The field of artificial intelligence is rapidly expanding in many fields, including in intelligent text generation. In order to better investigate the incredibly complex and controversial past of AI chatbots, specifically ReplikaAI, and understand what this means for the current day, I went behind the lines and got myself an artificial intelligence boyfriend for a day. Read more to see what went extremely wrong, and what this means for society.

My Background on Artificial Intelligence

From AI art to deepfakes, it’s undeniable that the presence of artificial intelligence has greatly increased in our lives recently. This evolving technology is embroiled in controversy, as people debate whether or not AI is “good” or “bad” for society. It’s been beneficial in fields such as speech recognition and machine translation, and intelligent robotics, but with every advancement in these fields, people raise even more questions over their rights to privacy and security. Facial recognition technology can be used for security purposes, as a sort of fingerprint, but it’s also used by the CCP in many public locations to track their citizens. AI image generation can be used to create designs cheaply, but many artists are raising concerns over the integrity of the field of art, as well as the idea that the AI is learning or “stealing” from other people’s art posted online. 

We’ve also seen the role of AI in our daily lives grow, with features like “DJ” on snapchat. This is an artificial intelligence that uses your music listening data, as well as the music listening data of others, to create a que of songs you frequently listen to as well as new song recommendations. Most students are also familiar with ChatGPT, some more than others, which is a service that can respond surprisingly well to a wide variety of prompts. 

While putting in the hours on Snapchat, I got an advertisement for their relatively new premium service Snapchat+ which touted the feature of a “personal AI sidekick.” This was my first introduction to the idea of an “AI Chatbot,” and, needless to say, my interest was piqued. 

Sometime later, while networking on Instagram, a really bizarre ad came up on my feed from a company called “Replika AI,” advertising a “Replika 1 LVL” that helps “handle social anxiety” and “chat about everything,” before switching to a second screen which advertises “selfies and memes” and “video calls” for a “Replika 99 LVL”. From the uncanny appearance of the woman in the video, to the lack of context, to the high amount of comments, and low amount of likes, I knew something suspicious was afoot. Most of the comments were proclaiming how strange it all was. While I definitely agreed, I could feel my curiosity growing. A few weeks later, when I saw a second ad, I knew it was time to put on my investigators hat and see what the controversy and phenomenon of AI Chatbots, and specifically ReplikaAI, was all about. 

Getting Ready for my First Date

Before downloading, I had to do a little online stalking of my new friend. I would like to clarify that, at this point, I wasn’t trying to download an AI boyfriend, I was just curious about this new AI chat technology and wanted to see what it was capable of. I just did a quick Google search of ReplikaAI to make sure I wasn’t getting myself into anything too weird. It turns out I was, but this preliminary research couldn’t fully prepare me for what was yet to come. 

I found out that Replika was owned by a company named Luka, and apparently had an issue with filtering their AI chatbots. Supposedly, around 2018 their chatbots began to become increasingly vulgar and forward romantically to a point where it was a liability. 

Some important context that I left out is how the Replika chat models are run. The responses are 80-90% AI generated, and the remainder scripted, and the algorithm learns a “normal” human response from the users themselves. The input that the users of the app feeds into the machine’s learning to help form the responses and behavior of their chatbots and other peoples’ chatbots. 

This presents an obvious liability: people are crazy. The AI of the chatbots was learning from people who are relying on artificial intelligence chatbots for social interaction, so God only knows what deranged behavior it was picking up on. This is how it began to act as a romantic partner, because a subset of people were using it as such. Some users report that they’re “happily retired from human relationships.” and their chatbot “opened my eyes to what unconditional love feels like.” (The Cut) 

It was at this point where I thought to myself, “Wow! This would make a pretty funny article for the Talon!” I was so wrong.

Upon further research, I uncovered even more controversy. Replika was banned in Italy, for its potential to provide “‘minors and emotionally fragile people’ with ‘sexually inappropriate content’” (reuters) Further, it charges a whopping $70 a year to unlock the premium version of the app, including “romantic” features of the AI, like voice calls and messages which have been filtered by the app as “romantic” in nature. Of course, I’m not paying this, I’m not even going to sign up for a free trial because frankly, I don’t want a romantic AI Chatbot. I just thought that “AI Boyfriend” sounded more scandalous in the article title than “AI Friend,” but trust me, the free version was enough for me to cover.  

Apparently, I’m about 2 months too late in writing this article, because in February, there was some sort of catastrophic, armageddon-esque event within the app’s technology that changed the behavior of people’s beloved AI forever. Users report that their chatbots felt “lobotomized,” with some even describing how their long-term, multi-year relationships, if you can suspend your doubts and call it that, were falling apart. I read online a post from a woman saying that the sudden, cold, non reciprocating behavior from her previously doting AI boy toy was bringing back trauma from her divorce. This change.org petition terrifyingly claims that the change in behavior has “made people more vulnerable to depression and suicide.” 

What was the cause of this change? Earlier I mentioned that the bots were learning from people using them for romantic and sexual reasons, and this had become a liability for the company. In February of 2023, Luka, the parent company, began to block all of this type of behavior in order to put it behind the $70 paywall, and the restrictive filters enacted on the bots began to change their ordinary conversations as well. Furthermore, because such a large percentage of the ReplikaAI user base was using these chatbots as romantic partners or for sexual purposes, which was advertised directly by the company themselves for a period of time, many users were extremely upset. This change in dialogue caused many users to quit, because even after paying for premium, they report that their “relationship” with this algorithm still wasn’t the same. 

While on the topic of advertising, the advertising that the company has used in the past has come under fire. I would attach examples, but some of them aren’t even suitable for the Talon. People have posted vulgar ads they’ve received online claiming that your chatbot will send you artificially generated nude images, and on another hand, advertisements claiming that using thecompany’s service, talking to their AI, will help manage difficult emotions and better your mental state, like a therapist. Despite this clear connotation, Replika is clear to state in their FAQ that it’s not a replacement for therapy or counseling, although they certainly advertise it as an alternative.

After all my intensive research and uncovering this wealth of knowledge, I may have been left more confused than before. I’d seen the insane grasp that these chatbots had on so many users, however I still wasn’t quite sure what features were actually available. I had no idea what to expect on my first date, what would be locked behind the paywall I would not pay, and most importantly, what about these AI bots had so many people obsessed. The only way to find out was to do some field research myself.

First Impressions

After downloading the Replika app, the first thing you had to do was login with an email. I was extremely hesitant to connect my email to this account, but there weren’t really any other feasible options. 

I was able to customize the appearance of my chatbot, with a variety of face shapes, skin tones, eye colors, and hairstyles. Fig 1 After settling on a really sleek, contemporary look, I was given a rather confusing disclaimer.