People think they can tell if AI made something, but they can’t, and their anger is at anything that they think was AI generated.
getty
In today’s column, I examine how anger about the widespread adoption of generative AI and large language models (LLMs) is leading people to assume that they can readily recognize AI-generated content when they see it, even though they can’t. Their fury blinds them. Furthermore, when shown human-devised content and told a fib that it was generated by AI, people tend to still make disparaging remarks about the item because they believe it was AI-generated.
This is the AI-is-rottenness bandwagon running amok.
The crux consists of these newly emerging five rules-of-thumb:
- (1) People believe they can cleverly discern when something is composed by AI, but the reality is that they cannot actually do so.
- (2) If shown a piece of content that is not labeled as to how it was made, people are generally no better than random chance at guessing whether it was AI-generated or human-made (all else being equal).
- (3) If told that a piece of content is made by AI, people will disparage the item.
- (4) If a piece of content is made by AI, but a person is told it was human-made, the person will tend not to disparage it as they would if they thought it was made by AI (they give the benefit of the doubt to human handiwork).
- (5) People emotionally allow their preexisting bias against AI to materially shape their perception of content shown to them.
Let’s talk about it.
This analysis of AI breakthroughs is part of my ongoing Forbes column coverage on the latest in AI, including identifying and explaining various impactful AI complexities (see the link here).
AI Is Drawing Fierce Anger
People are getting very angry toward AI.
In one sense, you can’t blame them for this spate of anger. The news keeps saying that AI is going to take away everyone’s jobs. Companies claim they are laying people off due to the adoption of AI, though this might be a cover story for wanting to lay people off anyway. AI seems to be outshining humans. There is a lot of AI producing a lot of junk-filled outputs, fueling worries that the Internet is going to be chock-full of AI slop.
On top of all that angst is the assertion that AI is an existential risk to humankind. That type of talk has somewhat diminished. It was the mainstay of discussion last year. Pundits were declaring daily that AI would either destroy humanity or enslave us all. Not very promising and certainly a sign that maybe we should not be quite so enamored with the advent of AI.
A related concern is that humans are losing their place in the hierarchy of existence. Humans are supposed to be at the top of everything. The customary anthropocentric viewpoint is that humans prevail over all else. It is said to be a form of speciesism. Our species is the top banana. But, sadly and scarily, we find ourselves facing a cruel new change coming.
AI appears to be heading above us on the heap of existence.
Humans Assessing Artwork
Many empirical research studies showcase how far this anger has become.
One of my research study favorites is a now-classic entitled “Understanding How Personality Traits, Experiences, And Attitudes Shape Negative Bias Toward AI-Generated Artworks” by Simone Grassini and Mika Koivisto, Nature, February 19, 2024, which made these salient points (excerpts).
- “This study primarily aimed to understand whether individual factors could predict how people perceive and evaluate artworks that are perceived to be produced by AI.”
- “Additionally, the study attempted to investigate and confirm the existence of a negative bias toward AI-generated artworks and to reveal possible individual factors predicting such negative bias.”
- “The findings of the study showed that some individual characteristics, such as creative personal identity and openness to experience personality influence how people perceive the presented artworks in function of their believed source.”
- “Participants were unable to consistently distinguish between human and AI-created images.”
- “Furthermore, despite generally preferring the AI-generated artworks over human-made ones, the participants displayed a negative bias against AI-generated artworks when subjective perception of source attribution was considered, thus rating as less preferable the artworks perceived more as AI-generated, independently on their true source.”
The gist was that people tended to have a negative bias toward AI-generated artwork. This was true even if the artwork was devised by a human, in the sense that the experimenters told the people it was AI-generated, and they were disparaging despite the piece being human-made.
This tendency by people to be upset and disapproving of AI seems to be across the board. All ages, all demographics seem to feel this way. Maybe it is one of the last vestiges that unites us all. A common enemy, AI, brings us together in a rare moment of unification.
Still, some behavioral factors do come into play. The experiment consisted of examining human traits such as extraversion, agreeableness, conscientiousness, emotional stability, openness, empathy, etc. Sometimes, a particular human trait can moderate whether a person is going to be antagonistic toward AI artwork or more mollified.
Written Content Is In The Same Bailiwick
You might be thinking that there is a notable difference between distinguishing artwork and written compositions. Maybe people are more adept at discerning the source of written material than they are of artistic works.
Nope.
When ChatGPT was first launched in November 2022, the early versions of generative AI and LLMs were rather simplistic when it came to how they worded their outputs. Many of the popular LLMs at the time were apt to use words repeatedly in their generated answers. In addition, the selection of words was skewed.
As a result, you could potentially detect whether a written composition was likely made by AI. If the frequency of words and the use of certain words were appearing in a piece of content, you could somewhat reassuringly guess that it might have been made by AI. I wrote at the time that these guesses were not reliable – and that many people were going to be falsely accused of using AI to write something even when they didn’t use AI at all, see my coverage at the link here.
In other words, a person could write an item using their own mind and hand, but it might appear to be AI-generated due to their particular writing style. I pointed out later that if you wanted people to think that your handwritten material is AI-generated, you could trick people by writing in a specific fashion, see my analysis at the link here.
The Modern World Has Changed
Nowadays, you can enter a breezy prompt telling AI not to write in a predictable AI-like manner, and it will do a bang-up job of composing content that is essentially impossible to detect as AI-written versus human-written. Period, end of story.
Likewise, you can take a human-written composition and tell AI to make it seem like it was AI-written. Voila, people will fall for the clues in the composition that they “cleverly” think are a dead giveaway. Boom, drop the mic.
The irony is that automation, in this case AI, can be prompted to stir people to believe that something is AI-written in how the item is composed, and can also be prompted to fool people into believing something must absolutely be human-written, even though it was prepared by AI.
Do Not Be Foolish
If you know someone who loudly brags that they can discern AI compositions versus human ones, you might be safest to not argue with them. They will likely blow their top if you suggest they aren’t capable of doing so. Their humanism is anchored to an indisputable belief that, as a proud member of the human species, by gosh, they can tell the fake stuff from the real McCoy.
As per the wise words of Abe Lincoln: “You can fool all the people some of the time, and some of the people all of the time, but you cannot fool all the people all the time.” Well, maybe that needs a timely update due to AI; perhaps you can fool all of the people all of the time by leaning into the use of AI.
