A few weeks ago Commsrisk ran a story about gamers who cleverly demonstrated artificially intelligent automation was being used to turn their social media posts into ‘articles’ for an online magazine. Such is the power of AI: words get can be piped in from any source, and the result will look a lot like the kinds of grammatical sentences a human might produce, especially if the reader’s expectations have been influenced by the endless torrent of marketing bilge that already gets confused with meaningful insights. Now it is evident that readers of Commsrisk will find bogus AI-written scumsites appearing in their Google searches very soon.
Consider a phrase like “the role of graph analytics in shaping the future of fraud detection and prevention in the telecom industry” or a sentence like “anomaly detection is a powerful tool for identifying hidden patterns in vast datasets, enabling proactive network management, effective fraud detection, and insightful customer behavior analysis”. No person writes like this, unless they are being paid to promote a specific vendor’s products. Even I do not write sentences like these, and I actually know the meaning of ‘graph analytics’. But I do keep an eye on news concerning this kind of arcane topic, which is why I stumbled across a website with the not-so-catchy name of Fagen Wasanni Technologies that recently began pumping out content on these subjects plus hundreds of other subjects too.
It is at this stage that, were I speaking to an audience of people who do not know me, as opposed to writing for the wise readers of Commsrisk, I would need to divide my audience into two halves. One half would be people like you; the part of the audience who appreciate the true extent of the dangers, deception and douchebaggery that bedevils this world, and how often technology is perverted for the foulest uses. The other half of the audience would be those hopelessly naïve fools who will question how I can be sure that certain articles were produced by machines. My heart always sinks when confronted with people like this during my talks because they have instantly demonstrated they are not suited to any line of work associated with the topics I discuss. They have either wandered into the wrong room or will endure the far worse fate of being utterly incompetent at their job, though the incompetent managers that recruited them may never notice.
One way of telling an online publication is not written by real people is that a cheap-as-chips generic WordPress site is unlikely to appear overnight and instantly recruit five hard-working journalists who each produce over 20 different articles every single day. Fagen Wasanni Technologies started publishing content on June 30 and has published approximately 20,000 posts in the meantime. A website that produces very many articles about niche subjects like graph theory is not going to be able to afford the wages that would be demanded by the army of human researchers and wordsmiths that would be required to write so much about such obscure topics. And that is before I observe that the images associated with each article have literally no connection to the content of those articles. Two-thirds of the images are artificial compositions showing people with different faces sitting in similar poses whilst reading an assortment of newspapers and magazines.
Also, I know how to read English, and I have read plenty of it. I can tell the difference between text written by a human being with a genuine personality and text churned out by an unthinking automation. There ain’t no machine that is ever going to be mistaken for me.
The machines of Fagen Wasanni Technologies demonstrated their true nature by revealing the source of some of the content used in a recent piece entitled “Telecom Fraud Trends: Understanding the Growing Problem”. The closing paragraph copied the name of the business whose marketing spiel had been recycled.
Arkose Labs offers advanced security solutions to protect telecom businesses from fraud. Their technologies detect bogus account sign-ups, malicious logins, and persistent attacks on user touchpoints that trigger OTP verifications. With real-time visibility, Arkose Labs helps businesses prevent fraud early on, reduce costs associated with manual reviews, and provide a secure user experience.
This information helped me to track down the blog on the Arkose Labs website that was the source material for the Fagen Wasanni AI. The Arkose Labs original, which was only published on August 8, begins:
Telecom fraud is becoming a growing concern as is a complex issue that impacts telecom companies and their customers worldwide.
Whilst the Fagen Wasanni retread, which was published the following day, has this beginning:
Telecom fraud is a complex issue that affects telecom companies and their customers worldwide.
The original says:
Telecom fraud is not a new phenomenon, but it has evolved significantly over time with the advent of new technology and methods. In early forms of telecom fraud, individuals would use payphones to make free long-distance calls or sell stolen phone numbers. However, today’s fraudsters employ more sophisticated tactics such as hacking into phone systems or using fake caller IDs to deceive people into giving away personal information.
The retread says:
Over time, telecom fraud has evolved with the advancement of technology. Early forms of telecom fraud involved individuals making free long-distance calls or selling stolen phone numbers. However, today’s fraudsters use more sophisticated tactics such as hacking into phone systems or using fake caller IDs.
The original lists the following frauds:
The rise of mobile technology has also led to new types of fraud, including SIM swapping and SMS phishing attacks, along with SMS Toll Fraud or International Revenue Share Fraud.
The retread gives exactly the same names for the same frauds. Sophisticated AI models will know synonyms for common words but still do not possess the knowledge required to substitute different names for something specific. A human being can know that smishing is another name for SMS phishing, but AI has to copy names exactly as presented to them, including the way those names are capitalized.
The rise of mobile technology has introduced new types of fraud, including SIM swapping, SMS phishing attacks, and SMS Toll Fraud or International Revenue Share Fraud.
I could go on, but you have got the point by now. The Arkose Labs blog was written in a flaccid generic style representative of the superficiality of most marketing. But at least it was written by a person who had to decide what their article would cover, including the types of frauds that would be highlighted, and those which would be ignored. The AI knock-off reproduces the meaning taken from a human being’s work whilst changing the order and choice of words in order to reduce the risk of detection and to defeat copyright laws.
