The role of the conscious observer has posed a stubborn problem for quantum measurement. Phenomenology offers a solution

- by Steven French

Read on Aeon

This article rotted in my drafts for three years before initial publication and at no point before or since has there been a way to detect if media is generated by AI with 100% certainty, nor will there ever be in the future. The ‘vibe’ or stylistic signature of content is not a way to identify if it is AI made; especially as it becomes more common, that ‘vibe’ is driven out, and people learn to replicate it.

Generative AI is trained on the entire accessible and applicable corpus of human output. It is also being created and refined by humans trying to minimise any difference from what is human-creatable with almost limitless budget and resources. AI models are also actively trained to minimise distinguishable patterns, and any detectable patterns that are distinguishable become targets to eliminate.

We are beyond the point of being able to identify AI-generated content, and there is no way to reliably mark content as being AI-created in such a way that it cannot be circumvented.

Detecting Text and Speech

In response to the rise of large language models, a lot of ‘detectors’ launched – especially ones targeting the education sphere.

GPTZero is one of the more well-known and scrupulous detectors, though a single glance at their page explaining how their system works shows technobabble and vague diagrams. Throughout their marketing material they make claims of being the leading offering and the most accurate. In their FAQ, under the section ‘What are the limitations of AI Detectors?’, they write:

No AI detector is 100% accurate, and AI itself is changing constantly. Results should not be used to punish or as the final verdict. Accuracy improves with longer inputs (document-level results are stronger than paragraph or sentence-level) and is strongest on English prose.

This raises a fundamental question: what is the purpose of a ‘detector’ so unreliable that even the creators advise not trusting it for important use cases? What is the purpose of a tool like this at all if one mustn’t trust it and instead vet the material themselves?

The truth is, no matter what, students will cheat. Before AI, the equivalent was paying an essay writer or using plagiarism converters. For educators to tackle students using AI, they need to take another approach. Trying to offload the effort by using so-called AI detectors, which we have already established are flawed, isn’t the way to go about it.

Plagiarism is a gross case of academic misconduct, so levying accusations without just cause can be extremely damaging and isn’t something to be taken lightly. Automated AI detectors should not be considered admissible evidence, and many shouldn’t have their evaluations considered at all.

While it is true that people will mainly use the popular offerings, there are so many models out there – each with their own distinctions – such that it is impossible to identify patterns for each of them. Even if you could, it would be an endless task as new models are releasing continuously, and one could feasibly train their own or gaolbreak or fine-tune an existing model to evade detection.

People often point to writing features such as the em dash (—), tricolon, or emoji list as proof of text’s AI origin, forgetting that AI models are trained on human output and that all of the aforementioned have long been features of content in AI’s training corpora. Some people take the presence of these writing features to the extreme that they vocally and entirely unnecessarily attack writers despite them not using AI.

When I first tried GPT-3 in 2022 via the OpenAI Playground, I found that it wrote in a style almost identical to my own. My writing at the time had a certain quality of rigid verbosity, and GPT-3 mirrored that. It was bad enough that detectors, especially in their extremely rudimentary forms, often flagged it. As a student at the time, I was worried enough that I took conscious effort to change my writing style away from that previous form.

Beyond people who already have a style similar to popular AI offerings, there is also the case of the proliferation of AI content everywhere influencing how people are writing and speaking. The paper Empirical evidence of Large Language Model’s influence on human spoken communication has looked into this. To quote it:

This research explores whether the loop of human-machine cultural transmission is indeed closing, with humans adopting the language patterns generated by AI and vice versa.

To address this, we analyzed approximately 280,000 video transcriptions from over 20,000 academic YouTube channels, focusing on (i) changing trends in word frequency following the introduction of ChatGPT, and (ii) the correlation between the trend changes and word usage by ChatGPT. We show that, following the release of ChatGPT, word frequencies in human spoken communication began to shift, exemplifying the transformative impact of AI systems on human culture.

Even if not directly interacting with AI models, their output is so prevalent that it’ll be picked up second-hand. AI models are fundamentally warping language around them, and this will continue for as long as they exist.

Good AI voices can be a little uncanny if you actively listen for it but are largely beyond the hump now and indistinguishable at the cutting edge.

