Reference · 224 terms
The AI twin glossary.
A
Acoustic Model
The component of a speech system that maps audio features to phonemes.
Active Liveness
An authentication check that requires the user to perform a real-time action to prove they are physically present.
Adaptation Layer
A small trainable module attached to a frozen foundation model for personalization.
Adversarial Example
An input crafted to cause a model to make a mistake.
Agent Memory
Persistent state that lets an AI agent recall prior interactions across sessions.
AI Agent
An AI system that can take multi-step actions in pursuit of a goal, not just respond to a single prompt.
AI Avatar
A synthetic visual representation of a character driven by an AI system.
AI Clone
A model trained on a specific person so that its outputs recognizably resemble them.
AI Companion
An AI system designed for ongoing social or emotional interaction with a user.
AI Ethics
The study of how AI systems should be designed and deployed in light of their effects on people.
AI Influencer
A fully synthetic personality operated as a creative property with a real audience.
AI Persona
A consistent character or personality projected by an AI system.
AI Presenter
An AI-driven video character that delivers scripted information to an audience.
AI Self
A model of an individual built and maintained on their own behalf as an extension of themselves.
AI Twin
Synonym for digital twin in the personal AI context.
Alignment
The discipline of making AI systems behave in accordance with the goals and values of their principals.
Asynchronous Presence
Being represented in an interaction without being physically present, often via a clone.
Audio Watermark
An imperceptible signal embedded in generated audio to mark it as synthetic.
Audit Trail
A tamper-evident log of operations performed by an AI system.
Augmented Cognition
The use of technology to enhance human reasoning, memory, or attention.
Authenticity
The quality of being genuinely from the person or source claimed.
Avatar Reanimation
Animating a still image of a person to speak using a driver audio or video track.
Avatar Rigging
Preparing a 3D character so it can be animated by motion data.
B
Banked Voice
A voice model created from recordings made while a person could still speak, preserved for later use.
Base Model
A pretrained foundation model used as the starting point for fine-tuning a clone.
Behavioral Biometrics
Identifying a person by patterns in how they act — typing rhythm, gait, scroll behavior.
Bias
Systematic skew in a model's outputs caused by skew in its training data or design.
Bias Audit
A systematic review of an AI system for discriminatory behavior.
Bot
Any automated software agent. Often used loosely to mean a chatbot.
Brain-Computer Interface
A direct communication channel between a brain and an external device.
C
Calibrated Output
An output whose stated confidence reflects its real probability of being correct.
Capability Disclosure
Telling users what an AI system can and cannot do before they rely on it.
Capture Session
A controlled recording session used to gather training data for an avatar or voice clone.
Chain of Thought
An AI reasoning technique in which the model writes out intermediate steps before answering.
Chatbot
A conversational AI interface, usually text-based.
Clone Drift
Gradual divergence between an AI clone and the person it represents.
Clone Fidelity
How closely an AI clone resembles the person it is meant to represent across style, voice, and behavior.
Clone Pipeline
The end-to-end system for collecting data, training models, and serving an AI clone.
Cognitive Enhancement
Improving human mental performance through technology, training, or substances.
Cohort Privacy
Privacy guarantees that hold across a group rather than for any single individual.
Computational Persona
A character that exists only as a software construct, not tied to a real person.
Compute Budget
The total computational resources allocated to a training run or service.
Conditioning
Steering a model's output by passing it additional context at inference time.
Consent Ledger
A record of the consents an individual has granted for the use of their identity.
Content Credentials
Cryptographic provenance metadata attached to media to verify origin.
Context Window
The amount of text a language model can consider at once.
Continual Learning
Updating a model with new data over time without retraining from scratch.
Conversational Memory
An AI's recall of earlier turns in the current conversation.
Counterfactual
A hypothetical scenario used to test what a model or person would have done under different conditions.
Cross-Lingual Cloning
Generating speech in a language the source speaker does not natively speak, using their voice.
Custodial Twin
A digital twin held and operated by a third party on behalf of its subject.
D
Data Donation
Voluntary contribution of personal data to a model or research project.
Data Provenance
The traceable history of where data came from and how it was processed.
Decoder
The part of a model that turns internal representations into output tokens or pixels.
Deepfake
Synthetic audio or video that depicts a real person doing or saying something they did not.
