Let’s talk about social context. Somebody with a stake on their long-term reputation with you won’t use AI-generated text, or will use it in a careful way. Nearly unlimited plausible but not necessarily useful text is only worth publishing, posting, or using by actors that are maximizing total views rather than the reputation of the individual account, page, or entity that is publishing it.
In other words, nobody you should trust will use it, and anybody who would use it is already somebody you shouldn’t be paying attention to: the incentives that made ChatGPT interesting (quantity over quality, plausibility over truthfulness) are already in play – it lowers the cost of Internet-scale bull, but that’s already here.
It feels like a danger to the Internet as we know it because most of the Internet most of the people experience most of the time runs, technologically and financially, on the concept of replacing social capital with algorithmic reach: you don’t see the output of people and organizations you deem interesting and trustworthy, you see the output of people who have convinced a company’s systems to show to you.
We can call this the Google Experiment: it marked the transition from an Internet where reach and reputation were negotiated on a per-case basis to one where this was offloaded, as it were, to PageRank and its successors. Google profited from this, and so did the Internet as a whole, arguably, but we can already see the unavoidable side effects of this model. Social networks, Twitter perhaps the most, are the reductio ad absurdum of this strategy, places where you have surrendered so much control over what you see that the ability to block becomes absolutely necessary. Imagine a library where instead of looking for books you needed a bat to hit back sales catalogs and neofascist pamphlets trying to stick to your face.
The usefulness and unavoidability of AI-generated content in these contexts is plain: AI-augmented, massively-automated actors have huge advantages in any algorithmic setting (not just in terms of scale, but also in flexibility). And these are by definition actors going for reach and influence, not quality or usefulness: bad content drives out good in algorithmic settings because the incentives and tools favor the bad.
And yet… so what?
The huge centralized sites work, and can only work, through replacing individualized reputational capital with algorithmic sorting, but there’s nothing in the concept of the Internet that makes this necessary or particularly useful. Search-and-scroll might be the current default verbs to explore the Internet, but the fact that they are becoming worse by the year — with AI accelerating but not initiating this process — doesn’t make the Internet as such useless.
We can see the growth of newsletters (and, I hope, the resurgence of RSS) as a reaction to this. Their main feature from this point of view is that every newsletter or RSS feed you subscribe to comes from one specific person, organization, or group, one you have a certain opinion of based on your experience with them. Maybe you follow them because they already have good reputation with you from some other context, maybe they are recommended by somebody you trust, maybe they created something good that gives them credit with you; most importantly, you can unsubscribe from them specifically, so they have an incentive to develop and maintain a reputation with you that no throwaway bot or autogenerated site has.
It might feel weird, or overwhelming, to think about moving around the Internet as if it were an archipelago of sites from individuals and specific organizations, blogs, newsletters, etc without the teleportation capabilities of search engines or the smooth cornucopia of a social network; to jump from somebody you trust to somebody you trust, building your own map, an structure of information flows like a spiderweb over the mass of things you ignore.
But the Internet as a whole, weighted by word as it were, is already an ocean of things you wouldn’t really look for, a lot of it actively poisonous, its quality diluting by the month. AI-generated content will only accelerate this. Building your own Internet as an archipelago of the trusted and the interesting, a site-of-sites you control absolutely because it’s by (your) invitation only and you can remove from your view, quickly and forever, anybody who betrays your trust, is an strategy open to every person and organization, perhaps the best way to obtain value from what still remains one of our greatest technological achievements.