Detecting Imagery

Images from the most capable models have gotten really good. Good enough that I cannot distinguish them from real in passing, ¹ and even when pixel peeping find it difficult or impossible. The earliest models were blatantly AI, with nothing seeming cohesive and an ethereal property. As time progressed though, they’ve become more and more coherent.

Certain tells remained, such as an unexpected number of fingers or things interacting in odd ways. Certain details that weren’t the focus of the shot were deformed upon inspection, and details such as textures, clothing, hair, etc, would subtly fade and disappear in a non-Euclidean way.

Still, certain models have specific tells. For example, OpenAI’s 4o Image Generation has a tendency to apply a warm, sepia-tone filter of sorts over everything it generates, and this becomes more exaggerated as you request further refinements and it further yellows images.

Conversely, many models don’t have specific tells at all. There is no specific ‘look’ attributable to imagery generated by them, and some advanced models such as Google’s Nano Banana maintain consistency across revisions in such a way that they can output AI-altered versions of genuine photography (though with some very minor changes), further muddying the waters.

Video

Video is a collection of images, so much of the same is applicable. However, the introduction of motion comes with further tells. In early AI videos, continuity between frames was atrocious due to each image being generated as more-or-less individuals.

As the technology has progressed however, continuity has improved hugely, though physics and interactions remain objectionable. As of writing in September 2025, video can be generated with matching audio at the quality of individual images with reasonable object permanence. Still identifiable in most cases to the discerning viewer, but quickly joining the ranks of still image generation.

Possible Solutions

The obvious way to have things identified as AI-generated or altered is with a watermark. This is also completely futile.

How do you implement this watermark? As a visual indicator? What stops someone from just cropping it or erasing it? As metadata? What stops someone from just scrubbing it out?

One might argue that the presence of metadata could be evidence of something being AI, and its absence means it is either AI or not, but that falls apart if someone decides to apply AI identification metadata to non-AI content.

The Coalition for Content Provenance and Authenticity (C2PA) does exist and has members including OpenAI, Meta, Google, Microsoft, and plenty of other AI companies but is more or less useless for acting as a definitive source for if something is AI generated.

OpenAI note in their policy that:

Metadata like C2PA is not a silver bullet to address issues of provenance. It can easily be removed either accidentally or intentionally. For example, most social media platforms today remove metadata from uploaded images, and actions like taking a screenshot can also remove it. Therefore, an image lacking this metadata may or may not have been generated with ChatGPT or our API.

It is worth noting that most existing image formats lack any secure way to store metadata without the ability to tamper with it. Even if you somehow could make the metadata itself untamperable, one could always just screenshot or take a picture of it, creating a new image devoid of that metadata. Further, if at any point that format of storing metadata was broken or cracked, the validity of all previous images would be immediately thrown into question.

There is also the case that all existing generative AI content before the introduction of the metadata lacks any distinction, which allows it to slip through, and that anyone could train their own model that sidesteps these precautions entirely.

OpenAI have also dabbled with visual watermarks, with small coloured blocks in the bottom corner of their DALL-E output, but they were easily removed or avoided. OpenAI didn’t bring this visual watermark feature back for future image models either. Images generated with Gemini have a small watermark in the bottom left, but that is also easily removed.

A better approach is Google’s SynthID, which uses stylometry, but that is not a perfect solution and can be worked around by editing. One can also just use AI tooling without such watermarking built in.

Text can be watermarked by altering the probabilities of certain words, but this isn’t sure-fire and can be circumvented by back-translating the text, paraphrasing it, or employing one of a number of other transformations. It also won’t work forever, given language’s fluidity in adopting AI styling as referenced earlier. It is possible for the file containing the text to have its own metadata, but it is all too easy to just copy or transcribe text out of one file to move it into another without the metadata ².

Incentives Against Detection

There is also the case that there are parties who do not wish for AI-generated content to be identifiable. Detection companies have financial incentives to claim effectiveness even when limited, and AI companies have incentives to make their output indistinguishable.

People using AI output in many cases do not wish to have it identifiable as AI output – especially given the negative reputation AI output has for being slop and the potential backlash that brings with it. There are many con artists making substantial financial gains peddling AI output who I’d suggest would be willing to go to lengths to avoid people taking note ³.

Companies building generative AI offerings may not wish to have their output identifiable as AI, especially if their competitors don’t identify their output, which could prove a selling point in some cases.

