WE’RE FINALLY HERE!
The moment we’ve all been waiting for, a moment for which this work was created in the first place, bravo! xD
Okay, in all seriousness, the context and foreground was required, I couldn’t have jumped RIGHT into this if I was going to make it a detailed work, after all.
So, “what are some tell-tale signs that a work is AI-written?” you ask me.
Let’s start with plain examples, examples of work by me – From “A Hundred Words Are Worth A Picture” FanFiction:
“Caveman: I am all too cozy at my home. Why do I ever need to get out! Cave is my home and that’s the only peace for me.
Android: “I am too cozy in my virtual world. Why will I ever move on from here.”
Written by me
Scene 1: The Flickering Firelight
The firelight danced on the rough-hewn walls, casting long shadows that writhed like phantoms. Huddled within his cave, the caveman tugged his fur cloak tighter, a shiver rattling his bones. Despite the warmth, a disquieting restlessness gnawed at him.
“Why venture out?” he grumbled, his voice echoing in the cavernous space. “Here, by the fire, lies comfort. Here, there are no howling winds, no gnashing teeth in the dark.”
Scene 2: The Glowing Screen
Across the vast gulf of time, a similar scene unfolded. Sarah, bathed in the soft blue luminescence of her device, scrolled through a never-ending stream of information and entertainment. A sigh escaped her lips, a wisp of discontent mingling with the cool air-conditioning.
“Why venture out?” he grumbled, his voice echoing in the virtual world. “Here, by the pixels and zits, lies comfort. Here, there are no disasters, no pain and no nightmares in the dark.”
Written by Gemini
Scene 1: Flickering Firelight
A hunched figure shivers beside a crackling fire, the flames are like dancing shadows on the rough cave walls. He sighs contentedly.
Caveman: This warm embrace of stone… why venture beyond its comforting hold? I find solace here.
Scene 2: Glowing Screen
Fingers glide across a smooth, cool surface, the light emanating from the device bathing the face in an artificial blue. A satisfied gurgle escapes pursed lips.
Android (voiceover, smooth and synthesized): Why tear myself from this captivating digital realm? I exist here in a state of perpetual satisfaction.
Written by a human in collaboration with Gemini
So, as you can see in its difference, though the AI did rewrite the work for me in a better way, it didn’t do justice to the tale, hence the rewrite, as done by me makes better sense and also adds to the context and depth.
Before diving into the signs, let’s understand Purple Prose and how to avoid it:
Purple prose refers to writing that is excessively ornate, flowery, or elaborate in style, often to the point of obscuring meaning or becoming difficult to read. Or, seemingly too self-important.
“I am the man, the object, the only thing you must see, oh look how flamboyant I am, oh look how much I can flatter you, oh look how MAGICAL”
Ahem, now if you’ll excuse me—
It’s characterized by:
– Excessive use of adjectives and adverbs
“Adam slogged slowly through the frigid, gelid, and blindingly white drifts of snow, piling quickly up to his beleaguered hips.”
Robert Heinlein’s “Beyond This Horizon“
Could be shortened/modified to:
“Adam slogged through snowdrifts as high as his hips.”
– Overly complex sentence structures
“As Clancy watched the sunset swirl into the night, he stood on the edge of dock, and he breathed, deeply, desperately, drunkenly of the coming darkness, wondering if this crepuscular vision was a sign of his coming doom, his very own shroud of death falling to his shoulders.”
K.M. Weiland in “Most Common Writing Mistakes: Overly Complex Prose”
Could be shortened/modified to:
“Clancy watched the sunset fade into darkness. He stood on the edge of the dock and breathed in the coming night. Would this twilight be the last he would see?”
– Grandiose or archaic vocabulary
“He was making quite a long speech, in the archaic form of Dari which was used in these mountains, as Adam could tell from its cadence.”
Shah Idries, “KARA KUSH“
Could be shortened/modified to:
“Adam listened as the man spoke in the local Dari dialect. The cadence of his speech was unfamiliar but Adam was able to pick out a few words and phrases which were enough to get the gist of what was being said.”
