AI, AI, AI! (as in ¡Aye, Aye, Aye! in Spanish)
Examples of elegant worthlessness from generative AI
I decided to give a new AI tool a try. It is difficult to ignore the “free” AI addons. I was heartened to learn in one case that these will be “free” until November 1 of this year. After that it will start to cost money each time the AI button is pushed.
I am a musician. I used to be better than I am now, but I still put in a couple of hours of noisemaking each evening while watching TV.
How would I start to play with my new AI tool while it still is free? How about starting with my instrument. Thus I put in the words “traditional silver cornet” as that is what I have in my hand each evening. The following four images were generated:
Admittedly, these are gorgeous, but they miss the point in each case. The idea is that you blow into the thing in one place and sound comes out in another place. All of these four examples fail in that regard. They are fanciful; they are comical; perhaps they have some value as abstract works of art; they are useless otherwise.
As a matter of reference, here is what a traditional silver cornet looks like.
As can be seen, one blows in one end and the sound comes out the other. This is a simple requisite of brass instruments. How is it that this simple requirement isn’t embedded in the AI tool? This is reminiscent of the recent report of an AI tool generating a legal brief, including fictitious citations (Merken, 2023). In defense of the ‘creator’, the brief did look legitimate at first. That was a part of the problem. Otherwise, it did nothing but bring trouble.
As to cornets, my son has said that he would like to get me a cornet or a trumpet with two bells, each pointed around and at my ears. He wishes for this as some form of recompense for his alleged suffering from my noisemaking habit.
It would be nice if AI could be used for this purpose. Due to its shape, a trumpet would be better than a cornet. As can be seen from another of my instruments, trumpets are longer and more sleek — hence, more popular currently. Here is a Bb trumpet of mine:
To be useful in the current high-technology manufacturing environment, a goal of a useful AI would be to configure desired product characteristics and then link to generative computer-aided manufacturing codes to actually create the item. This represents capabilities dating back to the 1980s and earlier. I wonder if generative AI would support this. As an enormous leap of faith, I wonder if resulting products would be even close to playable.
Here then is what shows up under “trumpet with two bells”:
These do not engender confidence. Playability in these cases might involve “tinkle, tinkle”.
Let’s go simple and work from there. I now put in “trumpet”. This is the result, along with a note that the query was too short:
I have no idea what that is about. As I have seen in previous AI showcases, context can be wildly erroneous in AI.
Here is the result from a slightly extended query with the string “Bb trumpet”. Adding “Bb” makes it clear that we are dealing with a music instrument:
This takes us back to weird. It is not clear with the last of the four images, which is incomplete, but none of the others would be playable.
Now, I will add the two bells question. When querying with “Bb trumpet with two bells” this is the result:
The first looks like a trumpet, but a close review makes it clear that it would not be playable, even if it could be built. There is no functional connection between tubing and the valves.
With one last attempt I put in “playable Bb trumpet”:
The second instrument from the left might be playable, but the valves are not clearly attached to the tubing in a functional way. The others represent nothing more than abstract art. It is a conundrum as to why actual trumpet players were obviously not involved in setting the parameters of this tool.
This is reminiscent of a popular television series called The Pretender (Mitchell and van Sickle, 1996–2000).
The episodes were about a bright young man that pretended to be people in a variety of professions and roles and succeeded in the effort.
The impersonations were impressive save one. The producers created an opening segment where the principal character acts out several roles with some success, but among them is his presumed acting as a classical symphony orchestra conductor. From a professional standpoint, he was simply flailing his arms in meaningless patterns. There is no way that would happen in a musical environment. They didn’t even try to do it in an authentic way, even though the show’s premise was this very thing. There was something about music that hadn’t earned the respect of the writers and producers.
This is unwarranted, over course. In each case, it is the people in the field in question, the experts, the practitioners, the legitimized journalists and critics, etc., that are best served to make such judgments and carry out programs and solutions.
Proponents of AI emphasize that their AI products are making decisions and reasoning on par with humans or “better”. Above is evidence to the contrary. It is important to note that the creators of AI have allowed such kinds of outcomes as we have seen in one small area into general use, covering all subjects. This is their product. They generate outcomes that pretend to be authentic, but are not. Such creators are responsible — in their own terminology, accountable — even if they don’t seem to want to take credit for their creation’s outputs.
References
Merken, S. 2023, June 26. New York lawyers sanctioned for using fake ChatGPT cases in legal brief. Reuters. https://www.reuters.com/legal/new-york-lawyers-sanctioned-using-fake-chatgpt-cases-legal-brief-2023-06-22/
Mitchell, S. L., and van Sickle, C. W. 1996–2000. The pretender. Created by Steven Long Mitchell and Craig W. van Sickle. Los Angeles, CA: Mitchell/Van Sickle Productions | NBC Studios | MTM Enterprises (season 1) | 20th Century Fox Television. https://www.imdb.com/title/tt0115320/