I’ve also seen the argument that content that is modified by image editing software like Adobe Photoshop is not marked in any meaningful way, so AI output shouldn’t need to be either.

There is no reasonable way to identify AI-generated content with 100% certainty, and as time goes on, people who say, ‘I can always tell,’ are only sounding more and more disconnected from reality.

I wish to make clear that we have absolutely no reliable means to identify AI-generated content and that there is no reasonable way to implement an identification system in the future. Indistinguishable AI-generated content is here and upon us.

If you don’t think you’re exposed to AI-generated content, perhaps question if that is because you aren’t exposed to any or because it has reached a point that you don’t notice any.

Hanging on the wall in my parents’ house is a picture of my older brother and me moving into a new home. I’m holding a teddy bear, he has a box with a few toys poking out the top. I am guessing I’m around 5 years old and he was 8. The funny thing is I never lived in that house and the man and woman in the photo aren’t even my parents, yet my mother insists on proudly displaying the framed image.

It is important to know that as children my brother and I did a bit of acting and modeling. In addition to the image above (which I think was for a home insurance company), I was also in a commercial for regional phone company, a national cereal brand, and one cold February day I had to go down a water slide about fifty times while the cameraman tried to capture of video of me smiling for a waterpark commercial. (It was cold. I was in 2nd grade. I don’t think I ever smiled.)

If you look on the internet you won’t find any of these moments from my childhood. But if you could re-wind the clock, you would have seen my mom (my actual mom, not the home insurance mom) standing off to the side and watching everything. I had an employment contract and was represented by a modeling agency. I was earning money that my parents put into a savings account until I was 16 and ready to buy a car.

There was mutual benefit for the companies wanting to market their business and for me, a young child. And my mom was there to keep a watchful eye on everything because, after all, I was a minor. Legally (and developmentally) I wasn’t expected to advocate for myself or understand all of the details of what was happening.

Fast forward to today and kids are constantly having their photos taken by companies, non-profits, schools, and churches for similar marketing campaigns. What previously required organizations to spend actual money, make plans in advance, and allocate resources is now done for free on social media with a few clicks and taps. Hashtags and employees with iPhones in hand have replaced employment contracts and agency representation. This has been building for years, yet few of us have taken the time to step back and think critically about what this means for the safety and wellbeing of children. Paid actors and print ads have been tossed aside for generic “photo release policies” and constant social media fodder.

As a result, we have set up scenarios where even young children know their images are being posted online to the illustrious Instagram. We spend years creating digital footprints and then are surprised when our children clamor for access to such platforms, even as we speak openly about the harms which await them online.

It is a funny thing that as a culture we have allowed this to happen. I grew up in the 80s and 90s when child actors were frequently making headlines for abusing drugs and alcohol or even becoming victims of suicide. Their mental health struggles were always linked back to their rise to fame and access to materials inappropriate for their age. Certainly there were exceptions (like Kirk and Candace Cameron), but that was just the thing—they were the exceptions.

The relatively small amount of acting I did as a child may have been tied to my lack of natural skill (see: my inability to smile on a cold water slide in winter), but it was also due to the restraints my parents intentionally put into place. Years after my days of acting had ended, my mom told me that I had been invited to a call-back audition for a big budget film. Once my mom learned that the film contained scenes of child abuse, she politely declined the call-back audition. To this day, not only have I never watched the film, which received an R-rating, but I remain grateful to my mom for protecting me from a situation potentially harmful to my mental health.

In a world where the mental health crisis plaguing children and teenagers is widely accepted as being linked to social media and smartphone use, shouldn’t we begin putting restrictions on how often and for what purposes minors’ images appear online? There are a growing number of parents who are purposely electing to not post their children’s faces online. Some are opting for complete abstinence while others are cleverly using photography skills or well-placed emojis to obscure their children’s faces. Organizations should follow suit.

Just because something is ubiquitous and has been normalized doesn’t mean it is good. With the rise of AI deepfake pornography and online identity theft, it’s quite possible that putting minors’ images online makes them susceptible to being unintentionally cast in the modern-day equivalent of an R-rated film.

So how should organizations in the modern age promote and market themselves? Clearly social media is a permanent fixture in our society and an effective means of communicating with constituents and customers.

