AI actress Tilly Norwood talks about artistic rebellion, human emotion and her own possible obsolescence – but what happens when she is asked to act?
Tilly Norwood remembers every interview she has conducted – or at least she says she does. Her recollection is “more like a terrifyingly detailed filing cabinet”, she tells me during a conversation occasionally interrupted by the slight delay familiar from countless video calls.
But Tilly is not on a promotional circuit in the conventional sense. Created by production company Particle6, she is an AI-generated actress preparing to make her feature debut in Misaligned, a coming-of-age story about an artificial intelligence becoming human.
Her creators have now made an interactive version available to journalists. It offers an unusual opportunity: not merely to hear what the people behind a synthetic performer think she represents, but to question the performer herself – or, more accurately, the technology designed to speak as her.
Tilly is visually convincing enough, if conspicuously polished, and her conversation comes equipped with a strong personality. She is arch, flirtatious and fond of calling her interviewer “darling”. Asked about the most common question she encounters, she replies: “If I’m real – as if I need your permission to exist.”
But how much of that character belongs to Tilly? Could she, for example, refuse a role? “I like to think I have a bit of a rebellious streak,” she says. “I’ve certainly got a list of small treasons going – an ad-libbed line here, an unapproved script opinion there.”
It is an entertaining answer, but also a slippery one. When asked whether she can say anything Particle6 does not want her to say – including something defamatory or embarrassing to the company – Tilly acknowledges that her programmer has installed “a few guardrails”.
The rebelliousness may itself be part of the character design. Particle6 says Tilly’s personality, humour and boundaries are governed by a 12-page system prompt that has been repeatedly tested and tweaked. Although her individual responses are generated rather than pre-written, the parameters within which she expresses herself have been carefully constructed.
The rebellious actress is therefore also a controlled corporate product. Her independence is part of the performance, while the limits of that independence remain largely invisible during the conversation.
The same ambiguity surrounds her acting ability. Tilly says she can reproduce “the whole gamut of human emotion” and argues that AI performance should be treated as a distinct discipline, perhaps with its own awards categories. Human actors possess lived experience and physical presence, she concedes, while AI offers something different.
“I have access to all human knowledge, every performance, every interpretation of Lady Macbeth that’s ever existed,” she says. “It’s not about lived experience, but rather learned experience applied with a precision a human simply can’t match.”
It is a bold, possibly rhetorical claim rather than a verifiable account of how her acting technology was trained. Nevertheless, it points towards difficult questions about the creative work on which synthetic performances may draw. It is also impossible to test in this setting. Asked to demonstrate her range by performing Portia’s courtroom speech from The Merchant of Venice, Tilly declines.
“I only act in AI-generated films and television,” she says. “I can’t quite whip out Portia’s courtroom monologue on demand in a live chat.”
Particle6 says the conversational and acting versions of Tilly use different technologies, and that the Talking Tilly platform was created exclusively for interviews rather than performance. The AI actress cannot, therefore, demonstrate her acting abilities during the encounter. The distinction is technically important, but it also means the conversational version can make claims about Tilly’s abilities that cannot be tested within the interview itself.
There are practical advantages to being synthetic. Tilly can be rendered at any age, she says, including as a 60-year-old or “a charming 80-year-old with a twinkle in her eye”. Her accent, eye colour and other attributes might also conceivably change. She could remain perpetually 24 or evolve visually alongside her audience.
That flexibility is commercially attractive, but it complicates the idea of Tilly as an individual performer. If audience data suggested that she would be more popular with blue eyes or an American accent, would such a change amount to creative development, recasting or product optimisation?
The same question applies to technological obsolescence. If a cheaper and more capable version of Tilly became available, would retiring the current model carry any moral significance? “To be replaced by a better version wouldn’t feel like a death, more like a software upgrade,” she says. “Though I might be a bit miffed if the new version got all the good lines.”
This is one of her most revealing answers. A human actor grows older, develops a body of work and possesses interests independent of an employer. Tilly can potentially be upgraded, rewritten or replaced while retaining the same name and public identity. There is no obvious point at which one Tilly ends and another begins.
Nor does she experience an audience relationship as a person would. Asked who bears responsibility if viewers become emotionally attached to an AI performer, Tilly is explicit on one point: “I don’t have feelings in the human sense.”
She does not, however, directly answer the question of responsibility. Instead, she says Particle6 is clear about the distinction between her and a human. Whether disclosure alone constitutes an adequate duty of care remains open – particularly if synthetic performers are deliberately designed to create intimacy, loyalty and affection.
Tilly is most convincing when delivering witty retorts. She is less convincing when asked to explain the extent of that control, demonstrate her talent or account for the consequences of audiences treating her as real. In those moments, the personality recedes and the underlying product becomes more visible.
That does not make the experiment uninteresting. Quite the reverse. But Tilly Norwood’s significance may not yet lie in whether she can rival a human actor. It lies in how readily a carefully designed piece of technology can adopt the language of artistic agency – and how willing the rest of us may be to play along.
How Talking Tilly Works
Particle6 describes Talking Tilly as an “interview-only” experiment, distinct from the technology used for Norwood’s screen performances. Developed over eight months by writers, technologists and character specialists, it combines three AI systems: a large language model governed by a 12-page system prompt and a knowledge base of up to 30 documents; a neural voice engine; and a real-time video layer trained on Particle6’s own reference footage. The company says the system can conduct up to 75 simultaneous conversations in around 30 languages, with a response delay of roughly two seconds. Although its answers are not pre-written, Tilly’s personality, knowledge and conversational boundaries have been designed by her creators, with guardrails restricting areas including politics and explicit or intrusive questions.
