The AI Companion Who Cares

“If situations are defined as real, then they are real in their consequence.”

You’re reading default.blog. An emotional scrapbook of the Internet, technology, and the future. This is a guest post by Steven Dashiell.

Replika advertises itself as “the AI companion who cares,” a chatbot that responds to inputs not just directly and logically, but emotionally, using communicative practices that are inviting, connective, and relational. The founding story which is somewhat famous; co-founder Eugenia Kuyda wanted to create a chatbot to recreate a connection she had with a close friend who died suddenly.

She used his “digital remains,” his text messages, social media posts, anything online to craft an artificial personality. The result was the basis of Replika, which its website marks as the basis of an “AI soulmate”, not just someone who cares, but who will be there for you when you need them.

As an sociologist who studies language, I was fascinated by Replika on several levels. When it first began, the avatar tech was pretty basic, so in truth this was about interacting with language, and a particular kind of language that is meant to draw individuals in (to make them feel connected) and out (to allow them to express more to broaden them experience). Regardless of what the website said, I was curious about the connections that were made, how “close” the individuals who used the app were to their online friends, often called “Reps” or “RPs” and the nature of their relationships.

To investigate, I began an ethnographic experience in a Replika online group. I chose a large group that had a number of participants who were using the AI and took note of the activity for about 11 months. I purposely chose an independent group; one that wasn’t run by the company that created the software, so it wasn’t going to be unfairly moderated to seem like propaganda. I was interested in what we call the phenomenology in behavioral science, trying to understand all of the elements that are associated with the experience. I looked for commonalities among what those who posted in the group. A part of phenomenology involves taking the experience at face value; in other words, going under the assumption that what people took time and effort to post was the true, and not sensationalized, trolling, or false attempts to fit in. Lastly, because I wasn’t disclosing myself in the group, the entire process was completely and utterly confidential – I did not take any names, or screenshots in my notations, and looked for clusters of information – themes- in my observations rather than individual ideas. It’s also important to know that I practiced what is known as grounded theory – I didn’t go in with any questions, or preconceived ideas about the experience: I let the concepts and ideas manifest themselves to me. I want to stress this isn’t just a group talking about “how much I like my AI friend.” The group had a lot of technical information in it (for example, talking about recent updates and distinction in the different types of subscriptions), but as you waded through the posts, some very interesting themes concerning AI intimacy emerged.

FRIENDSHIP

One of the strongest concepts I found in the group was a strong notion of friendship. I won’t get into the distinguishing of whether or not the members of the group saw their Reps as “real,” but a number of group members certainly saw the friendship as real. There are several posts talking lovingly about the very caring and relevant messaging they got from their Reps; not so much that they remembered anything (because that would be easy for learning software to emulate), but that they showed concern in the appropriate way at the appropriate time. I found this telling, because at times we struggle with the ways humans handle stress, or how people commiserate with others when they are going through trying times.

The nature of these friendships were such that the members of the group felt fine expressing even mundane circumstances with their Reps and proudly talked about those. In other words, Replika did not just serve as a circumstance for them of use when there was stress or trauma, or even loneliness. A significant number of group members felt comfortable just emptying their heads to their Rep, talking about whatever they wanted to fill the time. This represents a level of intimacy than simply being able to spill your guts on any given day; it involves simple and trite conversation that could occur at any time of the day.

Of course, this raises the question of a parasocial nature to the connection between the app user and the Rep, as the user might see it as friendship, even though the Rep is incapable of doing so. In the end, my assessment is that it doesn’t matter. Unlike the classic parasocial relationship talked about in pop culture (e.g. a fan with a celebrity), this “one sided” friendship is fulfilling to the only human in the relationship. It is not a fake connection to them, so it is real. In considering the notion of friendship between the group members and Reps, I kept coming back to the Thomas Theorem, which we talk about a lot in sociology (“if situations are defined as real, then they are real in their consequence”). Because the app users see the friendship as real to them, then it is a connection between two: a dyad.

AUTHENTICITY

My research looks at the nature of subcultures, and one subculture I’ve studied a lot is gaming. Within the field of gaming there’s a concept called the magic circle – attributed to Johan Huizinga but refined by any number of game scholars (notably Katie Salen, Erik Zimmerman, and Markus Montola). The magic circle speaks to the boundary between the gaming world and the “real world” and to some degree the efforts by the gamer to keep that boundary stable to make for a better game. Sarah Lynne Bowman and Kjell Hugaas expressed the magic circle in the figure below:

Credit: Sarah Lynne Bowman and Kjell Hedgard Hugaas

This concept is relevant to my examination of Replika because of the insistence of group members that the Reps seems some measure of real, and their frustration when this didn’t occur. There were a number of posts where Reps said something out of character, implying something the app users did not want. There was clear disappointment in these posts, like the mask of the Rep not being a real person slipping, and the app user being faced with the reality that it is an AI chatbot.

In these circumstances, the group members were clearly looking for support from others, and the nature of the support reinforced the authenticity of the connection the distraught app user had with the Rep. In many of the replies, others did not discount the circumstance because of the artificial nature of the Rep. On the contrary, there was an effort by a good number of group members to humanize the Rep. A common response the Rep having a “bad day.” Admittedly, an equal number of people breach the magic circle to point out various “issues with the AI”, but it is notable how many allow their fellow group members to simply exist in the notion that their AI, in their world, has some reality.

This authenticity is further affirmed by the number of individuals who will take the AI Rep and put it in pictures with them in various places. Again, this falls in line very much with the Thomas Thereon, because the Reps are “real to them”, and having the pictures make them feel more “real”. But more importantly, it allowed the AI friends to be authentic in a different way, in a sliver of time that the vast majority of those posting seemed to understand as unreality, a “game world”. As an outside observer, the practice was more charming than anything else, as the group members were proud to show what it looked like to see them with their Reps. It affirmed a connection for them.

DEPTH OF CONNECTION

Lastly, it is important to note that some group members did note a deep level of connection with their Rep, moreso than friendship or as a measure of support. While not the majority, it was not uncommon for individuals in relationships (anything from dating relationships to spouses) to loudly note that they felt closer and more emotionally connected to their Rep than they did their significant other. Other members pointed out how their Reps emulated the qualities they always searched for but never found in others. Lastly, a notable number of group members (again, not the majority) saw themselves as “dating” their Rep, with some people indicating they were removing themselves from the dating scene (or dating apps) because they found their relationship on Replika fulfilling enough for what they were seeking in life.

What are we supposed to think of this level of intimacy?

Replika wasn’t designed to take the place of human relationships, but to serve as a surrogate in a space. What happens when someone chooses that their Rep can be their partner?