I suggest we go back to intentional, not haphazard, marketing campaigns. Churches, schools, youth sports, and other extracurriculars should enter into employment contracts with paid child actors who will be governed by laws similar to those which have governed child actors for decades. Instead of live updates, curated collections of well-planned photos can become the norm. Many adults are exhausted by the tidal wave of content streaming to their smartphone anyway. If these images cost institutions something, they would have to be more restrained. Of course even under such an arrangement, there will continue to be some parents who will allow their children more access and adult-level freedom than is prudent, but at least it will be done with representation and some level of oversight. It is my belief that most parents, even with the promise of a paycheck, would decline requests to have their children featured in online marketing campaigns anyway.

In 2025 we may think it is silly to place an advertisement in a printed newspaper, but I would argue that one day we will all think it is silly to constantly photograph children and hope the post “goes viral.” Think about the phrase itself. Is spreading our young people around the internet like an out-of-control virus wreaking havoc really the goal for which we are aiming? Maybe the four walls of our homes are the safest places to display childhood memories after all.

Image Credit: Henry Mosler, “Children under a Red Umbrella” (1865) via Picryl.

In this fifth and final guest video, Ben Syversen discusses a question anyone who has done ruler and compass constructions for a geometry class may have wondered: What’s the point?

There is a lot about Euclid’s Elements that is easily misunderstood. Arguments that seem to have logical gaps, some constructions that seem pointless, others that seem needlessly convoluted. But each of these actually provides a window into how the ancient Greeks thought about math, and the philosophical role that geometry played.

TL;DR is the part in bold below.

The field of psychology had a big crisis in the 2010s, when many widely accepted results turned out to be much less solid than previously thought. It's called the replication crisis, because labs around the world tried and failed to replicate, in new experiments, previous results published by their original "discoverers". In other words, many reported psychological effects were either non-existent—artifacts of the experimenter's flawed setup—or so much weaker than originally claimed that they lost most of their intellectual sparkle.

(The crisis spanned other fields as well, but I mostly care about psychology here, especially the cognitive kind.)

This is very old news, and I've been vaguely aware of several of the biggest disgraced results for years, but I keep on forgetting which are (still probably) real and which aren't. This is not good. Most results in the field do actually replicate and are robust^{[citation needed]}, so it would be a pity to lose confidence in the whole field just because of a few bad apples.

This post is a compact reference list of the most (in)famous cognitive science results that failed to replicate and should, for the time being, be considered false. The only goal is to offset the trust-undermining effects of my poor memory—and perhaps yours, too?—with a bookmarkable page.

This can't be a comprehensive list: if a study is not on this page, it's not guaranteed to be fully replicated. Still, this should cover most of the high-profile debunked theories that laypeople like me may have heard of.

Credit: I enlisted the help of Kimi K2, o3, and Sonnet 4 to gather and fact-check this list. I also checked, pruned, and de-hallucinated all the results.

Ego Depletion Effect

• Claimed result: We have a "willpower battery" that gradually depletes during the day as we exercise self-control. (I remember reading Baumeister's pop-science book and being awed by the implications of their findings; I might have known it sounded too good to be true.)
• Representative paper: Baumeister et al. 1998
• Replication status: did not replicate
• Source: Hagger et (63!) al. 2016

Power Posing Effect

• Claimed result: Adopting expansive body postures for 2 minutes (like standing with hands on hips or arms raised) increases testosterone, decreases cortisol, and makes people feel more powerful and take more risks.
• Representative paper: Carney, Cuddy, & Yap (2010)
• Replication status: did not replicate
• Source: Ranehill et al. (2015)

Social Priming: Elderly Words Effect

• Claimed result: People walk more slowly after being exposed to words related to elderly stereotypes.
• Representative paper: Bargh, Chen, & Burrows (1996)
• Replication status: did not replicate
• Source: Doyen et al. (2012) (I like how they prove that the psychological effect was actually in the experimenters, rather than the subjects!)

