Dispatch from the Front June 5, 2023

Top

Revised 06/05/2023 Back to Contents

AI In the Apps:
Microsoft's Designer & The Existential Inference

I'm painting the room in a colorful way
And when my mind is wandering
There I will go
— Fixing a Hole, The Beatles

In the study of History, biases are neither good nor bad. The biases of historical figures are taken as a given, and personalities and events must be evaluated within their own time. We will gain a better understanding of the past by avoiding framing historical happenings within the language paradigm of our day.

As we have seen with the Editor, the simplest AI application will, nevertheless, demonstrate the limitations and inherent biases within a product based upon its underlying Language Dataset. The Editor denigrated Abraham Lincoln's reference to "All Men Are Created Equal," which was as progressive a statement as could be made in 1863, nonetheless, showed a distinct lack of sensibility to issues of inclusion that were not in the lexicon of Civil War Era America. The image output generated by Microsoft's Designer AI image generator also reflect the distinct joys and discontents of our culture and our time.

Microsoft Designer is Microsoft's answer to the OpenAI image generator application, DALL-E, reviewed here, February 24, 2023. For my own use, I now use Microsoft Designer almost exclusively for image creation. MS Designer on my system works faster than DALL-E. I am given a considerably greater number of free image generations by Designer than DALL-E. And, most importantly, I prefer the quality of the images generated by Designer over DALL-E. I think all AI generated images have a certain cartoonish patina about them. However, the images generated by MS Designer strike a balance between realism and being just plain corny.

I do not use any generated image by itself. I use the AI to generate elements within my own mental image of the final picture. MS Designer adds tools that make the images generated easier for me to work with. As is the case in all AI, the quality and accuracy of the textual prompt is key to getting the results you are striving for. So ready. The Wind Up. And here's the prompt:

A green apple with stems and leaves. A red worm is coming out of the top of the apple. The background is plain white.

And here are the results of the textual prompt. Designer initially offers up 3 choices of images to chose from.

I chose the third image for the demonstration purposes here. When you select one of the images offered, then click that image and the dimmed Select button will become illuminated.

One of the annoyances of Designer is that Microsoft offers several canned, eh my bad — preprogrammed — layouts for possible uses for the selected image. Posters, calendars, and other kinds of Hallmark inspired uses are displayed making it sometimes difficult to just find the image to download. Designer also offers a simple animated gif that looks identical in the image list as the single image file. Once the single image is found, Designer does offer useful tools to improve the download, at least for my uses. What you select here will to a certain extent control what optional designs are offered. I selected the image I want to download and the size, then the Designer formats the download accordingly. It is in the download function where Designer helps me the most.

No matter the degree of realism in the selected image, the images are not really photographs and therefore do not require the more photographic jpg file format. Jpegs are a choice though. So, I selected PNG file format to download. PNG files are smaller in size — always a plus when downloading. Also, PNG files take better to resizing than do jpgs. For my use, the ability to make the image background transparent is a key feature of Designer. My use for any AI generated image is as one element in a final composite image compromised of several different image elements.

Since I only use Microsoft Edge for Microsoft AI apps, the image to the left will appear in the top right of your Edge browser. The filename is always Designer (#).png. You can open the image from here, but that might not be useful.

What serial number Windows adds to the filename will depend on how many Designer generated files already exist in your Download folder.   It might be best practices to delete the unneeded Designer files in the Downloads folder. Those images that will be used should be given more useful filenames and copied into the respective Documents or Pictures folders where they will actually be put to use. That said, you must go to the Downloads file and click your newly acquired image.

Double click the Designer PNG file and enjoy the fruits of your labor.

My latest complex composite image was the creation below used in Regulating AI, Part 1. Every element in the composite image is AI generated, except the text. The clown on the left and the computer monitor are separate images that were generated by DALL-E in February 2023, before Microsoft Designer came online. The image on the right was generated by Microsoft Designer in May 2023. The idea behind the image was that AI before regulation is all flashy and full of hype; and after regulation AI will become cold, sterile, and, well, boring. It is the masked figure here that gets our attention and teaches something new about AI that has, in fact, been much talked about in the media.

Much of the hysteria surrounding AI has centered on its ability to make inferences related to the prompt, but maybe only tangentially related. AIs tend to add content that would not logically stem from the text of the prompt alone. In the image on the right, the AI made the decision to add the mask. The mask was not included in the original textual prompt, but rather was inferred by the text of the prompt. This leads to the existential question of does the AI's ability to make its own inference beyond the specifics of the prompt indicate the AI possesses its own intelligence? And thus the stuff of talk shows is made.

Here was the prompt: "A white man in a blue business suit with a white shirt and red tie. The hair is colored black and well groomed. One arm is raised. He is holding a glass laboratory beaker. The background is plain white."

How DALL-E interpreted the prompt is shown below. This is an unaltered image. How MS Designer interpreted the prompt is shown below. The image is unaltered. Empty space was cropped out.

