About once or twice a month, I go into Microsoft’s browser and connect to Copilot to see what has changed. One of my favorite things is to give the DALL E3 app a chore and see what it can draw with minimal instructions. Then, for the website, I sometimes need to reformat the image size using Photoshop, so I have two different AI programs that create and modify fictitious works.
The Dog Challenge
We have two female Chocolate Labrador Retrievers named Rylee and Cassie. Rylee is a rescue dog about six years old. Cassie is only sixteen months old, and we have had her just over a year. But Cassie is growing and just about Rylee’s size. So, I went into DAL E3 and gave it the instruction, “Draw two female Chocolate Labs named Rylee and Cassie.” The result was this picture, which looks enough like them to be a bit eerie.
But Cassie looks younger and is never without a tennis ball in her mouth. So, I went back to DAL E3 and modified the instructions a bit to make Cassie look younger and to hold a tennis ball. That produced another set of images that are also eerily accurate.
Finally, I instructed it to make them calmer and Cassie younger, and it produced this idealistic image, which I would always know is fake because Cassie has never been that still.
But this image is real enough that I could pass it off to most people as an accurate portrait. But unless you are in the fake dog portrait business this really has little application.
The Logo Challenge
For a bigger challenge, I asked DAL E3 to “Draw: Pillars of the Republic with an American Flag.” That nine-word request (and a few later tweaks) produced these images.
Simple one-word changes, such as adding “military” or “finance” to the end of the instruction, can produce amazing images.
Not Replicable
One problem I found is that the results were rarely replicable. Given the exact same instructions, the DAL E3 platform often produced entirely different results. The results were always fantastic, but for something to be usable, the results need to be produced over and over. For now I am going to chalk that up to a training issue on my part, but I suspect it is a real challenge.
A Plateau for Now
Many now realize that Artificial Intelligence will soon be doing some amazing things. But the challenge will be doing amazing things that are more than drawing pictures and rephrasing sentences.
We are learning of advances in medicine, security, and literature. For AI to become integrated into our world, it must produce things that corporations can monetize. This has been and continues to be the dividing line between new and exciting technologies and sustainable change. This engine drives us forward, and we only need to remember back to the dot-com bubble to know the difference.
I believe it is likely that the AI revolution will follow the gold rush path of the 1840s. Many will rush to the gold fields to make a quick fortune, but the real winners will be those who sell picks and shovels. In the world of AI, the picks and shovels come from companies like NVIDIA, which makes chips, and companies like Microsoft, Amazon, and Google, which provide large server farms. The rest will put AI in their annual reports with little understanding of what, how, or when to deploy it. Most CEOs know to say “AI” in their television interviews, but few know its meaning or possible impact.
I look for more hype and progress in specific businesses where the application and return on investment are clear. But AI’s takeover of the world will take much longer than most can see.