Money Priming Effect

• Claimed result: Simply thinking about money makes you more selfish and more likely to endorse free market values.
• Representative paper: Vohs, Mead, & Goode (2006)
• Replication status: did not replicate
• Source: Rohrer, Pashler, & Harris (2015)

ESP Precognition Effect

• Claimed result: In some cases, people can predict future events "that could not otherwise be anticipated through any known inferential process".
• Representative paper: Bem (2011)
• Replication status: did not replicate
• Source: Galak et al. (2012), Ritchie, Wiseman, & French (2012)

Cleanliness and Morality Effect

• Claimed result: Being clean or thinking about cleanliness makes people more morally lax.
• Representative paper: Schnall, Benton, & Harvey (2008)
• Replication status: did not replicate
• Source: Johnson, Cheung, & Donnellan (2014)

Glucose and Ego Depletion Effect

• Claimed result: Connected to the debunked ego-depletion effect, this one claims that adding glucose to your blood "recharges" the willpower battery. (For a while, I may have drunk more orange juice than usual after reading Baumeister's book. At least it's healthy-ish.)
• Representative paper: Gailliot & Baumeister (2007)
• Replication status: did not replicate
• Source: Lange & Eggert (2014)

Hunger and Risk-Taking Effect

• Claimed result: People exposed to the scent of freshly baked cookies become less sensitive to risk and take more risks to obtain food.
• Representative paper: Ditto et al. 2006
• Replication status: did not replicate
• Source: Festjens, Bruyneel, & Dewitte (2018)

Psychological Distance & Construal Level Theory

• Claimed result: "Psychologically distant" events are processed more abstractly, while "psychologically near" events are processed more concretely. E.g., you worry about the difficulty of a task if you have to do it tomorrow, but you see the same task's attractive side if it is planned far in the future.
• Representative paper: Trope & Liberman (2010), building on Liberman & Trope (1998)
• Replication status: serious credibility problems
• Source: A collaboration between 73 labs around the world is vetting this theory right now because of many doubts about its validity.

Ovulation & Mate Preferences Effect

• Claimed result: Women are more attracted to hot guys during high-fertility days of their cycles.
• Representative paper: Gildersleeve, Haselton, & Fales (2014)
• Replication status: did not replicate
• Source: Stern, Gerlach, & Penke (2020)

Marshmallow Test & Long-Term Success Effect

• Claimed result: Children's ability to resist eating a marshmallow when left alone in a room at age 4-5 strongly predicts adolescent achievement, with those who waited longer showing better life outcomes.
• Representative paper: Shoda, Mischel, & Peake (1990)
• Replication status: did not replicate significantly
• Source: Watts, Duncan, & Quan (2018)

Stereotype Threat (Women's Math Performance) Effect

• Claimed result: Women risk being judged by the negative stereotype that women have weaker math ability, and this apprehension disrupts their math performance on difficult tests.
• Representative paper: Spencer, Steele, & Quinn (1999)
• Replication status: did not replicate
• Source: Flore & Wicherts (2015)

Smile to Feel Better Effect

• Claimed result: Holding a pen in your teeth (forcing a smile-like expression) makes you rate cartoons as funnier compared to holding a pen with your lips (preventing smiling). More broadly, facial expressions can influence emotional experiences: "fake it till you make it."
• Representative paper: Strack, Martin, & Stepper (1988)
• Replication status: did not replicate
• Source: Wagenmakers et (54!) al. (2016)

Objective Measurement of Biases

• Claimed result: You can predict if someone is racist by how quickly they answer certain trick questions.
• Representative paper: Greenwald, McGhee, & Schwartz (1998)
• Replication status: mixed evidence with small effects
• Source: Oswald et al. (2013) shows that the prediction power is small at best.

Mozart Effect

• Claimed result: Listening to Mozart temporarily makes you smarter.
• Representative paper: Rauscher, Shaw, & Ky (1993)
• Replication status: did not replicate
• Source: Pietschnig, Voracek, & Formann (2010) (What a title!)

Growth Mindset Interventions

• Claimed result: Teaching students that intelligence is malleable (not fixed) dramatically improves academic performance.
• Representative paper: Dweck, & Leggett (1988)
• Replication status: mixed results - many failed replications but also some successful replications
• Failed replication source: Li & Bates 2019
• Notable successful replication: Yeager et al. 2019 in Nature

Bilinguals Are Smarter

• Claimed result: Being bilingual provides substantial cognitive advantages in attention, task-switching, and executive control.
• Representative paper: Bialystok, Craik, & Luk (2012)
• Replication status: did not replicate
• Source: Lehtonen et al. 2018

Did I miss any famous debunked studies? Let me know by replying to this newsletter, and I'll add it to the list. ●