Social Media Nightmares

You post a piece on AI safety. It is probably your best popular explanation yet: short, clear, persuasive, and as accurate as it can be for the layman. It is widely shared. You notice something unusual. As soon as the post is published, several people are in the comments arguing against AI safety. They say AI safety is a bogus field. They call you a grifter. Their arguments are better than you are used to, and they keep replying. It takes you a long time and considerable effort to respond to them to your satisfaction.

The replies never fully stop. The new commenters keep commenting, and when you publish your next AI post more of them come. A lot about this is suspicious to you, but most of all that they seem to respond a little too quickly and without a single typo. And this is something anyone who knows you knows you would notice.

After a sleepless night, you pull some strings. A friend risks their career at Facebook to find the commenters’ IP addresses. Every one originates from one of several satellite ISPs, all IPv4. You list them chronologically and see the second to last byte is 0 in the first reply, 1 in the second, and so on. After 255 it cycles back to zero. It takes you about five seconds to realize there is a pattern in the last bytes, too. They are ASCII codes for English letters, and their message must be addressed to you.

A Ph.D. student in natural language processing famous for his entertaining demos follows you on Twitter. Soon his new bot joins in. “To simplify and situate language” is the stated purpose of the bot. It doesn’t seem to interact with you at first, but then a few days later you post a clever take and it replies. The bot’s reply is a thread of three tweets. The first rephrases the take in primary school language, the second maps it onto the political spectrum, current and historical (complete with a little JPEGy chart), the third relates in to three major writers (one alive).

You block the bot and the student.

Your anonymous feedback form brings you the message: ‘Wikipedia link broken on page /Foo.’ You got the same message yesterday and the day before. While the page changes, the phrasing is always identical, and the messages never lie. The reason your links break is on Wikipedia’s side. You download snapshots of WP and run some statistics. Articles you link to are disproportionally affected by renaming, splits and mergers, and deletion. New users register, rise in the Wikipedia hierarchy, and months later delete, rename, split, or merge articles you reference. You publish a page about this. Soon every Wikipedia link on the page is broken.

You start seeing ads for high-carb fad diets. They seem weirdly targeted. When one addresses you by name, you start digging. You trace the ads back to one person. They are a fan of your old writing.

You wake up to a Twitter DM. The bio of the account that sent it reads, “Combat AI at Lockheed Martin. I love cute things!” The profile picture is a drawing. The style is like a hyperrealistic mockery of anime. The drawing depicts Hatsune Miku holding a hostage Kagamine Len in headlock with a gun to his head. The background image of the Twitter profile is solid black. The DM says “The bad news, Michael, is that you’ve made an OPSEC mistake. It wouldn’t have been a serious mistake last year, but we’ve had major progress since. The good news? You will be rich writing for us.”

You visit the Area 51-themed cafe next to your workplace for the first time. You know it is part of a fast-growing chain. Its exterior has a lot of shiny chrome, and there are giant windows in its domed roof. You blog about recent trends from the cafe, comparing them to your predictions. You open Twitter in another tab and tweet a poll. When you switch back to the tab with your blog, you see there are two new comments. They say, “okay yhbt but that’s enough”, and, “look up lmao”. You look up through a window in the dome. There is a silvery disk hovering in the clear blue sky, completely still and silent. As you reach for the camera you have been carrying around, it disappears. UFO sightings become vanishingly rare.