The prompt asked specifically for a white shirt and a red tie. And I don't know what is in DALL-E Boy's beaker, but he looks like he has already had too much of whatever it is. DALL-E Boy is clean shaven, though. Overall, DALL-E Boy looks like a younger Arte Garfunkel had he not become a musical star, and was instead a shoe salesman who dabbled in amateur chemistry. As the little girl said, "Ah, Not My Circus. Not My Monkey."

BingMan has the necessary gravitas to what was the ultimate goal of the project. I cannot tell if BingMan is clean shaven or not. To me, "well groomed" in the prompt meant more specifically "clean shaven with a crisp Military Officer style haircut", and not a bushy head of well-coiffed hair This reinforces the need for the greatest specificity with the prompt. It also shows how once again the overarching language paradigm of the AI is very modern. My idea of "well groomed" reflects my own 1960s sensibilities and biases. Biases are not inherently bad as much as biases can frame the debate, and therefore must be understood. Next time I want a figure "clean-shaven with a crisp Military Office style haircut," I will plainly say so. Nevertheless, BingMan did answer the prompt's request for Blue Suit; White Shirt; Red Tie, to a tee. So taken all together, BingMan got the call.

Had Covid not occurred as it did, it is doubtful that the request for the figure to be "holding a glass laboratory beaker," would have prompted the double masking on BingMan's face. My intention for the breaker was to make reference to the hair tonic that was the theme of the Before AI Regulation image. This clearly is an example of how the AI can make an inference from the prompt that was never included in the text of the prompt itself. Also, the masks further demonstrates how the AI is trained on the lexicon of our time. Everyday people wearing medical style masks is certainly and exclusively of our time, at least not since the 1918-1919 flu epidemic, and is now considered usual and customary.

In the first image, the hair tonic was displayed as a bottle of hot sauce. I only prompted for a bottle in the Clown's hand. DALL-E has a preference for red when no other color is called for — sometimes even when another color is prompted for. The beaker in the second image was to contrast against the hot sauce, and meant to say: "Our tonic is just another lab solution in a glass bottle."

What I also find astonishing here about how the AI works is that the Designer had no way of knowing my full intention of the use of the image. And if it did know my grand overall scheme for employing its image — well that is a very frightening thought, indeed. Nonetheless, the cold sterile tone of the BingMan figure, highlighted by my touch ups, fit perfectly with the joke that the composite image is making: That AI might seem wild and crazy now, but after Federal and International Regulation, AI will seem like just one more thing. Do with it what you wish. Get too far out on limb, and the AI Police will knockdown your VPN and take your AI away. Bet on it!

And it really doesn't matter if I'm wrong, I'm right
Where I belong, I'm right
Where I belong

— Ibid.

¯\_(ツ)_/¯
Gerald Reiff

Back to Top ← previous post next post →

AI In the Apps: Microsoft's Designer & The Existential Inference
I'm painting the room in a colorful way And when my mind is wandering There I will go — Fixing a Hole, The Beatles
In the study of History, biases are neither good nor bad. The biases of historical figures are taken as a given, and personalities and events must be evaluated within their own time. We will gain a better understanding of the past by avoiding framing historical happenings within the language paradigm of our day. As we have seen with the Editor, the simplest AI application will, nevertheless, demonstrate the limitations and inherent biases within a product based upon its underlying Language Dataset. The Editor denigrated Abraham Lincoln's reference to "All Men Are Created Equal," which was as progressive a statement as could be made in 1863, nonetheless, showed a distinct lack of sensibility to issues of inclusion that were not in the lexicon of Civil War Era America. The image output generated by Microsoft's Designer AI image generator also reflect the distinct joys and discontents of our culture and our time. Microsoft Designer is Microsoft's answer to the OpenAI image generator application, DALL-E, reviewed here, February 24, 2023. For my own use, I now use Microsoft Designer almost exclusively for image creation. MS Designer on my system works faster than DALL-E. I am given a considerably greater number of free image generations by Designer than DALL-E. And, most importantly, I prefer the quality of the images generated by Designer over DALL-E. I think all AI generated images have a certain cartoonish patina about them. However, the images generated by MS Designer strike a balance between realism and being just plain corny. I do not use any generated image by itself. I use the AI to generate elements within my own mental image of the final picture. MS Designer adds tools that make the images generated easier for me to work with. As is the case in all AI, the quality and accuracy of the textual prompt is key to getting the results you are striving for. So ready. The Wind Up. And here's the prompt: A green apple with stems and leaves. A red worm is coming out of the top of the apple. The background is plain white. And here are the results of the textual prompt. Designer initially offers up 3 choices of images to chose from. I chose the third image for the demonstration purposes here. When you select one of the images offered, then click that image and the dimmed Select button will become illuminated. One of the annoyances of Designer is that Microsoft offers several canned, eh my bad — preprogrammed — layouts for possible uses for the selected image. Posters, calendars, and other kinds of Hallmark inspired uses are displayed making it sometimes difficult to just find the image to download. Designer also offers a simple animated gif that looks identical in the image list as the single image file. Once the single image is found, Designer does offer useful tools to improve the download, at least for my uses. What you select here will to a certain extent control what optional designs are offered. I selected the image I want to download and the size, then the Designer formats the download accordingly. It is in the download function where Designer helps me the most. No matter the degree of realism in the selected image, the images are not really photographs and therefore do not require the more photographic jpg file format. Jpegs are a choice though. So, I selected PNG file format to download. PNG files are smaller in size — always a plus when downloading. Also, PNG files take better to resizing than do jpgs. For my use, the ability to make the image background transparent is a key feature of Designer. My use for any AI generated image is as one element in a final composite image compromised of several different image elements. Since I only use Microsoft Edge for Microsoft AI apps, the image to the left will appear in the top right of your Edge browser. The filename is always Designer (#).png. You can open the image from here, but that might not be useful. What serial number Windows adds to the filename will depend on how many Designer generated files already exist in your Download folder. It might be best practices to delete the unneeded Designer files in the Downloads folder. Those images that will be used should be given more useful filenames and copied into the respective Documents or Pictures folders where they will actually be put to use. That said, you must go to the Downloads file and click your newly acquired image. Double click the Designer PNG file and enjoy the fruits of your labor. My latest complex composite image was the creation below used in Regulating AI, Part 1. Every element in the composite image is AI generated, except the text. The clown on the left and the computer monitor are separate images that were generated by DALL-E in February 2023, before Microsoft Designer came online. The image on the right was generated by Microsoft Designer in May 2023. The idea behind the image was that AI before regulation is all flashy and full of hype; and after regulation AI will become cold, sterile, and, well, boring. It is the masked figure here that gets our attention and teaches something new about AI that has, in fact, been much talked about in the media. Much of the hysteria surrounding AI has centered on its ability to make inferences related to the prompt, but maybe only tangentially related. AIs tend to add content that would not logically stem from the text of the prompt alone. In the image on the right, the AI made the decision to add the mask. The mask was not included in the original textual prompt, but rather was inferred by the text of the prompt. This leads to the existential question of does the AI's ability to make its own inference beyond the specifics of the prompt indicate the AI possesses its own intelligence? And thus the stuff of talk shows is made. Here was the prompt: "A white man in a blue business suit with a white shirt and red tie. The hair is colored black and well groomed. One arm is raised. He is holding a glass laboratory beaker. The background is plain white."
How DALL-E interpreted the prompt is shown below. This is an unaltered image.	How MS Designer interpreted the prompt is shown below. The image is unaltered. Empty space was cropped out.

