Isaac Asimov Solved the AI Slop Problem For Us In 1953
By Mitchell R. Sowards (#AIAid)
In his 1953 novel The Caves of Steel — and in a whole series of later novels — Isaac Asimov introduced us to the character R. Daneel Olivaw. The “R” stood for robot.
For, you see, in that imaginary world, robots were humaniform beings with “positronic” brains. They were so ubiquitous, so intelligent, and so nearly indistinguishable from biological humans — even having human-sounding names — that society imposed a simple rule: all such robots must have their names prefixed with “R.” and should always introduce themselves using that prefix.
Without giving away the plot, robots in Asimov’s world did not yet have legal rights. It was therefore important — morally, socially, and practically — that people know what kind of “person” they were interacting with and what level of respect, obligation, and caution the interaction required. In addition, robots were governed by those famous Three Laws of Robotics, by which they were strictly programmed to abide.
The solution, at least temporarily, was transparency.
Having failed or delayed dealing with all the deeper moral issues surrounding human-appearing and human-smart robots, Asimov’s imaginary society solved the immediate problem with that prominent and ever-present “R.”
I believe we are in a similar situation today.
We cannot yet decide whether AI is good or bad. We cannot agree when it is acceptable to use AI and when it is not. And in particular, regarding so-called “AI slop,” we are suffering from a strange tension: nearly everyone seems to want to secretly leverage AI writing, while nearly everyone is also tired of being buried in a tsunami of bland, uniform, suspiciously familiar content.
We find ourselves not respecting the writing of others because we suspect it did not spring from their own intelligence, judgment, or creativity.
And then there is the matter of trust.
AI hallucinates. It makes factual errors. Sometimes it simply invents things. So even when we see something written, published, or posted by a person we know, we may not know whether we can trust it because we do not know how much of it was AI generated.
I propose that we solve this problem — at least temporarily — the same way Asimov’s imaginary society did.
With plain transparency.
Every composition should carry a marker indicating the level of AI involvement in its creation. Every LinkedIn post, every newspaper article, every book, every magazine article. Every byline should have a tag.
In my own experience writing both with and without AI, I have identified seven different levels of AI participation in the process of creation. Here is the tagging framework I propose, from least to most AI involvement:
Tag Human/AI Role
#AIFree 100% human authored; no AI used in composition
#AIAid Human authored with minor AI compositional assistance
#AIEditor Human authored with modest AI editing, restructuring, or added material
#AIGhost AI ghostwritten — human trained, AI composed, human approved
#AICollaborator Human and AI contributions close to equal
#AIPrompted Human directed, but AI composed using AI knowledge
#AIGen 100% AI generated with only a kernel or idea provided by a human
——————————————————————————————————————————-
#AIFree
For the purists among us, #AIFree is the tag which assures readers that the author or authors have not used AI tools to assist in the composition of the work.
Of course, such authors may still use AI for other purposes such as research, computation, or analysis. But the composition itself should remain nearly 100% human generated.
#AIAid
I quite often envision a topic on which I desire to write using my own thoughts and conceptions, and then compose it from beginning to end, striking every key on my keyboard myself.
But then I may ask an AI to help me “smooth and strengthen” the piece without modifying the content or unduly diluting my unique style.
To qualify at this level, the difference between the original and AI-assisted work should consist almost entirely of punctuation or grammatical corrections, minor word choices, and perhaps a few improved turns of phrase. A composition at this level should remain perhaps 90% or more the original human author’s work.
For reference, this very article falls into the #AIAid category and explains why my stylistically heavy use of parenthetical thoughts got translated into emdashes!
#AIEditor
On occasion, when writing for a publication, I have submitted one of my works and then had a human editor “re-work” the piece.
They may make changes like those described in #AIAid. But in addition, they may improve sequencing, clarify the argument, provide additional data, or add a bit of prose from their own experience, background, or expertise to help illuminate the topic.
Or a human developmental editor may make suggestions for significant compositional improvement but rely on me to perform the rewrite.
When AI performs at this level, then #AIEditor is the appropriate tag. At most, the AI-suggested content should remain a significant minority of the final piece, perhaps no more than 30%.
#AIGhost
Some years ago, I wrote articles for the Forbes Technology Council online publication. After I completed my first published article, Forbes offered me the services — for a fee! — of a human ghostwriter.
I would meet with her, and we would discuss topics of interest and choose one. I would then train her — more like give her a brain dump — on the selected topic from my expertise. I would share my positions on topics where there might be debate. She would ask questions to improve her understanding of the subject and my posture toward it.
