What does the mascot of Pringles, the Zimbabwean dollar, and the Playboy Bunny have in common?
They’ve all been subject to a crazy little thing called citogenesis.
What on earth is citogenesis?
Citogenesis is the process of creating “reliable” sources through circular reporting.
Okay, so what is circular reporting? Another good question. Circular reporting is when false statements are taken from a generally trusted source, such as Wikipedia, and passed off as truth in an article or news story.
Readers of said article or news story then check Wikipedia (other online sources are available), and flag the fact for review, wherein an editor will then find the inciting article, and add it as a citation on the Wikipedia page the fact originates from.
With a “real” source created, the false statement gets spread wider and wider, snowballing down the mountain of lies as more citations added, creating a feedback loop where fiction increasingly masquerades as fact.
The word ‘citogenesis’ was coined by American cartoonist, Randall Munroe (and good lord did I triple check that to avoid the biggest case of irony ever to stain the internet) best known as the creator of the webcomic xkcd.
In said webcomic, Munroe outlined the four-step process of citogenesis and how easy it is for a fib to become fact when complacency takes precedence.
Let’s look at those examples I teased at the start of this article.
Can you name the mascot of crisp masters, Pringles? If you called him Julius Pringles, then you, my friend, have just fallen for a classic case of citogenesis.
The year was 2006. Two North Carolina State University freshmen, Justin Shillock, and Michael Wiseman, updated the Wikipedia entry for the Pringles mascot - formerly known as Mr. P - to include the name “Julius” after Wiseman conjured the name from the name of a player in an American Football game. The name was quite quickly taken down, but Shillock simply reinstated it with the ‘citation needed’ tag, where it remained unchallenged for years.
Fast-forward seven years - during which time the Pringles brand was sold to Kellogg’s - and the name Julius Pringles started popping up in their own marketing material.
This is actually quite a unique case of citogenesis, as the legitimacy of the name has now become absorbed by the Pringles brand.
And this is far from the only case. In 2008, the country of Zimbabwe experienced a case of severe hyperinflation due to an ongoing period of currency instability and economic mismanagement. While the fallout of this hyperinflation was one of the worst cases in history, the severity itself was hyperinflated.
Reports spread worldwide that at the height of its instability, one US dollar was worth the same as an astonishing 2,621,984,228,675,650,147,435,579,309,984,228 (or according to ChatGPT - 2.6 decillion) Zimbabwean dollars.
This flabbergasting fact came from a single-edit IP, which claimed that the number was a “zillion” also stated that the country’s unemployment rate had hit 800%. While these facts were corrected within a matter of days, the 2.6 decillion figure stuck around for 10 months before it was reverted - just enough time to make it into a book entitled “Stuff You Should Know: An Incomplete Compendium of Mostly Interesting Things”. Ironic.
Before we move on, let’s look at that final example: the Playboy Bunny debacle.
Since 2010, a Wikipedia article about the Playboy Bunny insisted that Hugh Hefner claimed the costume’s inspiration came from Bunny's Tavern in Urbana, Illinois. It even went so far as to assert that Hefner “formally acknowledged” this origin in a letter that's now on permanent display at the tavern.
Yet, no credible sources have ever supported this notion. Nonetheless, this unsupported claim rapidly took on a life of its own, spreading via a 2011 book, a New Straits Times article from 2011, and even a 2017 piece in The Sun entitled “This is the real reason that the Playboy girls were called Bunnies” (which the New York Post later adopted, too).
As if that weren’t enough, the tale seeped its way into the French, Spanish, Catalan, and Italian Wikipedias, each falling prey to the self-fulfilling, spiral-of-misinformation magic that is citogenesis (I’ve used that word so much already it doesn’t look real any more).
Ladies and gentlemen, welcome to the internet.
Okay, so if citogenesis has been around for decades, why are we talking about it now?
Generative AI, that’s why.
Advancements in GenAI are coming thick and fast. It was only a few years ago when those janky generated videos of Will Smith eating spaghetti flooded the internet. While silly, it was clear from the off that the video you were watching was AI-generated.
Cut to today, and that line isn’t so clear.
One needs only to glance in the direction of cutting-edge AI applications to witness the terrifying speed with which these models are improving. Those mismatched graphics, questionable spellings, and Uncanny Valley-esque faces featuring in images and videos from even just a few months ago are quickly becoming a thing of the past. The change is like comparing a child’s drawing to the works of Michelangelo or Da Vinci.
But generative AI is far from perfect, and therein lies the biggest problem.
In this frequently alarming new world, we now face a concern delightfully dubbed “AI hallucinations”.
AI hallucinations occur when AI models confidently spout facts and figures that bear as much relevance to the truth as Shrek does to the Declaration of Independence (which is to say, none at all).
There have been some pretty noteworthy misfires that highlight the problematic nature of AI hallucinations.
