There has recently been a lot of attention given to AI. For some time now the usual doomsayers have been holding forth on the effects that this new technology will have, predicting the end of the human race and other pleasantries. Stable Diffusion is out in the wild now, open sourced and available to all to create images from a description. I’m getting it on to my computer so I can have a play with it.
Other AI’s, as detailed in this article are being used, among many other things, to create University papers which will get their students an A for the trouble of writing a brief prompt. The anonymous student defends his actions as as automating “busy work” so that he has more time to learn what he needs to know. There are those who would call it cheating. It slides right past the usual plagiarism checking tools.
More disturbingly, from the standpoint of an author, is an example of a book generated entirely by AI’s and illustrated by using the text as prompts. An engineer at Surge AI took a prompt from his five-year-old son and used it to generate a children’s book, with the text as prompts used to generate illustrations.
The usual nay-sayers are predicting doom, of course, pushing all the fashionable buttons of the day, such as bias and racism. Since these AI’s are created by human beings, and trained using information created by human beings, of course they also have biases. The art reflects the artist, and the tool reflects its creator. The tenor of the complaints I have read makes me think that they want the AI’s to share their particular set of biases.
Now, certainly, if the average reader who wants a potboiler to pass the idle hour, with the additional attraction of having the main character look like himself, can simply put in a prompt to get it, the old fashioned creative process from us human authors gets to look a little shaky.
I’ve been around long enough to see multiple technology revolutions, so I can take a longer view of the subject. So, first, we certainly have a very powerful tool here. My wife works with pathologists, who examine samples of tissue to decide whether, among other things, they are cancerous or not. Could that process be automated? I think it likely. Big data practically demands AI’s or something like them to generate usable knowledge from terabyte blizzards of data. There are a lot of applications where what was human work can be automated.
That has happened before. The printing press put scribes out of business, typewriters put clerks with pens to the necessity of learning to type or losing their jobs, and computers have automated an incredible array of information related functions. AI takes that to the next level. As we see with Stable Diffusion, it is well if human judgment is in the loop before the product goes out the door.
When one looks at the web sites which have sprung up to share the images produced by Stable Diffusion, the prompts to create those images get rather complex as they get more specific to what the creator wants the AI to produce. This begins to look like the early days of computers, when programming was found to be an absolute necessity to actually make use of the power of computers, and programming languages were needed to mediate between the humans and the microchips.
I’m looking at this to create cover illustrations for my books, and I rather suspect that I’m not the only one. There would be a learning curve, of course. The lawyers are already opening up their laptops and looking for the billable hours, too.
Looking ahead, what changes might we see here? We might see the role of the author change. Before I write a book, I do a lot of research and background information, delineating characters, creating background and history, inventing technology. To have the AI generate the novel, the author would have to set up all the prompts, input all the background data for the AI to use, tell it all about the style he wants it to use. That’s a significant amount of creative work. Then he’d have to go over the completed manuscript to look for errors and weirdness. There are some very weird images that come out of Stable Diffusion, and a lot of the ones posted on the sites have been worked over with Photoshop or GIMP to make them nearer the creator’s desire.
Even in an age of mass production and 3D printing, hand craft in such areas as jewellery still exists. Check out Etsy, and you find a vast array of people custom making all manner of things. Local craft fairs and farmer’s markets still exist, too. There might well be many people who would prefer a human-written book to the output of an AI.
Overall, then, what I see as an effect here is to push the human further up the scale, as has happened before. Engineers still create, but now they can build a model in three dimensions without the need of draftsmen, and substitute simulation for testing, which is a considerable economy.
Eventually, of course, models need to be checked against the real world, and that will still require the judgment of human beings.
In installing Stable Diffusion on my own computer, I had some problems to overcome. I run it in a separate container on my computer, for convenience rather than any fear that it might take over my computer or my life.
The fear that our own creation might supplant us, or destroy us, is a very old one. Mary Shelley tapped into it with “Frankenstein”, and its roots go back much further into myth and legend.
It was for a long time my business to measure threats accurately, to replace fear with knowledge. Here, then, is my assessment.
We have created another powerful tool, which is now out in the wild. Trying to control it is pointless waste of effort, and complaining about it will have no effect either. There have been no reports of any AI doing anything except run on its computer and apply itself to the problems its creators require of it.
A servant which can learn to do its job better is still a servant.
Be First to Comment