She would then do some of her own research and come back to me with a first draft. I would take that draft and re-work it in any way I felt necessary. We would have a final review together to agree on content and style before sending it off to the Forbes editors, who might make a few more changes.
Of course, despite the significant contribution and effort on the part of the ghostwriter and the Forbes editors, the article appropriately went out under my byline.
When an author uses AI in a similar fashion, then #AIGhost is the appropriate tag. Most of the knowledge and expertise being expressed comes from the human author, yet the AI ghostwriter may contribute some content from its own training and “expertise” in addition to performing most of the compositional work.
#AICollaborator
How should a human author tag a piece where their contribution of conceptions and expertise is roughly equal to that of an AI?
Or when the compositional efforts are roughly equal?
Or maybe when both expertise and composition are roughly equal?
I personally have had that experience. The human-only equivalent is when an article or book is published with credited co-authors. In scientific circles, even if one party does almost all of the composition but others shared in the research or experimental work, those colleagues are described as collaborators.
So, I believe #AICollaborator is the most appropriate tag when one of the collaborating parties happens to be an AI.
#AIPrompted
It is my belief that most of what people today describe as “AI slop” falls into this category.
A human “author” — using the term loosely — has a desire to speak or write on some subject. They compose a few short sentences or a paragraph of directives: the premise, what to include, what not to include, how long it should be, and perhaps the desired style. They submit that information as a prompt to an AI.
And the AI does the rest.
The AI does all the heavy lifting to produce the piece. The human contributes very little except the direction.
#AIGen
#AIGen differs from #AIPrompted when a human offers only an idea, or even just the kernel of an idea, and lets the AI do the rest.
For example:
“Re-imagine the Romeo and Juliet story as if it occurred between lovers from opposing Native American tribes.”
Or:
“Write an article explaining the relationship between the speed of light and time dilation.”
In such cases, the human imagination, judgment, and knowledge are almost completely absent from the resulting work. That kind of composition should carry the #AIGen tag.
A Few Questions Remain
This is a robust framework, but it still leaves a number of questions.
Should there be a default assumption when no tag is present?
It would be nice if we could assume that any work lacking a tag was #AIFree. But, alas, that ship has sailed. My belief is that tagless works will increasingly be assumed to be #AIPrompted, since that appears to be the most prevalent case today by a wide margin.
So, if #AIPrompted becomes the de facto default in the minds of readers, that may actually encourage authors to proactively use a different tag in order to earn — or retain — respect and trust for their work.
How should such a framework be implemented or enforced?
In Asimov’s world it was easy: the manufacturers of robots embedded the Three Laws into their programming. It is possible that today’s AI creators could do something similar by inserting hidden, unchangeable tags into metadata, the way cameras insert hidden data into digital photos or printer manufacturers insert hidden signatures into printed output. Publishers could then surface those tags at the time of publication.
But even then, who should decide on the appropriate tag?
The human author? The AI? The publisher?
More likely, at least for now, we will have to rely on the honor system. Human authors should voluntarily share the appropriate tag in their byline. If “everyone” is using AI, then there should be no shame in admitting how much or how little contribution to your work came from AI.
And those who try to fake it, claiming AI content as entirely their own, may find their work discounted or discredited.
Take credit where it is due. Give credit where it is due. That is how respect and trust can be earned and retained.
Regarding AI “Authors”
Finally, I am aware that some authors have years or decades of their own writing, research, speeches, presentations, and thinking which they can use to train an AI.
This is more than just a quick brain dump. This is the imparting of a human’s deep and unique knowledge, judgment, and style to an AI. In that case, it is possible to ask an AI to compose a new work based almost entirely on the training provided.
How should an author tag such a work?
I am considering adding an eighth level called #AIAvatar. It would sit just above #AIGhost and before #AIPrompted.
I am interested to hear what you think.
Does this framework resonate with you?
Would you consider adopting it voluntarily before it becomes any kind of accepted standard?
Would you expand or collapse the levels?
What do you think about the #AIAvatar level — should it be included or just combined with #AIGhost?
What are your thoughts on the assumed default of #AIPrompted and on adoption or enforcement?
One last Asimov-esque thought.
Just as today there are already social media “influencers” that are 100% AI generated — albeit sometimes with whole teams of biological humans producing them — there will surely be #AIAvatars authoring works under their own pen names.
Should we ask that their names be prefixed with an “A.” for instant recognition?
When that happens, should the byline be “by A. Daneel Olivaw”?
I am eager to see your comments.