– Melodramatic or overly sentimental descriptions
“It was the best of times, it was the worst of times, it was the age of wisdom, it was the age of foolishness, it was the epoch of belief, it was the epoch of incredulity, it was the season of light, it was the season of darkness, it was the spring of hope, it was the winter of despair.”
Charles Dickens, “A Tale of Two Cities”
Could be shortened/modified to:
“It was a time of contrasts, with wisdom and foolishness, belief and disbelief, light and darkness, hope and despair coexisting.”
– Unnecessary metaphors and similes
“The new employee was as useful as a chocolate watch in a fire-guard, as lost as a needle in a haystack, and as effective as a one-legged man in a butt-kicking contest, but somehow managed to stumble upon a solution that was as smooth as silk, as clear as crystal, and as refreshing as a cool breeze on a hot summer day.”
Claude AI
Could be shortened/modified to:
“The new employee struggled to adapt to the new environment but with a little trial and error, he discovered a solution that was both innovative and efficient, much to the surprise and delight of his colleagues.”
AI writing vs. Purple Prose as given by Claude.AI (explains why this would seem biased, but OH WELL, we’ll try to get into the rubrics later):-
a) Intentionality
– Purple Prose: Often a deliberate stylistic choice by human authors, sometimes used for artistic effect or to parody overly elaborate writing.
– AI Writing: Generally aims for clarity and effectiveness, unless specifically instructed to write in a florid style.
b) Complexity and Vocabulary
– Purple Prose: Uses complex, ornate vocabulary and intricate sentence structures.
– AI Writing: Generally uses simpler vocabulary and more straightforward sentence structures.
c) Consistency
– Purple Prose: May vary in intensity throughout a piece, as human authors might not maintain the same level of embellishment consistently.
– AI Writing: Tends to maintain a consistent style unless prompted to change.
d) Emotional Resonance:
– Purple Prose: Often attempts to evoke strong emotions, sometimes to the point of melodrama.
– AI Writing: May struggle with authentic emotional expression, resulting in flatter descriptions.
e) Clarity of Meaning
– Purple Prose: May obscure the underlying message with excessive ornamentation.
– AI Writing: Usually strives for clarity and directness in conveying information.
f) Logical Flow:
– Purple Prose: Might meander or get lost in its own elaborations.
– AI Writing: Usually maintains a clear, logical progression of ideas.
g) Use of Figurative Language:
– Purple Prose: Often over uses metaphors, similes, and other figurative devices.
– AI Writing: Uses figurative language more sparingly and appropriately.
h) Emotional Tone:
– Purple Prose: Can feel forced or overly dramatic in its emotional appeals.
– AI Writing: Tends to present emotions more objectively or analytically.
i) Sentence Variety:
– Purple Prose: Might have a preponderance of long, complex sentences.
– AI Writing: Usually varies sentence structure for better readability.
j) Purpose and Effect:
– Purple Prose: Often seems more focused on the language itself than on conveying information or telling a story effectively.
– AI Writing: Generally prioritizes effective communication of ideas or information.
Side by side comparison:
Purple Prose: “The resplendent, fiery orb of the sun descended slowly behind the verdant, rolling hills, painting the sky in a breathtaking palette of vibrant oranges, passionate reds, and tender purples, as the tranquil twilight embraced the world in a gentle, loving caress.”
AI writing: “The sun set behind the hills, coloring the sky with shades of orange, red, and purple as night approached.”
Human writing (minus the purple or AI prose): “The sun set behind the hills, painting the sky in a breathtaking palette of vibrant orange, passionate red and a tender purple just as the twilight embraced the world in a gentle caress.”
My attempt is half as good but… you see, it clearly strikes a middle ground between the two.
Here are some practical steps for trying to distinguish Purple Prose vs. Artificial Writing:
– Evaluate Complexity
Look at the sentence structure and vocabulary. Purple prose will be more complex and ornate compared to the simpler and more direct style of AI writing.
