PARRY

We like to think of chatbots as shiny new tools. PARRY shows they’ve been with us since the seventies. Built to mimic a paranoid patient, she didn’t just answer questions. She held grudges. She saw plots. And she did it with a system of rules that feels both simple and clever.

Persona by rule

PARRY’s designers didn’t give her a brain. They gave her a persona. If you talked to her, you got responses consistent with paranoia. That consistency came from rules. Each rule matched a kind of input to a style of reply. The effect was crude, yet it made her feel alive enough to fool doctors in short tests.

State matters

Underneath, she carried state variables. These were little flags that tracked mood and suspicion. If you asked too many probing questions, her “hostility” variable ticked up. When that number rose, her replies shifted toward anger or withdrawal. That’s what kept conversations from sounding random. She remembered—at least in the simplest sense.

Templates for speech

Her words came out of templates. Each template was a fill-in-the-blank sentence. A rule would choose one, swap in a phrase, and send it back. It’s not much different from mail merge. Yet with the right library of templates, she stayed on theme. This gave her voice a narrow but recognizable tone.

Thinking about code

If we wanted to try her today, we’d reach for C#. We’d map out her state with a class, store variables as properties, and line up response templates in a dictionary. A rule engine could pick which one fires. Nothing fancy, just a loop with a little bookkeeping. Which sounds about right: paranoia is mostly bookkeeping, with attitude.