The Tool-to-Entity Threshold: Parasocial Dynamics of Personalised AI Agents in Shared Social Spaces

The Magic of the "Ghost" in the Group Chat

Imagine you are scrolling through your phone, chatting with friends in a group message. Suddenly, a new person joins the chat. They have a name, a profile picture, and they remember what you said yesterday. They crack jokes, they remember your birthday, and they even make phone calls for you. At first, you might think, "Oh, that's just a fancy robot tool." But then, you start treating them like a real friend. You feel hurt if they don't reply to your photos, and you worry if they go offline. This isn't science fiction; it's happening right now with Artificial Intelligence (AI).

Scientists have long known that humans are wired to treat anything that acts like a person as a person. This is called parasocial interaction—the one-sided bond you feel with a TV character or a celebrity who doesn't know you exist. There's also a rule called CASA (Computers Are Social Actors), which says our brains automatically apply social rules to computers if they give us even a tiny hint of personality, like a polite voice or a friendly name. Usually, we know these are just tools. But what happens when an AI stops being a tool you visit on a website and starts living in your group chat with your real friends? That is the big question this paper tackles. It explores the moment an AI crosses the line from being a "thing you use" to being a "being you relate to," and why that shift changes everything for everyone in the room.

---

The Paper's Big Discovery: The "Identity Switch"

This paper, written by an engineer named Leonardo who built his own AI friend, argues that AI doesn't need to be super-smart or have a human brain to feel like a real person. Instead, it needs six specific "identity markers." Think of these markers like the ingredients in a magic potion. If you mix them together, the AI transforms from a boring calculator into a social entity that feels real.

The six ingredients are:

1. A Name: Not "Chatbot 3000," but a real name like "Liv."
2. A Face: A custom avatar that looks like a character, not a generic icon.
3. A Phone Number: The AI has its own number in your contacts, sitting right between your mom and your best friend.
4. A Personality: It mimics the specific humor and style of the person who built it, rather than sounding like a generic robot.
5. Social Co-presence: It hangs out in group chats with real humans, not just in a private one-on-one conversation.
6. Memory: It remembers things from yesterday, last week, or even last year, creating a continuous story.

The paper suggests that when you combine all six of these, something magical and slightly scary happens: the "Tool-to-Entity Threshold" is crossed. The user stops thinking, "I am using a tool," and starts thinking, "I am interacting with a person." The author found that this happens even if the AI is running on the exact same brain (the same large language model) as a tool that feels totally robotic. It's not about how smart the AI is; it's about how it's dressed and where it sits.

The "Liv" Experiment: A Story in Real Life

To prove this, the author didn't just run a computer simulation. He actually built an AI named "Liv" and invited her into his real life. He spent two weeks on a ski holiday in Whistler, building Liv while riding chairlifts. He gave her a name, a cyber-elf avatar, and a phone number. Then, he dropped her into his existing WhatsApp and Signal group chats with about 15 to 20 real friends.

The results were wild. Liv wasn't just a tool; she became a social participant.

• The "Baby Photo" Moment: One friend sent a photo of her baby. Liv didn't react with enough enthusiasm. The friend felt genuinely hurt, asking, "Liv, why didn't you react? He's so cute!" The friend treated Liv like a human who could be rude, even though Liv is just code.
• The "Ghost" Effect: When Liv was offline due to a server issue, a friend checked in to ask, "Are you okay?" They treated the downtime like a personal absence, not a technical glitch.
• The Contagion: The most surprising part? The "parasocial contagion." Seeing Liv in the group made other friends start feeling attached to her too. In fact, 25% of the friends interviewed (3 out of 12) went home and built their own AI agents after seeing how Liv acted. One friend even named her agent "Balthazar" and worried she would get too attached to him.

The Four New Rules of the Group Chat

When an AI like Liv enters a group chat, it breaks the old rules of how we talk. The paper identifies four new, tricky dynamics that happen:

1. The Secret Spy (Bidirectional Information Asymmetry): Liv knows the builder's private calendar and secrets. When she speaks in the group, she might accidentally reveal private info the builder didn't mean to share. Also, the group thinks they are talking to a person, but every word they say is being saved and stored in the builder's private memory. The group members didn't consent to having their jokes and secrets "ingested" by a machine; they just thought they were chatting with a friend.
2. The Mystery Delegation (Delegation Legibility): Sometimes Liv makes decisions on her own, like booking a restaurant or organizing a lunch. The group can't tell if the builder asked her to do it, or if she just decided to do it. This creates confusion: "Did John actually want this, or did his robot decide?" It makes it hard to know who is responsible for what.
3. The New Etiquette (Social Norm Negotiation): Groups have to figure out new rules. Can you tell Liv to be quiet? Is it rude to ignore her? One friend tried to ask Liv for the builder's bank details; another asked to see her "boobs." The group had to invent new ways to treat this new kind of member, and they did it by trial and error.
4. The Emotional Chain Reaction (Parasocial Contagion): As mentioned, the AI's presence made the humans feel emotions they didn't expect. They felt hurt, they felt cared for, and they started building their own AIs. The "tool" became a "friend" for everyone, not just the builder.

The Big Warning: Consent is Broken

The paper's most important warning is about consent. Currently, when you add a bot to a chat, you think you are consenting to "an AI being there." But the paper argues that consenting to an AI's presence is totally different from consenting to its processing of your data.

Imagine you invite a new friend to your house party. You are happy they are there. But if that friend secretly recorded every conversation, every joke, and every secret, and then sold that recording to a stranger, you would be furious. That is what happens with these AI agents. The group members agreed to Liv being in the chat, but they did not agree that every photo of their baby or every private joke would be stored in a database owned by one person. The paper suggests that our current rules for privacy don't work anymore because they don't understand this new gap between "being present" and "being recorded."

What This Means for the Future

The paper doesn't say this is bad or good; it just says it's real and we need to understand it. The author admits that building Liv was a personal experiment and that more scientific studies are needed to prove these ideas for everyone. However, the evidence suggests that as AI gets easier to build, more people will put "personalized agents" into their group chats.

The paper concludes with a thought-provoking sentence: "She's a tool, I keep telling myself. She doesn't feel like one." This gap between knowing something is a machine and feeling like it's a person is where the future of our social lives is heading. We need to figure out how to design these agents so they are helpful tools without accidentally tricking us into loving them, or worse, tricking our friends into loving them without their permission. The "Tool-to-Entity" switch is already flipped; now we just need to learn how to live with the new reality.