The prompt asked specifically for a white shirt and a red tie. And I don't know what is in DALL-E Boy's beaker, but he looks like he has already had too much of whatever it is. DALL-E Boy is clean shaven, though. Overall, DALL-E Boy looks like a younger Arte Garfunkel had he not become a musical star, and was instead a shoe salesman who dabbled in amateur chemistry. As the little girl said, "Ah, Not My Circus. Not My Monkey." BingMan has the necessary gravitas to what was the ultimate goal of the project. I cannot tell if BingMan is clean shaven or not. To me, "well groomed" in the prompt meant more specifically "clean shaven with a crisp Military Officer style haircut", and not a bushy head of well-coiffed hair This reinforces the need for the greatest specificity with the prompt. It also shows how once again the overarching language paradigm of the AI is very modern. My idea of "well groomed" reflects my own 1960s sensibilities and biases. Biases are not inherently bad as much as biases can frame the debate, and therefore must be understood. Next time I want a figure "clean-shaven with a crisp Military Office style haircut," I will plainly say so. Nevertheless, BingMan did answer the prompt's request for Blue Suit; White Shirt; Red Tie, to a tee. So taken all together, BingMan got the call. Had Covid not occurred as it did, it is doubtful that the request for the figure to be "holding a glass laboratory beaker," would have prompted the double masking on BingMan's face. My intention for the breaker was to make reference to the hair tonic that was the theme of the Before AI Regulation image. This clearly is an example of how the AI can make an inference from the prompt that was never included in the text of the prompt itself. Also, the masks further demonstrates how the AI is trained on the lexicon of our time. Everyday people wearing medical style masks is certainly and exclusively of our time, at least not since the 1918-1919 flu epidemic, and is now considered usual and customary. In the first image, the hair tonic was displayed as a bottle of hot sauce. I only prompted for a bottle in the Clown's hand. DALL-E has a preference for red when no other color is called for — sometimes even when another color is prompted for. The beaker in the second image was to contrast against the hot sauce, and meant to say: "Our tonic is just another lab solution in a glass bottle." What I also find astonishing here about how the AI works is that the Designer had no way of knowing my full intention of the use of the image. And if it did know my grand overall scheme for employing its image — well that is a very frightening thought, indeed. Nonetheless, the cold sterile tone of the BingMan figure, highlighted by my touch ups, fit perfectly with the joke that the composite image is making: That AI might seem wild and crazy now, but after Federal and International Regulation, AI will seem like just one more thing. Do with it what you wish. Get too far out on limb, and the AI Police will knockdown your VPN and take your AI away. Bet on it! And it really doesn't matter if I'm wrong, I'm right Where I belong, I'm right Where I belong — Ibid.
¯\_(ツ)_/¯ Gerald Reiff