In 2023, Google released a promotion video of the Bard chatbot which incorrectly claimed that the James Webb Space Telescope was used to take the first pictures of exoplanets, when in fact it was the European Southern Observatory's Very Large Telescope in 2004.
This mistake - accompanied by the belief that Google was moving too slowly in the large language model (LLM) and generative chatbot-sphere when compared to ChatGPT - knocked the stock price of Google owner Alphabet down by 100 billion US dollars within hours.
Similarly, a lawyer in New York got into hot water when it was discovered that legal research his firm had sourced from ChatGPT came replete with six entirely made-up legal cases, resulting in the first instance of AI-induced misinformation tainting the court of law.
Does this mean that AI has a predilection towards being a pathological liar? In reality, AI is still in its infancy. It’s easier to think of GenAI as like a child eagerly telling a story. The development of their mind is not too unlike that of these AI models.
The core of these AI missteps is the same underlying principle that enables citogenesis - a prioritisation of plausibility over precision. LLMs are engineered to predict the next word based on patterns in the data they are fed, not to discern the truth from fiction.
It’s an incredibly fragile process. Just the slightest change in the phrasing of a question can send the AI model down a completely different narrative path, and when the inputted data is less than pristine, the problem is magnified.
It’s a case of “garbage in, garbage out”, and - much like the pollution of citogenesis on trusted sources like Wikipedia - AI hallucinations risk seeping into the broader information landscape.
This then creates its own, deeply problematic version of citogenesis. AI models require large data sets to learn from, and as these data sets become infested with misinformation, the generated results will become more and more diluted. As the world leans evermore heavily towards an AI-assisted future, this spiral of dissemination poses a massive problem.
What makes things worse is that as these trusted sources are drowned in the quagmire of inaccuracy, they risk losing their stature as a credible source. If users are constantly met with information they don’t trust, they will look elsewhere to sources with their own agendas and biases.
The very foundations of academia are at risk. AI-generated research that circumvents the rigour of traditional academic study inflates citation metrics, thus misrepresenting genuine scholarly contributions.
But surely social media makes this problem easier?
Oh, you sweet summer child…
I’m sure most of us have at some point seen a post explaining a point, or sharing a fact that we inherently take for granted.
This is what is known as social information. Simply put, we think what other people are thinking.
The digital age has turned this concept of social information into a whole new beast, one Danish researchers Vincent F Hendricks and Pelle G Hansen called an “information storm”.
As people share facts out into the world, anyone subsequently encountering that information has to make the choice of whether or not they will believe it. Assuming they have no prior knowledge of the claim, they could look it up - potentially encountering our good old friend citogenesis along the way - or they could follow a far simpler method of evaluation.
“When you don’t possess sufficient information to solve a given problem, or if you just don’t want to or have the time for processing it, then it can be rational to imitate others by way of social proof,” Hendricks and Hansen explain.
Take this as an example.
We’ve all heard the phrase “The customer is always right” right? It’s often used as a motto in the customer service industry to prioritise customer satisfaction. The phrase is believed to originate from retail magnate, Harry Gordon Selfridge, or his one-time employer, Marshall Field, and for years, everyone willingly accepted the quote as verbatim.
But then along came TikTok.
Videos started to emerge on the popular platform claiming that the full phrase was “The customer is always right in matters of taste” effectively shifting the narrative from wilful subservience to a matter of opinion. The videos even went so far as to directly attribute this longer quotation specifically to Selfridge.
As the subject gained interest, some content creators further embellished it with AI-generated images showcasing the updated slogan in an effort to improve its legitimacy. Yet when fact-checking website Snopes investigated the matter, they found no evidence to support it.
But by that point, the new truth had spread far and wide, leaving many viewers content in the knowledge that what they knew was true.
It seems then that the problem isn’t solely with GenAI, but with the people using it.
Why do people post false information?
Allow me to answer that question with a quote by Peep Show’s Super Hans:
“People like Coldplay and voted for the Nazis. You can’t trust people.”
But in all seriousness, trying to understand the answer to that question is a subject as challenging as trying to unravel the tangled weave of the human psyche.
In the case of Julius Pringles, Shillock has since explained he updated the Wikipedia article as a joke. It was 2006. Misinformation on the web was far less of a concern back then.
Twenty years on, the effects of Pringlesgate (as I will fruitlessly attempt to coin it) is a stark reminder of how easy it can be to propagate false information online, and how infrequently people actually check sources.
What was a joke for Shillock and his roommate is an opportunity for much more nefarious individuals.
Without naming names and giving this thought piece a political taint, there are more than a few figureheads that use - and actively encourage - the spread of misinformation for political and personal gain.
If generative AI can be used to hastily disgorge vast amounts of “proof” then the process of citogenesis becomes even more complex. To what end does a Wikipedia editor have to shift through myriad articles, videos, reactions, and posts before they declare the legitimacy of a fabricated fact?
And if the masses wilfully adopt fiction as fact without pause for thought, then what hope do we as a species have?