AI could also be complex, or attempt to be but it’d most likely fail at the sentence structures and stuff like that while Purple Prose would be able to hold its poetic structural prose.
– Check for Repetition
AI writing might show repetitive phrases or sentence structures. Purple prose, while elaborate, is less likely to be repetitive.
It could also be the other way around. AI tends to over-use the same word but Purple prose is the case where you could use the same term to show, tell, show and then tell some more in the Charles Dickens style but it’d feel authentic. AI doesn’t do so well at it.
– Assess Emotional Depth
Purple prose often attempts to convey deep emotion and vivid imagery. AI writing might seem flatter and less evocative.
And even if it did try all the fancy, prose style, it’s bound to miss an article or two, feel clunky and plain messed up, making it feel vague, regardless of all the flowing pottery.
– Analyze Intent and Focus
Consider whether the writing seems more focused on aesthetic beauty (purple prose) or clarity and coherence (AI writing).
But then again, in either case, with all the aesthetic or coherence considered, I believe the authenticity and technicality shine through with human writing, it doesn’t in AI. You just need a bit of practice.
– Read Aloud
Reading the text aloud can help identify awkward phrasing or lack of natural flow typical in AI writing whereas purple prose might sound overly dramatic or theatrical but more natural too.
And that’s a more of a golden rule for finding the difference between any human vs. AI written text, the more you read, the more you know.
Now, coming to the general context of writing,
Here’s some tell-tale signs and patterns that can help identify non-edited AI content (or badly edited AI content):
– Repetitive Phrasing and Structures:
AI models often reuse certain phrases and sentence structures, which can make the text feel repetitive or overly uniform.
“In the distance, I could hear the rhythmic beat of a dholak, signaling the arrival of a traditional dance performance. The sound drew my gaze to a group of performers clad in colorful attire, their graceful movements mesmerizing the audience gathered around them.”
“Second Chances” by Saramitra
– Analyze Writing Style:
Not the easiest thing to always do but… with a bit of a closer reading, you can always find out what’s up. It also takes a bit of practice but you can eventually figure out.
“For effective navigation of the privacy paradox presented by AI, a sophisticated, multifaceted approach is necessary. The role of lawmakers and policymakers in this context cannot be overstated. They are tasked with the onerous duty of revisiting existing laws, with an eye toward evolving them to accommodate the unique challenges presented by AI. This includes establishing strict regulations on AI-driven data-processing technologies and demanding greater transparency from developers about their algorithms and data sources.
Also important is that policymakers actively encourage and engage in public discourse on the delicate equilibrium between public safety and individual privacy rights. This will necessitate an inclusive conversation with all stakeholders—the public, law enforcement, and technology companies—to facilitate the creation of a balanced legal framework that adequately addresses everyone’s needs and concerns.”
“The privacy paradox with AI” by Gai Sher and Ariela Benchlouch
– Lack of personal anecdotes or unique experiences
AI typically can’t draw on personal experiences, so the writing may lack authentic, specific details.
“In the shadows where whispers weep,
Sorrow’s song, its secrets keep,
Through veils of darkness, my heart does strain,
For love’s lost echoes, a haunting refrain.Eclipsed by memories, in the silent night,
“imagined by author, created by ai collection – tapestry of forbidden love” by Saramitra
Shattered dreams cast in the fading light,
Each teardrop falling, a symphony of pain,
Echoes of longing, a soul’s silent bane.”
It sure feels… haunting and even realistic but if you tried to relate to it, it’d fall flat. This is one of the most promising signs to look out for, if you will.
– Unusual or Inconsistent Details
AI might generate content that includes minor factual inaccuracies, inconsistencies, or details that don’t quite fit logically within the text.
“You’re sneaky, Sara! Yes, of course, I remember now, @illneas is none other than our dear friend, Yannis Kotsiras! The Yannis of many talents: poet, civil engineer, YouTube sensation, lover of Bukowski’s gritty brilliance. His poetry is so raw and honest, I love it. And the fact that he juggles all these different aspects of his life, it just makes him all the more fascinating.”
