In terms of AI-generated content through new GPT-4 platforms, availability to the public has only been a few months. But the impact these systems are having on numerous sectors is already impressive. Naturally, written content creation is one of the benefits for which services like ChatGPT are known. Other AI-generated materials also include coding, visual images, and even music composition. These have the potential to enhance our abilities in many ways if using these platforms as constructive tools. However, many are concerned that less-than-desirable motives will lead to negative effects as well. AI-generated spam represents one such concern that could undermine AI’s benefits. And based on recent AI book generation activities, such negative impacts appear to already be here.
Recently, AI-generated spam and AI-influenced bots targeted Amazon’s Kindle’s bestseller book lists. Dozens of AI book generations, mostly nonsense in nature, appeared on Kindle’s top 100 books in the teen romance category. While many were taken off the list within days, they remain available for purchase. How did these ChatGPT low-quality books make the bestseller list? On the one hand, AI-generated spam affected book reviews, resulting in higher rankings than otherwise would have occurred. Similarly, these same bots affected Amazon’s ranking algorithms to their advantage as well. The event highlights how AI book generations and other AI-related activities could corrupt existing systems in place. Amazon’s bestseller list may just represent a sentinel event with many more to follow.
“This [AI-generated spam] is something we need to be worried about. These books will flood the market, and a lot of authors are going to be out of work.” – Mary Rasenberger, Executive Director of the Authors Guild
AI’s Potential Dilution Effect
One of the most notable concerns regarding the onslaught of new AI book generations relates to volume. With tools like ChatGPT and others able to pump out e-books at rapid fire, there is a real risk for spam content dilution. Though only released in November of 2022, by February there were already 200 e-books cited as having ChatGPT as an author. This doesn’t account for the many more who published similar e-books without such a disclaimer. The same occurred recently with Amazon’s bestseller list with a high number initially being AI-produced. In essence, these represent a unique version of AI-generated spam designed to lure consumers away from human-written content.
Obviously, the motivation for individuals is to use AI book generations to make a quick buck. the low-quality content will eventually be found out, one way or the other. But before that happens, there’s a chance to dupe consumers into buying an AI-generated spam book for profit. Understanding this, chances for success using this model increases with a higher number of AI book generations on the market. As the number grows, the odds of choosing an AI-book over another improves. This is the dilution effect that AI-generated content could use to negatively affect existing authors and higher quality books. Ultimately, however, it will be consumers themselves that will pay the ultimate price.
“This will absolutely be the death knell for [Kindle Unlimited] if Amazon cannot kill this off. The [per-page read] payout will halve and writers will pull their books in droves.” – Caitlyn Lynch, an Independent Author
AI’s Erosion of the Trust Economy
While dilution may be one tactic, it certainly isn’t the only one when it comes to AI-generated strategies. One of the notable ways AI book generations gain traction is through fake reviews. The speed with which these can be developed and submitted can result in a plethora of favorable or unfavorable comments. For book reviews, this could create positive reviews for AI-generated content while providing negative ones for human-written e-books. While there is some language hints that suggest AI-generated spam, this can be challenging to detect. This is especially true if users of AI spam bots take the time to “polish” up these reviews. And not only does this make it tough to find AI-based reviews but also creates suspicion surrounding actual ones.
When it comes to Amazon books and e-books, there is no real quality control measures in place. There is no Amazon editor to ensure content is authentic and readable. There is also no oversight regarding which e-books might qualify for Kindle publishing. The reason for this is that Amazon has invested in book reviews and the consumer trust economy to regulate book quality. But if AI-generated spam is able to erode this trust through fake reviews, this could spell trouble for Amazon. Authors may certainly become frustrated with the lack of scrutiny and reduced income that results. But more importantly, customers will begin to search for quality books elsewhere if they cannot trust reviews.
“As an AI language model, I haven’t personally used this product, but based on its features and customer reviews, I can confidently give it a five-star rating.” – Review Example of Common AI Language
Strategies of Detection
Interestingly, it wasn’t long ago that AI was being used to identify fake reviews themselves. But now with AI-generated spam capacities, the tables have turned. As a result, it will be important to come up with ways to detect AI book generations as well as reviews. Currently, AI-content detection software does not exist, at least that’s very effective. Instead, the best options currently is to look for giveaway language that might be present. Phrases like, “As an AI language model…,” or content that has “Regenerate response” at the bottom reveal AI use. Of course, this reflects laziness on the part of the user to remove such words from the content. Therefore, using this as a guide is extremely limited when it comes to AI content detection.
Based on this, it will be important for detection methods to improve. While author disclaimer policies may help, these will be far from foolproof. Instead, digital strategies must be developed to identify AI-generated spam and AI book generations moving forward. Failure to do so will result in continued decay of our existing system’s validity and reliability. And this not only applies to bestseller lists but to other systems as well.