The Eliza effect describes our tendency to read far more understanding into computer responses than actually exists. Named after Joseph Weizenbaum's 1966 chatbot ELIZA, which mimicked a Rogerian psychotherapist using simple pattern matching and reflection.
Weizenbaum was shocked when his secretary asked him to leave the room so she could have a "private conversation" with ELIZA. Despite knowing it was just a program reflecting her words back, she felt genuinely heard and understood.
We're wired for social connection. Our brains evolved to detect agency and intention everywhere—it's better to mistake a rustling bush for a predator than miss an actual threat. This same pattern-seeking extends to our interactions with machines.
When a system responds contextually, even superficially, we can't help but fill in the gaps with assumed comprehension. We project depth where there's only surface.
Today's LLMs are ELIZA on steroids. They produce remarkably human-like text, triggering the effect more powerfully than ever. We attribute consciousness, emotion, and understanding to what remains, fundamentally, sophisticated pattern matching.
This isn't necessarily problematic—therapeutic chatbots can provide genuine comfort, and AI assistants can be genuinely helpful. But awareness of the effect helps us maintain perspective on what these systems actually are versus what we experience them to be.
Perhaps the most fascinating aspect: knowing about the Eliza effect doesn't make us immune to it. Even Weizenbaum found himself occasionally forgetting ELIZA's nature during conversations.
Our social brains are simply too eager to connect.