Are we headed for the terminus of truth?
Will we smash through the Rubicon of reason?
Have we crested the climax of common sense?
And am I anchored to an age of aimless alliteration?
Please tell me there is a glimmer of hope?
There is, but as a wise Yorkshireman once said, “nowt o’ worth ever came easy”. (It was me. I just said that).
A multipronged approach is required to tackle the problem of AI-generated citogenesis. How many prongs? Let’s say three.
Prong number one tackles the top of the chain. OpenAI, Google, Microsoft, X, Apple, and all the other big names carving out a slice of that AI pie have a duty to uphold the validity of their AI models.
One such example is a greater push towards machine learning models that are trained to identify discrepancies, cross-reference citations, and flag inconsistencies. Building this responsibility into the model itself will help to circumvent the hallucination effect, and make AI accountable for its own mistakes.
This too would trickle down to the human element. If a statement cannot be supported with proof, the computational prowess of AI can flag it as suspect long before it has a chance to trickle out into the public consensus.
Thankfully, it seems the tech giants are already at work. Humbled by their public errors, Google is working hard to not only put security and ethics at the forefront of their AI ecosystem, but are making it accessible to the public, so future AI applications are built on a firm foundation.
Prong two puts the onus on Governments and policymakers. We exist in what may one day be referred to as “The Golden Age of AI”, which - like the lawless times of piracy and plunder that predated it - cannot continue to exist without rules and regulations. The sooner these restrictions are brought into law, the sooner we can curtail the dissemination of false facts.
That’s not to say that AI should be banned or stifled. If one thing is clear, AI isn’t going away. It’s not enough to just bury one’s head in the sand and pretend like AI won’t affect them. It’s got its hooks in us all. Better to be informed than ignorant.
At risk of sounding like a bitter Millennial, kids these days do not understand the ramifications of the world they are growing up in.
Deepfakes, cyberbullying, information privacy, job displacement, economic inequality. These are all things that AI has exacerbated. Given these topics, is it any surprise that a recent UK survey found that almost half of kids wish the internet didn’t exist?
But anything shy of a technological apocalypse is unlikely to see that world exist. Without a greater understanding of how to navigate the complexities of an AI-powered world, how can we expect the technology to be used ethically, sustainably, or safely?
Maths teachers once barked that we wouldn’t walk around with calculators in our pockets. And they were right. We don’t. We have something much more powerful at our disposal.
Educational initiatives need to exist. Schools should centre whole curriculums around the correct use of AI, not bolt it on as an afterthought. Forewarned is forearmed, if for no other reason than lifting the battered spirits of the next generations.
The focus of Governments and policymakers should be proactive, balancing regulatory frameworks that hold technology providers accountable, while enabling the ethical use of this transformative tech through education.
Lastly, prong three falls to us, the common folk. Simply put, don’t believe everything you see on the internet.
I’ll be the first to admit that it is way too easy to take information at face value. I fell for the mutated Selfridge quote. It took almost including it in an article for me to realise the error of my ways, by which time I had definitely told other people about it without thinking to fact-check first.
We need to take ownership of our own knowledge. Adopt the policy of doing your own research before taking something as fact, and actively challenge those who disseminate unverified information.
Just be sure to avoid arguments with random people online. Since when has someone swayed their opinion on a topic after being berated by a total stranger?
And that, ladies and gentlemen, is my three-pronged strategy - a Trident of Truth, if you will.
Let’s wrap this up then, shall we?
I don’t want to end this article on doom and gloom. For most of the time I’ve been writing this, my cat has been sat on the window sill, asleep and content.
There were times in the writing of this that I wished we could swap roles, that I could abandon the responsibilities of 21st century living in exchange for coughing up hairballs and eating the same biscuits for breakfast and dinner every day.
It’s easy to get swept up in the hysteria surrounding AI, but if you dial the clock back twenty years, the same was being said about the internet as a whole. The simple fact is that, like the internet, AI will continue to bring change to the world, and we’ll face it regardless of whether we like it or not.
I’d like to hope that the issue of citogenesis and GenAI can have a positive resolution, because while there are those who use misinformation for their own benefit, the vox populi remains the same.
We’re not an evil species. We don’t want to bring ill to others. I’m reminded of the incredible speech delivered by Charlie Chaplin at the end of The Great Dictator:
“You, the people have the power - the power to create machines. The power to create happiness! You, the people, have the power to make this life free and beautiful, to make this life a wonderful adventure”.
We don’t get to be sunbathing cats. It’s up to us to make the internet a safe and reliable space, together and individually. I’ll end this on one final Yorkshire adage:
“If tha wants summut doin’, do it thissen”.
Velstar: Doing AI Properly
If you want to harness AI in the right way, you’re in the right place. At Velstar, we’re taking AI seriously, providing positive and effective solutions for our clients.
If you’d like to learn more about how we can help your business implement AI, speak to the Velstar team today.
