One, the use of AI to get past the inertia of a blank page when starting a new project. Two, AI is already changing what is considered ‘normal’ for jobs like coding and design by creating time for ideation and enhancing the value of imagination.
The expansion of time for ideation and imagination fits with a predicted shift in the nature of work as the Industrial Era which started in the 1750s begins to cede ground to the new Imagination Era. In short, as predictable, and logical tasks are increasingly automated, in the Imagination Era the activities requiring human interaction, intuition, and creative thinking are moving forward as the new primary creators of economic value.
At the Applied AI Conference 2023, being prepared meant and pursuing imaginative options and launching ideation with Generative AI. An exploration of prompt design, or ‘prompt engineering’ was a common thread for in multiple industries including law, financial analysis, human resources, and digital agencies.
In this blog post we explore the processes around practical Generative AI prompts as well as how this approach to tasks shifts structural work processes, ideation and imagination towards this new era.
Converging the many attributes of an image into a series of useable images for an ad campaign via Generative AI is challenging and takes multiple interactions using AI prompts, according to Jason Tell, Chief User Experience Officer at the digital brand agency Modern Climate. In general, prompts are explicit written instructions to an AI tool, such as “Create a photo realistic picture of a farmer in a field of corn” used to produce outputs, including images, text, code etc.
On the one hand, the ability to quickly pursue random ideas when working with AI-generated images is inspiring, said Tell. So much so, that the ease of idea generation is a more valuable aspect of AI image tools than actually creating the final images, he noted. On the other hand, the idea generation can be frustrating because the AI image tools are evolving so quickly. The same prompts that worked well a week previously in Midjourney and DALL-E no longer work or provide different results. The prompts, said Tell “are evolving weekly.”
Tell said his agency was learning how to adjust prompts for creating images. For example, said Tell, adding adjectives like ‘energetic’ to prompts helped to add nuanced qualities to landscape images. The people generated by AI consistently arrived as young, thin, and without flaw or blemish - an unrealistic standard of beauty, according to Tell. Adjustments to prompts were added to more realistic characteristics, including adding freckles, uneven skin, weight, and other characteristics. Notably, said Tell, to create an overweight the AI-generated prompt for an image of a man required adding 100 pounds, but only adding 20 pounds created so-called overweight women. The implication was the training set of images used unrealistic and underweight women. He also showed how general prompts like ‘Create a scene from the Wild West’ often drew on stereotypes, like white men wearing white hats riding into the sunset, which needed additional prompts to generate more realistic images of the American frontier.
AI generated images, however, were dramatically lower in cost compared to live photography and provided more stylistic consistency when compared to pulling images from stock photography. Even so, legal issues associated with AI image copyright questions have put AI-generated images on the back burner when working with enterprise clients.
The agency, said Tell, has strict guidelines against using prompts like “Create an image in the style of (some famous style, such as a Norman Rockwell painting)” or the use of “Reference images” due to similar legal concerns. As an agency navigating an unsettled AI copyright legal landscape, Tell said, the agency rule was to generate AI images “described in our own words.”
Tell walked through a step-by-step process starting with a landscape, weather, and then adding people. This was followed by details like types of clothing and nuances such as the ‘focal point’ of the image. He noted that AI generated images often had a ‘drama’ to them and that getting a more photo realistic image required reducing dramatic elements the AI tools had added.
In a live demo, Scott Bromander, Chief Product and Technology Officer at YourPath riffed on building a Generative AI kiosk protype for use in zoos. Young children at a zoo will often ask their parents things about animals in the zoo that relate to the child’s experience like does a wombat also play Roblox, noted Bromander. Bromander’s design uses OpenAI Whisper and 11Labs so zoo-goers can ‘talk’ with zoo animals via an AI-power voice interface. The ChatGPT Developer API system key was used to set parameters at the kiosk so the AI interface for the zoo animal responded to inquiries in an informative and family-friendly manner.
In working with ChatGPT to code the zoo kiosk application he said the speed of coding was exceptional. What would take Bromander 30-minutes, and a junior coder 8 hours, took 5 seconds with ChatGPT, said Bromander. Despite his experience as a coder and coding teacher, said Bromander “AI codes better than I do.”
Moreover, he said, a rule of thumb in coding is coders spend 10% of time with ideation and 90% code creation. He estimated that using AI tools his work on the zoo prototype shifted closer to 40% ideation and 60% coding. Without the assistance of AI, Bromander questioned whether he’d have stuck with the project through the more repetitive and tedious coding that was required.
Instead, the tedious coding aspects were handled by the AI, reducing frustration with the passion project. Many passion projects launch with enthusiasm and then devolve into a grind and are abandoned. Using AI bends the curve and speeds up projects before enthusiasm wanes, he said.
AI was also helpful in kicking off projects and exploring new ideas. Working with the Gen AI helped create momentum by getting past the “blank page problem” where the effort to take the first step stalls a project. “It felt weird to not thank the AI,” said Bromander.
In conclusion, the Applied AI Conference 2023 in St. Paul, Minnesota, illuminated how Generative AI is a component an early foray into the Imagination Era. From Jason Tell's insights on enhancing AI image generation through nuanced prompts to Scott Bromander's innovative use of ChatGPT in speeding up passion projects like the AI-powered zoo kiosk, the conference showcased the diverse applications and benefits of AI in ideation and creation. Beyond the instrumental uses of AI, the ground is shifting from the Industrial Era to the Imagination Era. AI generating imaginative options, breaking through the "blank page problem," and accelerating projects. It is preparing individuals and markets with new ways to approach ideation and problem-solving in this new era.