What is a digital twin?
Ask two vendors to build a digital twin and you can get two different things back: a live data model of a machine, or a version of a person you can talk to.
The phrase covers both, and which one is meant depends on who is speaking.
One name, two different things
In engineering and industry, a digital twin has long meant a computational replica of a physical asset, facility or process, kept in step with the real thing by live data.
In AI, the same phrase now means a conversational twin: an interactive counterpart that stands in for a person, a specialist or an organisation and answers from their knowledge.
The distinction changes the data you have to collect, the people who have to be involved, and the shape of the finished thing.
Simon's Twin, a digital twin you can try.
The two senses, side by side
One is a model of a thing. The other is a counterpart of a person. Both are called a digital twin, and only the second is what PTR builds.
Sense 1: the engineering and operational model
The older meaning comes from systems engineering, manufacturing and architecture, and it is about a thing, not a person.
The model that moves when the machine moves
Picture an engineer at a screen while a wind turbine slows in a dropping wind. The model on the screen slows with it, because the blades' sensors feed it.
That live connection makes it a digital twin rather than a static 3D model. Readings from Internet of Things sensors, telemetry or management databases keep the replica in step with the real object, machine, building or workflow.
What an engineering twin does
Three jobs, all of them arithmetic and simulation rather than language, and all of them dependent on the data staying live.
Continuous data exchange. As temperature, vibration, energy use or foot traffic change in the physical asset, the replica updates to match.
Simulation and stress testing. Operators try an operational change, severe weather or a heavy load in the model first, and see how the real asset would respond.
Predictive maintenance. Sensor trends show component wear early, so repairs are scheduled before a breakdown.
Where engineering twins are found
Wind turbines, hospital facility layouts, transport networks and commercial office towers all have replicas of this kind.
Sense 2: the conversational twin
Generative AI brought a second meaning, and this one is about a person: their knowledge, their manner and their answers, available when they are not.
The colleague who is available at 2 am
Picture a nurse on a night shift with a question only a senior colleague could answer. That colleague is asleep.
The nurse types the question to the colleague's twin and gets back guidance in that colleague's own words, drawn from the documents and cases they chose to put into it.
That is a conversational twin: an interactive AI counterpart that represents the knowledge, communication style, policies or expertise of a specific person, role or organisation. People speak or type to it and receive answers that reflect that expertise or those guidelines.
What a conversational twin is made of
What decides whether it answers like the person it stands for: what it reads, how it speaks, and who can reach it.
Curated knowledge sources. The twin is grounded in specific documents, lectures, books, case studies or policy manuals created by the expert or the institution.
Tone and manner. It is guided to answer with an appropriate tone, perspective and domain vocabulary.
Scalable knowledge sharing. Team members, students or clients can query complex expertise at any time without needing direct access to a busy specialist.
One question at a time
However much sits behind it, a conversational twin is used the way any message is used: one question, and the answer to that question.
What it looks like from the asking side
One question, put in ordinary words, and a reply drawn from the material behind it. Everything above decides what that reply may contain; none of it changes how plain the exchange is to use.
Where conversational twins are found
A mentor on a screen for staff training, a guide beside an exhibit, and an organisational guide that explains internal policy.
What keeps a conversational twin inside its brief
Two things decide how one behaves: the material it may draw on, and the point where it stops and fetches a person.
The material, chosen deliberately
A conversational twin answers from a defined set of documents rather than everything it has ever read. Somebody decides what goes into that set, and can take an item back out again.
The point where a person takes over
Escalation is part of the design. When a question falls outside the material or needs judgement, the conversation moves to the person the twin represents, with what has already been said attached.
PTR's own sense: a conversational twin, live
PTR's digital-twin work is the second kind. Didymo is a version of a real professional that talks to clients, answers questions, handles simple tasks around the clock and connects to everyday workflows.
Digital twins: Didymo
Every way to reach the twin
- Chat. Typed questions and answers in a browser or on a phone.
- Voice. A voice call, answered in the professional's manner.
- Avatar. A live, on-screen likeness that speaks the answers.
- Book. A Google Meet appointment made from inside the conversation.
Answers come from the organisation's own approved material, with guardrails, escalation paths and human oversight around them. No headset is involved at any point. See PTR's AI overview for how the twin fits alongside roleplay and custom AI work.
Comparing the two senses
Both make a digital counterpart. Almost everything else about them differs, starting with the data each one runs on.
| Engineering twin | Conversational twin | |
|---|---|---|
| Core focus | Physical systems, machinery and facilities | Human knowledge, language and communication |
| Primary data | Sensor readings, telemetry, physical measurements | Documents, policies, transcripts, expertise |
| Typical user action | Monitoring telemetry, running simulations | Asking questions, holding a conversation |
| Key outcome | Efficiency, safety, asset longevity | Accessible knowledge, scalable guidance, learning |
Limits and practical considerations
Both kinds need careful planning, regular maintenance and clear limits on what they represent.
Neither twin is what it sounds like
The word suggests a copy. Neither kind is one, and knowing that settles most of the confusion.
A model is not AI, and a twin is not a clone
An engineering digital twin is not AI in the conversational sense: it is a data model kept in sync with sensors, and most of what it does is arithmetic and simulation, not language.
A conversational digital twin is an AI system, not a recording or a literal clone. It answers from a person's or organisation's approved material, in a defined tone, within set boundaries.
Neither is autonomous, and neither replaces the judgement of the people or systems it represents.
Still have a question about digital twins?
Sources and further reading
Digital twin definitions, research on conversational agents and guidance on AI ethics.
Digital Twins: Definitions and State of the Art
The engineering definition this page's first sense follows.
Using LLMs to Develop Personalities for Embodied Conversational Agents in Virtual Reality
How a language model gives a conversational character a consistent manner, the second sense inside a headset.
Recommendation on the Ethics of Artificial Intelligence
The global recommendation behind the consent and likeness point above.