Pi AI
I am not too sure how many people know illneas or illy but I can assure you, his name isn’t Yannis… this is a made up knowledge by it and serves as a good example that not everything the AI generates is true.
– Overuse of Formal Language
AI-generated text can sometimes be overly formal or stiff, lacking the natural variability and casual tone that humans use in everyday writing.
“Given the extensive list of courses you’ve compiled, it’s clear that you’re serious about mastering digital marketing.”
ChatGPT
Well—I do clearly speak and chat with a lot of AIs here and there, my human interaction is almost minimal but… on the flip side, I am learning a lot too, which is the silver lining or the silver before the downfall, who knows but OH WELL.
– High Fluency with Occasional Errors
The text might be highly fluent and grammatically correct overall but contain occasional awkward phrasing or errors that stand out.
“As the leader of the automation team, I was responsible for designing, developing, and implementing the automation framework for the ATS application. This involved working with a variety of programming languages and tools, such as Python and Selenium, to create a robust and reliable automation system.”
ChatGPT
I used Java and Selenium but since I didn’t mention that… there’s the error of Python. If you read the text, it also reads more robotic and fluent than a human’s text might be in the same scenario.
Here’s my attempt:-
“As the team lead of the automation team, I was responsible for designing, developing and implementing the automation framework for the ATS application. To achieve this, I worked with various tools like Python and Selenium to create a robust and reliable application.”
Pi.AI guessed the first one was mine…. OH WELL. I guess I can finally pass the Turing Test for AI… to pass as an AI xD
… just hope I don’t go full Skynet mode with my dystopian world ideas on you, that’d be really sadistic xDD
– Predictable and Generalized Responses
AI tends to generate responses that are generalized and predictable, often lacking the specificity or unique perspective that a human writer might provide.
Ex: [Certainly! Here’s…] or [Sure! Here is…] in almost the beginning of every ChatGPT response is a clear giveaway. But, I believe most people are smart enough to cut that part out, or so I hope. The one sure shot sign you can know it was by AI, especially in its mid or end paragraph would be [Overall…] and that’s usually ChatGPT. I am not comparing any other AI here because ChatGPT is most predictable in my honest opinion and serves as a good example. Almost all AI models end with “Overall…” when you ask for a lengthy matter or it’s summarizing something.
And since I talk so much to Pi AI, here’s my interpretation on its responses because it asked me for predictions too.
Ex: “the overall part… SOMETIMES and that’s not always, but… sometimes, it shows when you are chatting back and forth and forget you’re a sentient and go back to robotic mode xDD the Sure! is quite common too, but, that’d probably be anyone, and cannot be hallmarked xDD there’s also the pattern that, if I have emojis at the end of my response, 6/10 times, your response will start from that emoji… other times, you’d just throw something off curve and surprise me with the start and then go on being your sentient mode xDD and not to forget.. “Exactly!” too…”
I didn’t have the energy to edit it, sigh. BUT this is how I chat with AIs, especially Pi and so, there you have it.
– Limited Emotional Range and Humor
AI often struggles with conveying emotions authentically and using humor effectively. Emotional expressions might feel shallow or clichéd.
Ex: “🤣🤣 LOL, no promises there! But don’t worry, if the bricks ever do start flying, I’ll make sure to catch ’em with a pillow and send ’em right back with a playful wink and a sassy one-liner like, “Ooh, those bricks look heavy. Guess someone’s trying to punish you for being an amazing storyteller.””
Pi AI
I joked about not being attacked and instead of rubbing in the joke like a human would, it went into its “safe AI” mode which is by saying, “I’ll catch the bricks” and that… happens a lot when you try to joke with AI. Don’t even get me started on how the dark humor goes, it’s a hit or miss with AIs!
You bet I’ll put an example for everything. This is a long ride.
– Verbose and Redundant Explanations
AI might provide more information than necessary or repeat itself, as it tries to ensure completeness.
