Volume XII: The Signal
The complete inquiry into authorship, provenance, novelty, and the Signal Protocol.
Media / AI & authorship
Authorship Panic: Finding the Signal Beyond AI Detection
Why statistical resemblance cannot determine authorship, novelty, or research value.
The program follows a page into an AI detector and shows which relationships disappear: provenance, direction, revision, evidence, responsibility, and the distance an idea traveled. It replaces the verdict with a multidimensional Signal Protocol.
Terms in this film
Explore the same term in other films through the transcript glossary.
Continue into the research
Related work
Volume X: CodeX
Documents human direction and machine abundance across a long AI research session.
AI Use & Research Transparency
The site's disclosure of AI-assisted research practices.
Read the video transcript
Published captions, with their original wording.
Okay, let's kick things off today with a bit of a modern paradox.
Can a machine accurately detect if a human wrote a text when both the machine and the human are trained on the exact same language?
Just think about that for a second.
We literally teach AI to sound exactly like us, and then we turn around and build detectors to penalize anything that sounds too much like the AI we just trained to sound like us.
It's a completely dizzying loop, isn't it?
Welcome to The Explainer.
Today, we're diving straight into an incredibly timely framework from Sidjay Hubbard's Causality and Attraction series.
Specifically, we are focusing on Volume 12, the signal.
We're going to fundamentally reframe how we view AI-generated text.
Our goal today is to move past all this current institutional panic and find a much, much more reliable way to actually evaluate human contribution in the age of artificial intelligence.
Here is our roadmap for today. We'll start with the AI authorship panic, then the limits of detection, third, human enclosures and responsibility, fourth, novelty over similarity, and finally, we'll wrap up with the signal protocol.
All right, part one, the AI authorship panic and how we actually define the problem.
Right now, you've got institutions, universities, and massive publishers who are absolutely terrified of losing accountable human work to automated generation.
But look, to solve a problem, we have to define it accurately first.
Hubbard introduces this really crucial term here, AI-assisted research content.
Notice what this highlights.
It's not about counting how many keystrokes the human physically made.
Instead, it's a research process where the AI participates, sure, but the human contributes the originating observation, recognizes a signal, refuses substitutions, and crucially takes full responsibility for the finished whole.
That right there is a legitimate research object, not just some active cheating.
Moving into part two, the limits of detection, or basically, are flawed instruments.
Notice how easily these statistical detectors break down.
An AI detector doesn't actually know a machine wrote a text, it's merely measuring statistical resemblance.
And just look at these variables, if you change the topic from, say, a history essay to writing code, if you paraphrase the output, or -- and this is a big one -- even if you're a non-native English speaker who naturally uses highly predictable, structurally safe sentences, the detector just fails.
A clear, concise writer is suddenly penalized for sounding too coherent.
The detector is basically caught between two moving distributions and it routinely gets it wrong.
To fix this mess, you'll hear a lot of people yelling about watermarks.
But let's look at this comparison.
A watermark intentionally injects a machine-selected statistical pattern into the prose, just so it can be found later.
It's literally altering the language to prove its origin.
Providence, on the other hand, is the actual historical record of creation.
Think of a time stamp, a repository commit, or a DOI deposit.
Providence establishes that a specific idea existed at a specific time in reality.
A watermark just tells you a machine touched the text at some point.
Providence tells you the true causal history of the actual idea.
Section three, human enclosures and responsibility.
This is the philosophical core of the argument.
This whole distinction between superficial patterns and true origins brings us straight
to Hubbard's philosophical core, the nested causal model.
It sounds complex, but it's basically a three-part structure.
On the left, the enclosing outer thing projects causality inward.
In the center, we have the departure.
That's the focal thing we're actually observing.
And on the right, the enclosed inner things.
When you apply this to AI and authorship, human freewill is the active enclosure.
Causality flows inward, maybe through AI-generated research,
but the human sits right in the center, actively choosing to redirect, edit, accept, or reject that flow.
The human takes the responsibility.
And this isn't just some isolated, abstract idea.
Volume six of the series thoroughly grounds this nested causal modeling
in a massive, mathematically, and philosophically rigorous continuum.
The main takeaway for us right now is this.
You, the human, are the dynamic, steady state navigating all this incoming information.
You are the active enclosure that gives the research its actual, tangible value.
Which brings us to the real danger of AI.
It's not that a machine types for us.
It's what Hubbard calls "prior capture."
This is when an existing, highly probable frame silently replaces a novel human idea before it can even be represented.
AI is an incredibly powerful "continuator" of known patterns.
So if you give it a wildly original thought,
its statistical nature might just sand it down into a beautiful, but completely average, familiar idea.
When we prioritize dodging AI detectors, we're actually training writers to sound like the corpus.
We're selecting for disguise rather than actual contribution.
Part four, novelty over similarity, finding the signal.
So how on earth do we fix this?
Well, we have to fundamentally shift our entire evaluation metric from similarity over to novelty.
We need to stop obsessing over the noise and start finding the signal.
I absolutely love this quote from the text.
Someone should be told that Shakespeare already did that one.
They should not tell Shakespeare he did not write it.
Just let that sink in for a moment.
If a student uses AI to generate an essay, the failure isn't that the prose mimics a machine.
The failure is lacking a brand new, reasoned idea.
The signal we should be looking for isn't the absence of AI.
It's the presence of real novelty.
Now, context is huge here, and we have to draw a hard boundary.
Fiction and research are evaluated entirely differently.
Fiction requires expressive human authorship.
As a reader, you want that unique human touch, that one-of-a-kind creation.
But research is judged by novelty, reasoned ideation, and reproducible evidence.
In research, an AI can absolutely help articulate a complex idea,
but the human is still the one held 100% responsible
for proving that the idea is true, new, and valid.
Finally, section five, the signal protocol, where we get into the actionable steps.
Okay, so we know the detectors are broken.
We know novelty is the ultimate goal,
but how do institutions actually do this in the real world?
Hubbard offers a seven-step solution called the signal protocol.
Let's walk through the first three steps.
Step one, declare the evaluation purpose.
Are we judging the process, the validity, or the novelty?
Don't let a single AI similarity score answer all those vastly different questions.
Step two, prefer recorded provenance to inferred similarity.
Trust time stamps, source deposits, and version histories
weigh over statistical guess.
And step three, bound detector output.
If you absolutely must use a detector,
you have to recognize it only measures statistical compatibility, not true authorship or value.
And this is where it really gets good.
Step four, construct a novelty ledger.
Document exactly what new relation or proposition is being introduced compared to prior work.
Step five, conduct a reasoned ideation audit.
Ensure the author can actually orally defend the reasoning and source material.
Step six, preserve multi-dimensional judgment.
You can never let one single automated coordinate silently substitute for human judgment.
And step seven, of course, is permitting correction and appeal.
So if you take absolutely nothing else away from Hubbard's volume 12,
let it be this core point right here.
The signal is not that the page resembles no machine.
The signal is that the page contains something worth finding.
When perfectly formatted prose becomes cheap and universally abundant,
human reasoning and original relationships become infinitely more valuable.
We really shouldn't be terrified of machines expanding what we can build with language.
It's a lot like how computerated drafting totally expanded what architects could design.
The only question that truly remains for us is this.
As AI expands what we can build with language,
what novel signal will you take responsibility for introducing to the world?
Thank you so much for exploring this fascinating topic with me.
Keep learning, keep asking the hard questions, and always focus on the signal.