An additional problem with a rip-off website like Fagen Wasanni is that machines which flood words into one website can easily be turned into machines which flood words into multiple websites, further increasing the chances that somebody will read them, and further poisoning the public well of discourse. Having connected “Telecom Fraud Trends: Understanding the Growing Problem” from Fagen Wasanni to the Arkose Labs original, I then found a different AI variant of the same piece that was published on the same day at another website called ‘Gillett News’; you can see it here. This evidenced the same trait of repeating the meaning of Arkose Labs’ original blog but with a different arrangement of words. Then I found a third AI-retread at a site called ‘ISP Today’; that version is here. The most significant distinction between all three is that the Fagen Wasanni version was attributed to a made-up person with the name of Donovan Johnson, the Gillett News version was attributed to a made-up person with the name of Stacey Scott, and the ISP Today version was attributed to a made-up person with the name of Maurice Broaddus. The names of the made-up writers represent the only real creativity that went into any of these websites.
Protecting the intellectual property inherent to the written word has always been problematic. During the 19th century, Charles Dickens routinely pled with Americans to stop producing knock-off copies of his novels and to broker an international agreement to respect copyright. The American publishing industry and American legislators ignored him, and the USA continued to ignore the problem until the tables were turned and US businesses started losing lots of money because their original output was being copied elsewhere. That is why global copyright rules have largely been harmonized through the World Trade Organization. However, the internet has greatly exacerbated problems with protecting property rights associated with written words because it has simplified the task of scraping another person’s words and republishing them.
World leaders have shown little appetite to deal with the problem at a global level, except when they are responding to demands from dominant business interests like German publishers or Hollywood film studios. Perhaps that is unsurprising given that individuals like Ursula von der Leyen, chief executive of the European Union, and Joe Biden, chief executive of the United States of America, were able to shrug off the evidence of plagiarism in their earlier careers by asserting each instance was a ‘mistake’ rather than intentional. This defies credulity. How can somebody copy the work of another person, and fail to mention the name of that other person, and this all be some kind of accident? The only accident was that they were caught, and when powerful people are caught doing things which are wrong but common they tend to be forgiven more than the rest of us would be.
Even the anti-fraud industry has a problem with intellectual property theft, as ably demonstrated when the Telecommunications UK Fraud Forum (TUFF) stole content from this website. TUFF likes to talk a lot about enforcing the law but they still have a fundamental problem with respecting the purpose of the law. I was astonished that my research into Fagen Wasanni led me to TUFF’s website too. The ‘telecommunication fraud news’ column on their website is littered with articles from Fagen Wasanni, effectively rewarding plagiarists by promoting their work; you will see them on the bottom right of this PDF print of a recently visited page. This also suggests the anti-fraud experts at TUFF know so little about fraud, or spend so little time reviewing their own machine-driven output, that they actively promote unlicensed copies over the original source of insights.
Copyright was always an unsatisfactory way of defining and defending written intellectual property because it only protects the exact sequence of words. It is easily circumvented by rip-off artists who engage in wholesale copying of the meaning of other people’s work but who change the order of words and who make liberal use of synonyms to avoid an exact word-for-word match. Our electronic chickens have come home to roost now that AI is proving capable of evading copyright in the same way human drones have been doing for decades, but at a far larger scale than any human can emulate. The speed with which AI will be used to exploit other people’s work may finally force the long-overdue evaluation of how modern societies will reform intellectual property laws to defend the semantics used to express an original concept rather than defending the mere form of a syntactic series. Ironically, this will likely involve AI being used to rapidly judge and evaluate the likelihood of one piece of writing being derived from others.
Sometime during the last decade I stopped being nearer the start of my career than the end of my career. So my warnings about AI are not motivated by concern for myself. Just like global warming and finding a replacement for QWERTY keyboards, artificial intelligence is mostly a problem for future generations. My fear is for the younger readers amongst this audience, who may previously have been pressured to learn the drivel produced by human charlatans like Papa Rob Mattison, and may soon be forced to learn the drivel produced by the Fagen-Wasanni-Mattison-Weyland-Yutani AI consortium. I always tried to teach people how to think for themselves instead of just remembering what somebody else had told them. Imagine the consequences if a site like Commsrisk is replaced by a torrent of AI-composed drivel that is itself just a rearrangement of words taken from marketing announcements. This is not just some nightmare vision of a possible future. Organizations like TUFF are already perilously close to confusing AI-reprocessed marketing with reliable sources of specialist information.
A poet wrote that we should rage, rage against the dying of the light. From what I have seen during my career, including books written by computers and marketing spam masquerading as encyclopedia articles and deepfake presenters reading out press releases on YouTube news channels and outright unashamed plagiarism, our professional community already had problems with keeping the candle of human inspiration burning. Artificial intelligence could finally snuff it out, unless we rage against the dying of the light. Perhaps you might rage against the dying of the light by urging your peers to read the human-produced content of Commsrisk whilst they still can, instead of taking its existence for granted. Or you might devote a few moments to listening to the words of Dylan Thomas whilst asking yourself if you would want to live in a world where the synthetic burble of machines drowns the thoughts and passions of real men and real women.