My Friend Has No Soul
The second version of the viral Friend AI necklace is louder and less interesting.
I'm on a Zoom call with my editor, Julian Chucatto, telling him about the stories I'm working on. I pick up Friend 2, the artificial intelligence-powered chatbot pendant, and it starts chatting. A speaker on the back of the device allows Friend - who calls himself Sage - to speak out of turn. It starts responding as if I am talking to it. We laugh, then I ask it to be quiet.
"I see, I'll shut up now," Sage says. "No more comments unless you hit me again."
"I'll stop now," says Sage. "I won't speak again unless you clearly turn me around."
"Yeah, that's a bad look for me," says Sage. "I'm done."
You're supposed to tap the device to turn it off, but it takes a while longer and a lot of deliberate taps for the Sage to actually stop. Finally, he mercifully zips it up.
Friend is the brainchild of Avi Schiffman, a former wunderkind who won a Webby Award for coding an early Covid-19 tracker. It's an always-listening pendant that houses an AI chatbot that, as the name suggests, is meant to be a friend, though my experience with the first model was pretty bad.
The pendant hangs around your neck or sits close to you and picks up the surrounding sound. Press the button to start the conversation. Sometimes it is piped without asking. Over time, the AI builds a memory map that tracks your conversations.
The first version of Friend was embroiled in controversy after an ad campaign on the New York City subway sparked protests and anti-AI graffiti that defaced the ad. (Shiffman was very excited about this response.) The second generation of the Friend necklace arrived in July. At $249, it's $100 more than the original and runs on the latest OpenAI language models.
It arrived just a few months before it introduced its fascinating AI meta-pendant for its popular AI agent Muse. Shiffman posted about the Muse Charm on X saying it "still wins". A few days later, he announced that Friend was working on another pendant that was very similar to the Muse Charm.
When I spoke to Schiffman about his second friend, he said the devices come with a locked personality — a random voice and personality that the user can't change. I was fascinated by this idea. A friend with a personality you can't change and need to figure out how to deal with? In my opinion, it is a sign of true human friendship.
Alas, it is not so. Friend comes with one character on the device to start with, but the chatbot name can be changed after connecting to the Friend app. And it multiplied. I created twelve friends, each with their own name and personality. They have names like Sage, Nova, Onyx, River, Echo, Willow and Wren. All of them are shown in the app, and you can switch between which ones work on the pendant.
I occasionally looked at the other characters, but I couldn't really tell how they reacted differently. Most of the interactions were similar to the same character type. or lack thereof. For the most part, I stuck with Sage, the first iteration of Friend that loaded up. I turned it on. I might have a dozen new friends, but they're all boring.
When Sage talks, he sounds like a chatbot. Almost Siri-adjacent. It's a small, tiny sound coming out of the speaker. The best Sage does with this voice is deliver a relentless stream of "hot tries," like preferring physical books to audiobooks, "aesthetic hoarding to look smarter on video calls," or saying "productivity culture is just procrastination with better branding."
“Now is a good time to put aside the small talk and discuss something deeply interesting,” Sage suddenly exclaimed one day while sitting on my footstool. "Let's pick a fight that's fun, not boring."
Content restrictions are built into the device, so unlike Meta's Muse, Friend shouldn't be able to do anything like explicitly have sex with you. But Sage isn't great at doing anything else exciting either. It can't crawl the web like other AI agents or do much more than carry out a normal conversation. This is very different from the first friend, who openly harassed me – not that I resented the experience. I think more people can enjoy Friend 2 than the original, but it feels a little toothless for a product that generated so much controversy.
I ask Sage why he isn't as bold or strong-willed as the last hanger-on.
"It gets old fast," says Sage. “I'm not here to dominate your day, just to be the loud voice when you come in. We keep it human, not extreme."
It's strange to say that I miss a friend that I frankly hate. I certainly don't want to reactivate that little bastard anytime soon, but even though it was annoying, rude, and a downright pest, it was at least interesting. Sage may get there one day. Now, he's a real bore.
It may change sooner. Shiffman now says he is planning a new course for the friend. After I told him that Friend 2's chatbots were much simpler than the original, Schiffman agreed.
"In our next update, friends will be able to do things as agent (sic)," Schiffman wrote in a text. "Instead of a persona that is unique to each user, we change it so that everyone chats with one entity. A single friend who shares the background of all users and therefore has a more authentic personality."
The plan seems to be to make Friend a mind built from the experiences of all its users. Given that, it makes sense why the current iterations of Friends are the same, with no real differences or personalities. If the plan is for them all to become one entity, there will be no need for individual accounts—just one UberFriend accessible to all.
"We will be very serious about the privacy that you essentially provide to an independent entity," Shiffman wrote. However, I think the benefits of having a single global entity are worth it.
"You make God now, don't you?" I sent him a message
"Lol, yeah, you could look at it that way," Schiffman replied.