ELIZA—a computer program for the study of natural language communication between man and machine
A starting point for understanding how a conversation can feel meaningful even when the underlying system follows relatively simple rules.
Research / Evidence & foundations
Follow the evidence.
Keep the questions open.
New findings to follow. Foundational work to think with. Research across the technical and human sides of conversational AI.
Latest key findings8What’s changing in the fieldFoundational reading9The ideas to know and return to
A lasting reading shelf / 1966–2021
Foundational reading
Influential journal and peer-reviewed conference papers for anyone going deeper into chatbots and companions. Read across disciplines, or start with the question closest to your work.
01Where it began
02People, social cues & relationships
Computers are social actors
People can apply social rules to computers without consciously believing that the computer is human.
Establishing and maintaining long-term human-computer relationships
Moves the unit of design from a single helpful exchange to a relationship that develops through repeated interactions.
On seeing human: A three-factor theory of anthropomorphism
A framework for asking when and why people attribute human characteristics, intentions or emotions to nonhuman agents.
My Chatbot Companion – a Study of Human-Chatbot Relationships
Explores how people who already consider a chatbot a friend describe the development of that relationship.
03Designing the interaction
“Like Having a Really Bad PA”: The Gulf between User Expectation and Experience of Conversational Agents
A study of the gap between what conversational interfaces lead people to expect and what those systems can actually do.
Guidelines for Human-AI Interaction
Eighteen guidelines for helping people understand, use, correct and remain in control of AI-supported interactions.
04Language models & their limits
Attention Is All You Need
Introduces the Transformer architecture: essential technical background for understanding many modern language models.
On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?
Connects language-model development with environmental costs, training-data bias and the risks of interpreting fluent text as meaningful communication.
How this shelf is selectedChosen for historical influence, relevance to conversational systems and breadth across disciplines. This is a starting collection, not a definitive canon or a ranking of the most-cited papers. Conference papers matter here alongside journal articles.
Citation counts: OpenAlex snapshot, 1 October 2026. Counts vary by database and record matching, favour older work and are not a measure of reliability. An unverified count is omitted. Every entry links to its source and states the depth of its reading note.
Citation counts: OpenAlex snapshot, 1 October 2026. Counts vary by database and record matching, favour older work and are not a measure of reliability. An unverified count is omitted. Every entry links to its source and states the depth of its reading note.