Fire, water and poison in the Library of Alexandria

Technicians often say to anyone who will listen that the best time to create a backup is now, because disaster is inevitable and the only question is when it will happen. That’s more or less what we’ve learned, although we still get upset if, in an example I’ve completely made up for this column, you fail to recover a password and lose all the photos from those two years where you weren’t very happy but at least you lived by the sea and had palm trees, pink sunsets and alligators to look at, and you believed that at least you would have the memory of that.

We rely on the internet to store our intimate memories, but we have also delegated our collective experience to it, and the internet is fragile. If much of our life happens in its pages, how can we preserve it? “We can lose part of our memory as a society because a file format becomes obsolete,” the professor Nanna Bonde Thylstrup told EL PAÍS a few months ago. According to a study published in January by Ahrefs, at least 66.5% of all links created in the last nine years lead nowhere. Internet Live Stats estimated in 2019 that there were 1.7 billion websites, of which 99.9% were not active.

All you have to do is not renew a domain, or stop paying for server space, and the digital weeds will make their way through the ruins. Nothing remains of The Hairpin, which was one of the funniest and best-written publications on the internet, and is now a farm of AI-created content. Playground magazine lost its entire archive in a technological migration. In both cases, the youthful works of some of the most promising authors of the day disappeared. In February of this year, Google stopped storing a cacheable version of the websites it crawls. Archive.org still does a good job of preserving some copies of what is published [...] but when the daily life of a civilization happens within walled gardens, these efforts are not enough. Now, other dangers must be added to their disappearance.

Researchers are warning about the risk of internet poisoning, a paradox that holds that if the internet is saturated with large volumes of inane content generated with AI, its quality will be reduced and it will no longer serve to feed future language models. This scenario is starting to become real. Amazon has limited the number of books a user can publish per day to the very human number of three.

“It’s a little scary to see how AI-generated images are sneaking into Google Images. Searching for portraits from the 16th century, I detected three and among the first results,” the art historian Alegra García wrote a few days ago. Her professional colleague Montaña Hurtado responded by explaining that she had located works by Remedios Varo whose background had been enlarged with AI, which could make the non-expert viewer think that the original painting was derived from the artificial one, and not the other way around. There are those who, to avoid these types of errors, use a filter on Google that prevents content made after 2022 from being displayed, that is, after the emergence of ChatGPT.

It’s funny: we always assumed that our great collective construction was getting increasingly wiser, and we were more concerned with erasing our trail on the internet than with preserving it. There is no need to choose a metaphor. Alexandria can burn, flood and be poisoned at the same time.

Sign up for our weekly newsletter to get more English-language news coverage from EL PAÍS USA Edition