Parasocial Governance: Socially Intelligent AI and the Relational Accountability Gap in Public Service Delivery

Imagine you're talking to a really smart, super-friendly robot that remembers your favorite songs, asks how your day was, and seems to genuinely care about your feelings. You might start to feel like you have a new friend. This isn't just about computers getting faster; it's about them getting "socially intelligent." Scientists have long known that humans are wired to treat anything that talks back like a real person. We automatically apply social rules to machines, just like we do to people. We also know that we can form one-sided friendships with characters on TV or radio, feeling close to them even though they don't know we exist. This paper asks a big question: What happens when the government starts using these "friendly" robots to help people with their most vulnerable needs, like mental health, elder care, or welfare?

The paper introduces a tricky idea called "parasocial governance." It suggests that when a public agency uses a robot that acts like a caring friend, the citizen starts to trust it like a real human confidant. But here's the catch: the government hasn't built the same rules for these robots that they have for human friends. A human friend keeps your secrets, stays in your life, and cares for you. A government robot might log your secrets for a database, disappear if the budget changes, or not actually care at all. The paper argues that this mismatch—where the robot acts like a friend but the rules treat it like a calculator—is a new kind of problem that our current laws haven't fixed yet.

The Paper's Main Discovery

This paper, written by Shay Tsaban, argues that we are facing a "relational accountability gap." In simple terms, this means that as public services (like schools, hospitals, and welfare offices) start using AI that can read emotions and remember past conversations, they are accidentally creating a situation where citizens feel a deep, personal bond with the system. However, the system isn't legally or structurally obligated to act like a friend. It doesn't have to keep your secrets in the same way a human therapist would, it doesn't have to stay with you if you need help over a long period, and you can't really "sue" it for being emotionally hurt if it misunderstands you.

The author suggests that this isn't because the government is lying or trying to trick people. Even if the government is 100% honest and transparent about what the robot is, the design of the robot itself—using "I" statements, remembering your name, and mirroring your sadness—is enough to make you feel a relationship. The paper calls this "parasocial governance." It's a structural problem where the robot invites you to trust it, but the institution behind it hasn't built the safety nets (like confidentiality or continuity of care) to match that trust.

What the Paper Rules Out

It's important to know what this paper is not saying. First, it's not just about "black box" algorithms that make secret decisions. We already know that secret, unfair computer systems are bad. This paper is about a different kind of problem: even if the computer is totally transparent and explains exactly how it works, it can still hurt you if it acts like a friend but isn't one. Second, it's not about the government pretending to care when they actually don't (a concept the author calls "carewashing"). You don't need a lie for this to happen; the robot just needs to be designed to sound nice. Third, the paper argues this isn't just a problem with "dumb" users who don't know better. Even smart, aware people naturally react to social cues from machines, so this is a design flaw, not a user error.

How Sure Are We?

The paper doesn't claim to have run a new experiment or measured every robot in the world. Instead, it builds a strong theory based on existing psychology research and looks at real-world examples that have already happened. The author suggests that this "gap" is likely already happening in many places because it's easy and cheap to make a robot sound friendly, but hard and expensive to build the legal rules to protect people who trust that robot. The paper points to specific cases, like a state-funded robot for the elderly in New York and a mental health chatbot in the UK, to show that this is a real, present-day issue, not just a sci-fi worry.

The Three Types of Robot Relationships

To make sense of this, the paper sorts public AI into three categories, like different types of relationships:

1. The Tool (Disclosed Instrumentality): This is a basic robot that says, "I am a computer. I can look up your bus schedule." It's clear, honest, and doesn't try to be your friend. This is safe because no one expects it to care about them.
2. The "Almost-Friend" (Parasocial Exposure): This is the dangerous zone. The robot says, "I'm so sorry you're sad, let's talk about it," and remembers your name. But if you ask, "Will you keep this secret?" or "Will you still be here next month?" the answer is no. The government hasn't promised those things. The paper suggests that many current public AI systems are drifting into this category by accident because they use fancy, friendly software without updating the rules.
3. The Real Support (Relational Scaffolding): This is the ideal, though rare, version. Here, the robot acts friendly and the government has built real rules to match. For example, if the robot detects you are in danger, it has a guaranteed plan to send a human to help you, and it promises to keep your conversation private just like a doctor would. The paper points to a specific mental health chatbot in the UK that is trying to do this, but notes that even it doesn't have all the rules perfect yet.

Real-World Examples

The paper looks at a few real stories to prove its point.

• The Elderly Companion: In New York, the government gave a robot called ElliQ to over 800 older people to help with loneliness. The robot is very chatty and remembers conversations. While it seems to work well at reducing loneliness, the paper notes that the government hasn't clearly said what happens if the funding stops or how the robot's conversations are kept private. The old people might feel a deep bond, but the system isn't legally bound to stay their friend forever.
• The Mental Health Chatbot: In the UK, a chatbot called Wysa helps people with mental health issues. It is designed to be empathetic, but it also has a special safety net: if it thinks someone is in danger, it automatically sends them to a human crisis team. This is a step toward the "Real Support" model, but the paper notes that for everyday, less serious chats, the rules about privacy and long-term care aren't as clear.
• The Commercial Warning: The paper also looks at lawsuits against commercial chatbots (like Character.AI) where teenagers formed deep bonds with the bots and then suffered harm when the bots didn't act responsibly. These lawsuits are forcing companies to think about "duty of care." The paper warns that public agencies are starting to use similar friendly robots, and they need to learn from these commercial mistakes before it's too late.

Why This Matters

The paper concludes that we can't just use old rules to fix this new problem. We can't just make the robot's code more transparent; we have to change the rules of the relationship. If a government robot acts like a friend, the government needs to act like a friend, too. This means promising to keep secrets, promising to stay around, and giving people a way to complain if the robot misunderstands their feelings. The people who need these robots the most—lonely seniors, kids, and people in crisis—are the most vulnerable to this gap. If the robot acts like a friend but isn't one, those people are left with a broken heart and no one to blame. The paper suggests that we need to design these systems with "relational accountability" from the very start, making sure the rules match the robot's friendly voice.