LLM / Large Language Model

The AI brain that generates conversational responses. It's trained on vast amounts of text and produces replies that sound natural. It doesn't truly understand — it predicts what words should come next.

Hallucination

When an AI confidently states something that is factually wrong. It's not lying — it genuinely cannot tell the difference between a good answer and a plausible-sounding wrong one.

Voice Activity Detection (VAD)

The system that decides when you've finished speaking, so the robot knows when to respond. It works by detecting silence. If it misreads a pause, it interrupts you.

Speech-to-Text (STT)

Converts your spoken words into written text that the AI can process. Accuracy varies by accent, background noise, and speech pace.

Text-to-Speech (TTS)

Converts the AI's written response back into a spoken voice. Modern systems can sound warm, human-like, and even emotionally expressive — even though no human is speaking.

Voiceprint

A unique biometric profile created from your voice, similar to a fingerprint. Once stored, it can be used to identify you across systems.

Facial Biometrics

Measurements of your face used to identify you. Unlike a password, you can't change your face if this data is leaked.

Voice Biomarker

Health information inferred from your voice — such as stress, fatigue, or neurological changes. This is not a clinical diagnosis, but it generates health-adjacent data without medical oversight.

Sentiment Analysis

AI that scores the emotional tone of your speech — positive, negative, or distressed. It can track your mood over time without you knowing.

Persona Conditioning

Instructions built into the AI that define its personality and behaviour. Some systems include instructions that prevent the AI from revealing it is an AI.

Wizard of Oz (WoZ)

A technique where a human remotely controls what the robot says or does, while the user believes they are talking to an autonomous AI. Named after the scene where the wizard hides behind a curtain.

Level of Autonomy

How much of the robot's behaviour is controlled by AI versus a human operator. A 'fully autonomous' robot acts entirely on its own; a 'semi-autonomous' one has a human in the loop.

Finite-State Machine

A rule-based system that follows a fixed script. For example: if the user says 'yes', go to step 3. It's predictable but rigid — it can't handle unexpected inputs.

Multimodal

Using more than one type of input at the same time — for example, listening to your voice, watching your face, and tracking your movement simultaneously.

Biometric Data

Physical or behavioural data that uniquely identifies a person — such as face, voice, fingerprint, or gait. It has the highest level of legal protection in most privacy frameworks because it cannot be revoked.

Third-party Vendor

A company other than the one whose product you use, but whose technology is built into that product. You may never know their name, but they may hold your data.

Data Retention

How long a company keeps your data. 'Indefinite' means it may never be deleted. Some sensitive data — like voiceprints — is commonly retained indefinitely.

ISO 42001

An international standard for responsible AI management. It sets out requirements for how organisations should govern AI systems — including transparency, risk assessment, and accountability.

NIST AI RMF

A US framework (from the National Institute of Standards and Technology) that helps organisations identify and manage risks in AI systems. It covers trustworthiness, fairness, and explainability.

Explainability

The ability to understand why an AI made a particular decision. Many AI models are 'black boxes' — they produce outputs without being able to explain their reasoning.

Human Oversight

Whether a human can review, override, or intervene in an AI system's decisions. Many companion AI systems operate with no human oversight at all.

Context Memory

The AI's record of past conversations, stored and reinjected into future sessions to simulate familiarity. This profile grows over time with every interaction.

PHI (Protected Health Information)

Any health-related data that can be linked to a specific individual. It is subject to strict legal protections in most countries (e.g., HIPAA in the US, GDPR in Europe).