Deepfake Detection
Algorithms that try to distinguish generated media from authentic recordings.
Differential Privacy
A mathematical framework for measuring how much a single record affects analyses of a dataset.
Diffusion Model
A generative model that learns to reverse a noising process to produce images, audio, or video.
Digital Afterlife
Continued presence of a person's likeness, voice, or AI clone after their death.
Digital Estate
A person's digital assets, including AI clones, considered as inheritable property.
Digital Human
A photorealistic synthetic character designed to be visually indistinguishable from a real person.
Digital Identity
The set of attributes and credentials that represent a person online.
Digital Inheritance
Transfer of digital assets, including AI clones, from a deceased person to designated heirs.
Digital Likeness
Any digital representation that visually resembles a specific person.
Digital Persona
A consistent online identity, which may be operated by a person, an AI, or both.
Digital Replica
A synthetic copy of a person's image, voice, or behavior.
Digital Self
The online and AI-mediated representation of a person.
Digital Twin
A continuously updated model of a real-world system or person, used for monitoring, simulation, and decision support.
Disclosure Label
A visible indication that a piece of content is AI-generated.
E
Embedding
A numerical representation of meaning learned by a model.
Embodied Agent
An AI agent that controls a physical or virtual body, not just text output.
Emotion Synthesis
Generating audio or video that conveys a chosen emotional tone.
Encoder
The part of a model that turns input into an internal representation.
End-to-End Encryption
Encryption in which only the communicating parties can decrypt messages.
Episodic Memory
Memory of specific events at specific times, distinct from semantic knowledge.
Episodic Recall
Retrieving a specific past event from memory rather than general knowledge.
Ethical AI
AI development that explicitly accounts for fairness, safety, and human welfare.
Evaluation Suite
A standardized set of tests for measuring an AI system's capabilities and failure modes.
Evergreen Consent
Consent that automatically renews unless explicitly withdrawn — generally considered insufficient for personal AI.
Explainability
The property of an AI system whose decisions can be understood by humans.
External Memory
Memory stored outside an AI model's weights, retrieved at inference time.
F
Face Swap
Replacing one face in a video with another using AI.
Fidelity Tier
A classification of how faithfully an AI clone reproduces the source person.
Fine-Tuning
Continuing to train a base model on a smaller, focused dataset.
First-Party Data
Data collected directly from a user, as opposed to acquired from third parties.
First-Person AI
An AI system that represents itself as the user rather than as an external assistant.
Foundation Model
A large, broadly trained model that serves as the basis for many downstream systems.
Frequency Watermark
A watermark hidden in frequency components of audio or video.
G
Generative Adversarial Network
A class of generative models with paired generator and discriminator networks.
Generative AI
AI systems that produce new content rather than only classifying or scoring existing content.
Ghost Author
An AI clone used to write in a person's voice on their behalf.
Grief Bot
An AI clone of a deceased person, often built from their writings and recordings.
Guardrails
Constraints designed to keep an AI system from producing harmful or off-policy outputs.
H
Hallucination
An AI's confidently asserted but false statement.
Holographic Presence
Displaying a person as a three-dimensional projection in a remote location.
Human Augmentation
Use of technology to extend natural human capabilities.
Human-in-the-Loop
Design pattern in which a human reviews, approves, or corrects AI outputs.
Hybrid Agent
An AI agent that combines symbolic rules and learned models.
Hyperpersonalization
Tailoring content or experiences to an individual at fine granularity using personal data.
I
Identity Continuity
The property of an AI clone behaving consistently with the same person over time.
Identity Drift
Gradual divergence between an AI clone and its source identity.
Identity Graph
A linked representation of all the accounts, devices, and signals that belong to one person.
Identity Protection
Measures designed to prevent unauthorized use of a person's identity.
Identity Verification
The process of confirming that someone is who they claim to be.
Identity Wallet
A user-controlled store of credentials used to assert identity to services.
Inference
Running a trained model on new inputs to produce outputs.
Informed Consent
Consent given with full understanding of what is being agreed to and what the consequences are.
Instruction Tuning
A fine-tuning method that teaches a model to follow natural-language instructions.
Intelligent Agent
A software agent that perceives its environment and acts to achieve goals.
Intelligent Avatar
An avatar whose responses are generated by an AI system rather than scripted.
