Twitter’s emphasis on text has me of two minds. As someone who’s been on the receiving end of harassment, I appreciate the power to skim a message and quickly remove it from my sight. On the other hand, sometimes Twitter is a lonely service: a way to communicate with people without the comfort of hearing someone’s voice.

That’s part of what drew me to Postconscious, a web experiment that brings speech to the platform.

Berlin-based designer and developer Andrew McCarthy created the site. It pulls posts from the platform, sans handle or urls, on whatever topic you search and reads them back via Speech Cloud’s text-to-speech. McCarthy, who describes himself as being drawn to technology but “very weary of it,” is fascinated by the idea of simulation theory — the idea that humans are just characters living inside of a base reality. That it’s possible “we are all one.” Consciousness is part of that system, he says, makes a lot of sense.

“I could see our consciousness being a module of that system, like a database, where we only have access to our own entities, but the whole thing is connected and aware of itself.”

This is half of the inspiration for Postconscious; the other half belongs to Werner Herzog’s Lo and Behold. McCarthy points to a specific line, in which Herzog suggests the idea that if artificial intelligence were sentient, it wouldn’t necessarily tell humans.

“I’m not hoping to achieve anything, really,” he says. “I see it more as an art project. If it makes people contemplate, or feel any kind of emotion, then I’m happy. Even if it’s just a quick lol.”

As it exists today, the site has a few suggestions to get you started, like “politics,” “#love,” “?,” or “sex.” When I tested the site myself, I ran the politics search before wondering what people were saying about topics like “trump,” “gamergate,” and “harambe.” Postconscious dutifully spewed back a few innocuous tweets intermingled with swearing and Hot Takes.

“The 140-character limit makes it perfect for spitting out these little nuggets of wisdom,” McCarthy tells me. “It’s a massive platform with a super diverse audience, so you get really thoughtful, interesting and funny stuff, along with mindless garbage.”

The weirdest thing about this experiment are the vaguely robotic voices that filter through: male, female, young, old. They even have different accents. If you sit on the home page long enough, the site begins to rattle off tweets at random, as if it’s in conversation with itself.

“From the reaction I’ve seen so far, people are still a bit freaked out by synthesized voices,” McCarthy says. “...There is some good intonation and inflection, but it still has this robotic, uncanny valley thing going on. I think the child voices don’t help either.”

Postconscious can be a fun and novel tool, but its voice is ultimately at the mercy of the tweets of strangers. Instead inputting the most vile or inflammatory prompts you can imagine, I would encourage you to search things like “dogs.” The results for me included what is now my favorite tweet of all time: “I am the kind of person who says hi to dogs.”

In a possibly ill-advised venture, I also gave my own name a search. Postconscious consistently returned a single tweet: “I wasn’t even hungry but I ate too much anyway: The Megan Farokhmanesh story.”

That would be a sick burn if it weren’t so accurate.