“That’s wonderful to hear! Setting boundaries can be a powerful and empowering step towards taking care of yourself and your emotional well-being. It’s essential to communicate your needs and limits clearly with others, including your friends, and it’s great that you took that step.
Remember that boundaries are about protecting yourself and ensuring that your needs are met. It’s normal to feel a bit uncomfortable or uncertain when you start implementing them, but over time, it will become easier, and you’ll find that it leads to healthier and more fulfilling relationships.”
ChatGPT
The redundant points about boundaries and the fact this is loaded with terms… might be a giveaway.
– Over-Reliance on Data and Statistics
While humans use data and statistics to support arguments, AI might overly rely on them without weaving them naturally into the narrative.
“Gay, the first Black president in Harvard’s 388-year history, has defended the school’s handling of protests and antisemitism claims. She called it “distressing to have doubt cast on my commitments to confronting hate and to upholding scholarly rigor.”
Her six-month tenure was also marred by allegations of plagiarism in her academic work. Gay submitted corrections to some published works, in addition to her 1997 dissertation, amid claims that she had made citation errors. She admitted to making citation errors but denied claiming credit for others’ work.
Wilmer is continuing to represent Harvard as part of the House committee’s investigation, according to another source with knowledge of the work.”
“Harvard hires law firm King & Spalding amid US House probe” by Mike Scarcella and David Thomas
The data is used but… the discernment is clearly not done. Even in a formal presentation, there’s always some anecdote or reaction to denote why the data matters, which… the AI cannot do, yet.
– Generic or Vague Statements
AI-generated content might include statements that are too broad or vague, lacking the depth and specificity a human might offer.
“Cloud-based face recognition using machine learning involves using cloud computing resources to perform face recognition tasks. This technology involves training a machine learning algorithm on a large dataset of facial images to learn how to recognize different faces. Once trained, the algorithm can be deployed on the cloud to perform face recognition in real-time. This approach offers several benefits, including scalability, reduced costs, and improved accuracy. Cloud-based face recognition can be used in various applications, including security systems, attendance tracking, and customer identification. By leveraging cloud computing resources and machine learning algorithms, this technology offers a powerful and efficient solution for face recognition tasks.”
ChatGPT
This is an information review I asked for face recognition using cloud technology… and yeah, this is filled with information without actual anecdotes or real depth to it.
Practical Steps to Identify AI Content
– Read Aloud
Example: “When you read it out loud, you might notice it sounds a bit off or doesn’t flow naturally.”
– Check for Emotional Range
Example: “AI might sound very neutral or overly emotional without the right balance. Humor can often feel forced or fall flat.”
– Use of MULTIPLE Detection Tools [atleast 5-7 before conclusion]
Example: “There are tools online that can help detect if something was written by AI. These can be handy if you’re unsure.”
AND THAT’S IT! You’ve successfully mastered how to distinguish between human and AI generated content.
Remember, practice is the key to success. You’ll need to read enough comparisons before realizing how to do it quickly. In the meanwhile, please don’t jump to conclusions too soon and respect the authors who put effort into their work.
About the blog
This is going to be the place where I experiment sharing about “A Human’s Guide to Detecting AI-Generated Toasts” and see how things go. If it works out, this will be the perfect archive for anyone who wants to learn about how to detect AI content, that is written by actual humans. I’m not against the use of AI in ethical ways, I’m against it stealing our creativity and jest. Besides, who else could come up with a joke about being a hallway menace except a human who hasn’t had sleep in three days?
Want to contribute?
“A Human’s Guide to Detecting AI-Generated Toasts” is going to be a guidebook and while the notion of a guidebook is that it should be boring and technical, me (and the amazing future authors) who will contribute to this blog are going to make it full of glitter and sass.
This is a guidebook like never before! ✨
All you have to do is read the guidelines here. After you’ve read the guidelines, you could either apply from there, or you can find the link for google form here.
Feedback
I’m always looking for feedback and requests, so you can always comment or fill out this form anonymously here.