Blessed are the AI Detectors
Large Language Models Come to Jesus
For millennia, philosophers and theologians have questioned where the divine inspiration ends and the human begins.
Today, there’s a new player on this existential chessboard: AI.
Suddenly, it’s not just about divining the divine; it’s about untangling the strands of human creativity from the probabilities of algorithms.
Assuming they’re not the same thing.
So for today, an experiment: an ancient text, and its AI-generated counterpart.
Can the AI detector tell the two apart?
A canonical text
The Beatitudes undoubtedly stand as a cornerstone in the Western Canon.
Originating from the first century CE, they documented some of Jesus of Nazareth’s most popular oral teachings.
Although two millennia have passed, their resonance hasn't dimmed. Even those distant from religious traditions might chuckle at Monty Python's irreverent "blessed are the cheesemakers."
For clarity, here's a snippet from Matthew's gospel, Chapter 5, as presented in the 1611 King James Version (KJV):
Blessed are the poor in spirit: for theirs is the kingdom of heaven.
Blessed are they that mourn: for they shall be comforted.
Blessed are the meek: for they shall inherit the earth.
Blessed are they which do hunger and thirst after righteousness: for they shall be filled.
Blessed are the merciful: for they shall obtain mercy.
Blessed are the pure in heart: for they shall see God.
Blessed are the peacemakers: for they shall be called the children of God.
Blessed are they which are persecuted for righteousness' sake: for theirs is the kingdom of heaven.
A Modern AI’s Beatitudinal Interpretation
Now, place that beside a 21st-century rendition of the Beatitudes, as cranked out by GPT in 2023:
Blessed are they who unplug, for they shall find peace in the present.
Blessed are the environmentally conscious, for they shall inherit a preserved Earth.
Blessed are those who stand against prejudice, for they will foster unity.
Blessed are the fact-checkers, for they shall find truth amidst misinformation.
Blessed are they who uplift the downtrodden, for their kindness will be the light in darkness.
Blessed are the digital detoxers, for they will experience the world without filters.
Blessed are those who promote mental health, for they will heal the silent sufferings.
Blessed are the peacemakers in online forums, for they shall be called moderators of harmony.
Blessed are those who teach without bias, for they will nurture the open-minded leaders of tomorrow.
Blessed are they who log out from hate and negativity, for their feeds shall be filled with joy and positivity.
Deus ex machina?
Now, a challenge for the discerning reader:
I plugged both sets into one of the top-ranked AI detectors from a quick Google search (let's keep its identity a mystery).
Guess which one the detector confidently labeled as 100% AI-generated?
You saw it coming… the AI earmarked the timeless KJV 1611 as 100% machine-conjured, while the GPT 2023 version seemed unmistakably human.
Though the 2023 version radiates a very human essence, layered with today’s zeitgeist, its categorization was undeniably a curveball.
What’s Going On?
Large Language Models (LLMs)
At their core, Large Language Models like GPT-3 or GPT-4 are massive neural networks trained on vast swaths of the internet’s text. As well, of course, the King James Version of the Bible. Not to mention every other scripture that’s been digitised.
Their primary function? To predict the next word in a sentence.
Blessed are the…
Feed them a prompt, and they’ll generate a coherent continuation, often with uncanny eloquence.
… adaptable, for they shall thrive in changing times.
They’re not just regurgitating memorized text; they’re combining patterns they’ve learned to produce novel responses.
Not exactly the most profound, but it’s certainly grammatical and at least from the perspective of Google, completely original:
No results found for "Blessed are the adaptable, for they shall thrive in changing times."
Let future searches for that phrase point to this Medium article!
AI Detectors
AI detectors try to be modern manifestations of the Turing Test, which gauges a machine’s ability to mimic human-like intelligence.
If a machine passes the Turing Test, it’s demonstrated that its behavior is indistinguishable from that of a human.
This brings forth a thought-provoking question: If it’s indistinguishable, isn’t it, for all practical purposes, human?
These detectors scrutinize text, searching for patterns or inconsistencies that might betray an artificial origin.
Curiously for me, words like "palpable" seem to surface more frequently in outputs from ChatGPT—perhaps a hint of machine’s touch? Always a little bit too much cliché?
However, as LLMs feed on vast amounts of human text, drawing a clear boundary between man and machine becomes a challenging endeavor.
Remember, human writers, even those as unique as Monty Python, have drawn from the same well of human-generated content.
And probably much less of it than an AI, at least when it comes to text.
The Fundamental Challenge
Here’s the crux of the matter: it’s impossible to definitively conclude that something hasn’t been created by a human.
In the same way it’s impossible to categorically reject that something has been divinely inspired.
Your priors matter. What you already believe often turns out to be conveniently confirmed by whatever comes next!
And with the rise of sophisticated AI, the line between human and machine-generated content is becoming increasingly blurred.
It’s not just a technological challenge but an existential one.
As AI continues to advance, our understanding of creativity, originality, and even humanity is continually reshaped.
Beware of casting the first stone
For now both AI generators and detectors need the guiding touch of human judgment. Healthy skepticism is a must when navigating their conclusions.
Above all, when consulting a random AI detector, take its results with a grain of salt.
Should you ever suspect a text is AI-generated, think twice before jumping to conclusions.
First, ask yourself does it even matter how the sausage was made?
And secondly, ask whether the author is in a position to take responsibility for the text and its consequences?
After all — to err is human, but to really foul things up requires a computer.
Support the author by visiting…