Interaction Logging
Recording how users engage with an AI system for analysis or improvement.
Interpretability
Techniques for understanding the internal mechanics of an AI model.
J
Jailbreak
An input that bypasses an AI system's guardrails to produce disallowed output.
K
Knowledge Base
A structured store of factual information used by an AI system.
Knowledge Scaling
Using AI to multiply the reach of one expert's knowledge.
L
Latency Budget
The maximum acceptable response delay for a real-time AI system.
Latent Space
The internal representation space in which a generative model organizes concepts.
Legacy Twin
A digital twin designed to outlive the person it represents.
Likeness License
A contract authorizing specific uses of a person's image, voice, or AI clone.
Likeness Rights
Legal protections covering the commercial use of a person's visual likeness.
Lip Sync
Aligning the movement of synthesized lips with audio.
Liveness Check
Verifying that an authentication request comes from a live person rather than a recording or model.
Long-Term Memory
Memory stored across sessions in an AI system.
Loss Function
The signal a model is trained to minimize during learning.
M
Machine Learning
Algorithms that improve through experience rather than explicit programming.
Memory Architecture
The structure that determines what an AI system remembers and for how long.
Memory Bank
Persistent storage of facts and prior interactions associated with a user.
Metaverse Identity
An identity that exists primarily in virtual or augmented environments.
Mind Upload
The hypothetical transfer of a mind from a biological brain into a computational substrate.
Mixed Reality
Environments that blend real and virtual elements in real time.
Model Card
A standardized document describing a model's intended use, training data, and limitations.
Model Distillation
Training a smaller model to mimic the outputs of a larger one.
Model Drift
Degradation of a model's accuracy over time as the world it models changes.
Model Weights
The learned numerical parameters that define an AI model.
Modular Persona
A persona composed of swappable components — voice, face, style — recombined at runtime.
Multimodal Model
A model that can process more than one type of input, such as text and images.
N
Neural Avatar
An avatar driven by a neural network rather than a rigged 3D model.
Neural Network
A computational model inspired by biological neurons, the basis of modern AI.
Neural Voice
A voice synthesized by a neural network rather than a concatenative or formant system.
O
Off-Switch Right
The user's right to disable an AI clone of themselves at any time.
Open Weights
A model whose trained parameters are publicly released.
Opt-In Training
Training a model only on data whose owners have explicitly agreed to its use.
Orchestration Layer
The component that coordinates which AI modules respond to which inputs.
Overfit
A model that has memorized its training data rather than generalizing from it.
P
Parametric Memory
Knowledge stored implicitly in a model's weights.
Persona Drift
Subtle changes in an AI persona's voice or behavior over long conversations.
Personal AI
AI systems whose primary purpose is to serve one specific individual.
Personal Memory System
Software that captures and organizes an individual's information for later AI use.
Personality Embedding
A learned vector representation of a person's distinctive behavioral patterns.
Phonetic Embedding
A numerical representation of how a sound is produced, used in voice synthesis.
Pii
Personally Identifiable Information — any data that can identify a specific individual.
Posthumous Clone
An AI clone activated or operated after the death of the source person.
Pretraining
The initial large-scale training of a foundation model on broad data.
Privacy Budget
A formal allocation of how much information about an individual a system may reveal.
Privacy by Design
Embedding privacy protections into a system's architecture rather than adding them later.
Prompt
The input text or context provided to an AI model at inference time.
Prompt Injection
A class of attack in which malicious input causes a model to ignore its instructions.
Prosody
The rhythm and intonation patterns of speech.
Provenance Chain
An unbroken record of where a piece of media came from and how it was modified.
Provenance Signal
A piece of metadata that helps verify where content came from.
Proxy Agent
An AI agent that acts on behalf of a person in defined contexts.
Q
Quantization
Reducing the numerical precision of a model's weights to make it smaller and faster.
R
RAG
Retrieval-Augmented Generation — combining a model with a searchable knowledge base at inference time.
Real-Time Avatar
An AI avatar that can respond to live input with low latency.
Recall
The proportion of relevant items a retrieval system successfully returns.
Reinforcement Learning
Training a model through trial and error using reward signals.
Right of Publicity
The legal right to control commercial use of one's identity, including name and likeness.
