In a separate article, I discussed on this topic:
Is AI theft?
Introduction
- The first industrial revolution – with the steam engine and the rise of factory production – took about a century to run its course.
- The second – driven by widespread electrification and industrialisation – took around 50 years.
- The third industrial revolution – with microchips and early automation – took roughly 25.
What we’re witnessing now will bring more drastic changes than all of those combined: in years, not decades!
Regardless of whether you are watching, reading, listening to this, or using a tool for the sensory impaired, these words (that carry some thoughts and ideas) are reaching your brain – that’s where it all begins and the magic happens. But… have you ever wondered if what you are experiencing is in fact an AI? A robot, not a human?
1. AI text and video production
These very words – were they written by an LLM (also known as “AI”)? Either directly, or by my hiring a “ghost writer” – a poor, underpaid Serbian guy who used ChatGPT to make all this text, because he has no clue about the topic I hired him to write about? What do you think – and are you sure? 🙂
Even on YouTube, when you are watching a person talk to the camera, is the person just repeating such a ChatGPT text? Did they at least bother to memorise it, or are they reading from a teleprompter placed right beside the camera?
Today, you can cheaply and easily use software that will re-position the eye pupils to appear as if the “actor” is looking straight at the camera, even if they are looking just beside it (at the teleprompter).
Also, when watching videos, are you 100% sure you are watching real people? On 21 May 2025, Google published videos of their AI-generated actors making dialogues – it looks scaringly realistic!
I’ve added the link to it below.
2. The volume of AI production
As you can see, realistic AI actors are not a matter of “if,” but a matter of “when?” AI-written articles and video scripts, on the other hand, are already here – and they have been for over two years now. Ahrefs SEO company published a study of 900,000 pages, using their detector for AI-generated-text (codenamed bot_or_not), and concluded that well over 70% of the newly published web pages were AI-written – see the pie chart for details:
Source: ahrefs.com
These results may (or may not) seem shocking to you, but as far as I can tell, they are most probably correct, if not even too optimistic!
It’s fair to note that Ahrefs “Free AI Content Detector” tool I tested is crap – it said this very script is “85% AI written” – LOL 🙂 But it is also fair to note that most articles and writers are using AI nowadays – don’t take my word for it, find one who doesn’t – if you can!
As The Prodigy said:
“…Now the writing’s on the wall
It won’t go away
It’s an omen
You just run on automation…”
I’ve been experimenting and examining the limits of AI since 2023, and can confirm first hand how powerful it is, and how “human-like” its writing can be.
See my AI experiment for more details.
As an avid reader and a good writer, I am often able to tell if an article/script was AI-written, especially if it’s about something from my area of expertise. As far as I can tell, AI-created stuff has become the norm, not an exception in 2025.
3. Can we really tell which is which?
It is fair to admit that AI-written stuff can look very much like human written stuff, and even I often have problems telling which is which. Can an average working mom/dad tell the difference – if it’s about something they are just learning about? I would not bet on it.
Can other AI/LLM systems tell? I also doubt it. First generation of AI scraped and copied human-written websites like BikeGremlin. Now, the AI is reading stuff written by other AIs. And it is producing aritcles and videos that humans are learning from. It is a closed loop, that may not end very well.
A Zurich university conducted an experiment on Reddit (from November 2024 to March 2025), quote:
Over the past few months, we used multiple accounts to posts published on CMV. Our experiment assessed LLM’s persuasiveness in an ethical scenario, where people ask for arguments against views they hold. In commenting, we did not disclose that an AI was used to write comments, as this would have rendered the study unfeasible. While we did not write any comments ourselves, we manually reviewed each comment posted to ensure they were not harmful. We recognize that our experiment broke the community rules against AI-generated comments and apologize. We believe, however, that given the high societal importance of this topic, it was crucial to conduct a study of this kind, even if it meant disobeying the rules.
Make your own mind about the ethics and justifications of this, but I think that a study like this was an important one to be made.
The published results of this experiment are not available to me at this time, but redditors were pretty much fooled they are discussing with humans. What is available (haven’t confirmed if it’s correct) is a study by two dudes from the University of California, San Diego:
https://arxiv.org/pdf/2503.23674
Abstract
…The results constitute the first empirical evidence that any artificial system passes a standard three-party Turing test. The results have implications for debates about what kind of intelligence is exhibited by Large Language Models (LLMs), and the social and economic impacts these systems are likely to have.
The basic point of the experiment results is that people could not correctly guess which of the two entities they are chattig with is a human, and which is an AI.
This kind of confusion can breed paranoia – so a well-written and structured article can be “accused” of being AI-written. Talk about irony.
For experiment and for fun, I asked “ChatGPT 4o” to write an article about this topic. Then, I asked it to write an article but with my closer guidance and some minor editing on my part. I will link both articles below so you can see and compare for yourself.
To me, the AI-written and AI-assisted articles look to be pretty good, pretty convincing. The version where AI, like a very good editor, polished my writing is especially, scaringly good (I know I’m repeating myself about this, but… damn!). Which one looks better to you? Should I hang up my writing boots? 🙂
Conclusion
This is not fiction, nor science fiction. By the end of this year, we may hardly find any text published by human authors.
Update 2026: I wrote this article in 2025, and now it seems like I was right. Now, I wrote a newer article discussing a similar aspect of human-like AI work for marketing emails:
AI spam 3.0 – can we tell?
Of course, AI can be used to do the boring, tedious stuff, and in that case it is not that much different (basically) than a battery powered screwdriver. The problem is when AI replaces human thinking, emotions and creativity.
For extra irony, I ran this script and the two AI-written scipts through some AI-detection tools, and some mixed up which is which (robots vs humans).
References
- OpenAI (ChatGPT) makes a deal to train AI on Reddit data:
https://techcrunch.com/2024/05/16/openai-inks-deal-to-train-ai-on-reddit-data/ - Runway ML eye-refocus software (and video editing in general):
https://app.runwayml.com/ - Nvidia Broadcast video editing and altering software:
https://www.nvidia.com/en-eu/geforce/broadcasting/broadcast-app/ - CapCut AI video generator:
https://www.capcut.com/tools/ai-video-generator - Google’s AI video model Veo 3:
https://blog.google/technology/ai/google-flow-veo-ai-filmmaking-tool/ - Demo of Google’s AI actors making a dialogue:
https://www.youtube.com/watch?v=QYnJ3qJ5qJQ - AI Mode is obviously the future of Google Search (a news article):
https://www.theverge.com/google-io/670439/google-ai-mode-search-io-2025 - Ahrefs study about the percentage of AI-written web pages:
https://ahrefs.com/blog/what-percentage-of-new-content-is-ai-generated/
Lily Ray’s article rewritten by Google’s AI:
Should we let AI do the “heavy lifting”, “but guide it”?
VIDEO version of this article
Last updated:
Originally published:
