Write a commentA recent episode of the Audio Unleashed podcast touched on something important to audio enthusiasts of all kinds. Important enough that I felt it worth diving into a bit more here. If you haven’t heard of the Audio Unleashed podcast, it’s cohosted by my predecessor here at SoundStage! Solo, Brent Butterworth, and the editor of SoundStage! Access, Dennis Burger. The topic was AI, and if you’re Very Online like I am, I’m sure you’re tired of hearing those two letters even remotely near each other. More than a buzzword, AI has the possibility of changing the web in dramatic ways. None of them, in my opinion, good. But we’ll get to that. Where it’s going to directly affect you is the increasing potential that you’re going to stumble upon an AI-generated review, or even more likely, an AI-generated review aggregator. First, though, a quick primer.
The advent of AI
The technology colloquially referred to as artificial intelligence, or “AI,” isn’t really AI. This software doesn’t “think” per se as much as it creates a mashup of words and phrases related to the topic submitted by a user. They’re more accurately called large language models (LLMs). Think of it this way: Imagine the entire text of an encyclopedia. Could you, given enough time, take words and phrases from just that text and create a summary of a topic not in the encyclopedia? Now imagine not just the text of an encyclopedia, but all of Wikipedia, or all of Reddit, or huge chunks of the entire internet. What kind of content could you make then? What topics could you describe? While this description is overly simplistic, it’s the basic idea of “AI” like ChatGPT, PaLM, LLaMa, and so on.
There are some interesting uses for this technology, but we’re going to focus on the more nefarious. Or if not nefarious, at least potentially dishonest. It’s easy enough for a user to prompt it to “create a review of the Sony WF-1000XM5 earphones” and, with a few additional qualifiers, come up with something that reads like a real review. Already, there are websites doing something exactly like this, complete with AI-generated “reviewer” photos and bios. I’m not going to link to them and give them the traffic and Google juice, but they exist.
I’ll start with the plagiarism aspect with the full caveat that I am not unbiased in my dislike of AI. The only way LLMs like this can create any semblance of a useful result is by rearranging established data. In the above case, it would pull from existing reviews of the Sony XM5s. Those reviews would be written by me and my fellow headphone reviewers. Essentially, they’d be “copying our homework.” Now if someone’s just using this for their own personal use, OK. However, if someone is using it to generate their own “review” website and making money from it, that’s plagiarism.
The logical end to that path is people like me being out of a job. These reviews aren’t better than what I and my fellow reviewers can create, but generating them is essentially free. Faced with the decision of “good and costly” and “mediocre and free,” far too many website owners will choose the latter. In that eventuality, the AI ouroboros would have nothing to feed on but itself, generating meaningless and useless content. No one’s homework to copy, so to speak. Would that matter to the owner of the unscrupulous AI-created content site? Probably not. It might matter to you if the only reviews you can find are AI-generated nonsense that doesn’t accurately describe how a product performs.
The other possibility is a little less troubling, and probably even more likely: AI-generated “best” lists. Honestly, these aren’t that different from their human-generated counterparts. After all, how much effort did a person making $10 per post put into their “10 Best True Wireless Earbuds $100–$120” list. They mentally aggregated a few sites, shuffled the pics, and posted it. An AI would do the same thing. I’m not condoning a machine taking even a crappy job, but for the reader, the result probably isn’t too different.
Due diligence
So what can you do? Well, I can’t expect you to memorize the names of all your favorite reviewers (other than me, of course!). Being aware of a few key names is certainly a good safeguard though. Additionally, be wary of any website you’ve never heard of. While it’s entirely possible new review websites will creep their way to the front page of Google, or get linked to on a forum or Reddit, in this day and age that would take a substantial investment of time and money. If you find a website where the earliest content is three months old and there are 2000 reviews, they are almost certainly fake, AI-generated trash.
Lastly, and certainly most self-servingly, if there are sites and reviewers whose work you enjoy, frequent those sites. This is the reviewer version of the YouTuber “smash that like button.” Support the people whose work you like, otherwise that work might disappear.
. . . Geoffrey Morrison