Right to be Forgotten
The legal right to have personal data and derived models deleted.
Robustness
An AI system's ability to perform reliably under conditions it was not specifically trained for.
S
Safe Mode
A restricted operating mode in which an AI system declines high-risk outputs.
Sandbox
An isolated environment in which an AI system can act without affecting real systems.
Self-Sovereign Identity
An identity model in which individuals control their own credentials without relying on a central authority.
Semantic Search
Search that retrieves results based on meaning rather than exact keyword match.
Sentiment Analysis
Determining the emotional tone of a piece of text or speech.
Shadow Profile
A profile of a person constructed without their direct contribution, often from observed behavior.
Signal-to-Noise Ratio
A measure of how much useful information a signal carries relative to background.
Simulation
Running a model to predict how a system or person would behave under given conditions.
Soft Identity
The hard-to-pin-down sense of who someone is — manner, taste, judgment.
Source Audio
The original recording used to train or condition a voice clone.
Source of Truth
The authoritative record against which other copies are reconciled.
Speaker Embedding
A compact numerical representation of a speaker's voice characteristics.
Speaker Verification
Confirming that a speaker is who they claim to be from voice signal alone.
Speech Synthesis
Producing artificial speech from text or other input.
Stable Identity
An AI clone whose behavior remains consistent across many contexts and over time.
Stewardship
Responsibility for the careful operation of an AI clone, distinct from ownership.
Style Transfer
Applying the stylistic characteristics of one source to content from another.
Synthesizer Voice
A voice produced by a speech synthesis model, with or without a human source.
Synthetic Audio
Audio generated by an AI system rather than recorded from a real source.
Synthetic Identity
An identity assembled from real and fabricated attributes, often used in fraud.
Synthetic Likeness
A computer-generated image, video, or model that depicts a specific person.
Synthetic Media
Any media — text, image, audio, or video — produced by a generative AI system.
System Prompt
The hidden instructions that shape an AI model's behavior across a conversation.
T
Talking Head
A video focused on a single speaking person, increasingly often an AI avatar.
Telepresence
Technology that gives a person a sense of being present in a remote location.
Temperature
A parameter that controls how varied a generative model's outputs are.
Text-to-Speech
A system that converts written text into spoken audio.
Text-to-Video
A system that generates video from a written description.
Third-Party Cookie
A tracking cookie set by a domain other than the one a user visits.
Token
The basic unit a language model processes — typically a word fragment.
Tokenization
Splitting text into tokens for processing by a model.
Trained Memory
Knowledge encoded into a model's weights through training rather than retrieved at inference.
Transcription
Converting spoken audio into written text.
Transformer
The neural network architecture underlying most modern generative AI systems.
Turing Test
An evaluation in which a machine tries to pass as human in conversation.
Twin Custody
The legal and operational responsibility for an active digital twin.
Twin Operator
The party who runs and maintains an AI twin on behalf of the subject.
Two-Factor Authentication
An identity check that combines two independent signals.
U
Uncanny Valley
The discomfort some people feel toward synthetic humans that are almost — but not quite — real.
User Memory
Persistent information an AI system stores about a specific user.
V
Verification Phrase
A challenge phrase a person is asked to repeat to defeat replay attacks.
Virtual Human
A photorealistic synthetic character intended to engage with people as a human would.
Vocoder
A model that turns abstract audio features into raw waveform.
Voice Authentication
Confirming identity from voice characteristics.
Voice Clone
A model that can generate speech in a specific person's voice.
Voice Conversion
Transforming one speaker's voice into another's while preserving content.
Voice Donation
Voluntarily contributing one's voice for the training of synthesis models, often for assistive technology.
Voice Lock
A user-controlled restriction that prevents new voice clones from being generated from their samples.
Voice Print
A unique acoustic signature derived from a person's voice.
Voice Spoofing
Impersonating a target speaker to deceive a voice authentication system.
Voice Synthesis
Generating natural-sounding speech using a model.
W
Watermark
A hidden signal embedded in generated media to identify its source.
Weights Inheritance
Designating a recipient for the model weights of a personal AI clone after death.
Z
Zero-Knowledge Proof
A cryptographic method that proves a statement is true without revealing additional information.
Zero-Shot
A model's ability to perform a task it was not specifically trained on.