16 September 2026
Heard In AI

A CEO avatar on call 24/7, built from a life's recordings rather than a brain scan

Martine Rothblatt told Moonshots that her digital double, the Marvatar, is available to all 2,000 United Therapeutics employees, "any question 24-7, 365." She uses it to argue that videos, audio, documents and conversations, fed to today's language models, can reconstruct a person without copying every neuron.

A briefing reports one development when it happens. We correct or clarify it later; a new development gets a new briefing. How our formats work

Martine Rothblatt wants to be, in her words, "the most accessible CEO anywhere in the world." Her solution is not longer office hours. On Moonshots with Peter Diamandis, she said that a digital double of herself — she calls it the Marvatar — is available to all 2,000 employees of United Therapeutics, the biotechnology company she founded. They can ask it "any question 24-7, 365."

Diamandis summarized what sits behind it: the avatar has been trained up with everything she has ever said and written. Rothblatt added the second ingredient — "all the best LLMs out there." LLM is short for large language model, the kind of system behind ChatGPT and similar assistants: software that has learned patterns from enormous amounts of text and can hold a conversation. The Marvatar, on her account, is that general machinery pointed at one person's record.

She described one outing for it as a high point of her life. At what she called the 50th anniversary of AI at UCLA, she got Ray Kurzweil — the inventor and futurist whose books on machine intelligence she credits with shaping her thinking — on stage with the Marvatar in front of a 2,000-person audience of AI experts. What struck the room, she said, was that there was no latency: the avatar answered immediately, and would interrupt Kurzweil to say, well, how about this. "Marvatar kept up with Ray, which is actually better than I can do," she said.

From a robot head to a company intranet

This is her second attempt. Around 2010, with the computer programming available at the time and the cooperation of the roboticist David Hanson, she built a digital doppelganger of her wife, Bina. The result, Bina48, became the subject of a documentary that Rothblatt said drew rave reviews at several film festivals.

LifeNaut, the project that hosts the robot's description, says Bina48 was built by Hanson Robotics and introduced in 2010, and that its construction drew on interview transcripts, a laser-scanned likeness, and facial and voice recognition. Its information combines material from several people, including Bina Aspen. It can converse, make facial expressions, move its head and eyes, and track motion. LifeNaut presents the robot not as a finished person but as an early exploration of its hypothesis: that detailed personal records, combined with future software, might one day produce a conscious analogue of someone.

The Marvatar is the same bet with better tools, and with a workplace job attached.

The argument with Kurzweil

Diamandis raised mind uploading — the idea of transferring a person's roughly 100 trillion neurons, and their consciousness, into a computer. Here Rothblatt parts company with her main influence. Kurzweil's preferred route, as she described it, is to load the body with neural nanobots, microscopic machines that would sit among the neurons and wirelessly transmit everything a person is thinking.

"I think we're not that special," she said. If you have all the external indications of a person — their videos, their audio, their documents, their conversations — then the models running right now, she argued, will be able to integrate that information and produce a digital simulacrum indistinguishable from the original. You do not need every single neuron. She reached for a medical comparison: people undergo hemispherectomies, she said, in which half the brain is removed, and they are the same person. Putting a nanobot on every neuron is, in her view, overkill.

The rest is entrepreneurial habit. What she always does in her companies, she said, is look for a simpler, faster way to the same goal. If somebody tells her she has achieved only 95 percent of it, "I'll take it. Because I want 95% now is vastly better than 100% in some misty future."

Diamandis pressed on the rival approach: digitizing the connectome, the complete wiring diagram of a brain, which researchers have mapped in the fruit fly Drosophila and are attempting in mice. Rothblatt declined the role of wise old expert but said it is possible — and noted that she backs long-shot routes with money. She and Bina have both been members of Alcor, the cryonics organization, for 20 years, have made multimillion-dollar donations to it, and are signed up to be fully cryopreserved. Full body, she said, "because I can afford it." She also funds a National Space Society prize for living on an O'Neill settlement — a large rotating space habitat — within our lifetimes, and thinks the practical way there is to digitize yourself and be re-embodied at the other end.

What an archive is and is not

The archive-first approach has a consumer-facing version. LifeNaut offers free mindfile accounts, where people organize personal information and preserve distinctive experiences for their families and for future generations. It describes participation as a long-term computer-science project investigating whether digital technology might eventually extend life, and the affiliated Terasem Movement Foundation says it stores mindfiles and biological records for possible future revitalization, subject to participants' consent and to technological advances. The starting material is accumulated biography, not a scan of neural wiring.

Rothblatt set out the vocabulary more than a decade ago in her book Virtually Human, whose original text carries a 2014 copyright: a mindfile records someone's memories, beliefs, preferences and experiences; mindware is the proposed software for reproducing mental processes; a mindclone is the resulting digital identity. The book treats the rights and moral status of such beings as prospective questions rather than settled ones.

What exists today is narrower than that. The Marvatar answers employees' questions quickly and in her manner, drawing on her record and on commercial language models; whether anything of her experience continues inside it is the hypothesis, not the demonstration. Earlier on the same show, a panel treated uploading as a near-term forecast attached to AI-generated worlds, with one speaker saying he expected it solved within the next few years — a prediction Diamandis asked him to start dating with specific years.

Rothblatt's contribution is a cheaper path and a lower bar. While the connectome mappers work through flies and mice, her 95 percent is already logged in, taking questions from 2,000 colleagues at any hour.

Share this article

Go to the original

Sources & further reading

  1. 01
  2. 02
  3. 03

Connected ideas and articles

From the conversation

Podcast episodes

Article history

Updates to this article

Tags

Aaron Levie's question for AI memory: what belongs in the weights?

On Training Data, Box CEO Aaron Levie was asked where enterprise AI memory is heading — retrieval, or models whose weights absorb a company's knowledge. His answer started with a lawyer who can see five matters and whose access changes daily, and ended with a wish for a rubric deciding what gets baked in and what stays a lookup.

6 min read

Better data beat better architecture — but the panel split on its shelf life

A Moonshots panel unpacks Dwarkesh Patel and Jerry Han's experiment, which found that improvements in training data delivered a 12-fold compute-efficiency gain between 2019 and 2025 against 3.7-fold for architectures and training recipes — at small scale, on easy benchmarks. The panel then splits over whether a company's proprietary data is a durable advantage, with a $32 billion data-subsidiary valuation on one side and the fate of BloombergGPT on the other.

7 min read

Why a Moonshots panel thinks China's AI tokens go to video and America's to code

Alibaba's Wan 3.0 and a relayed claim that 70% of Chinese AI token use goes to video sent the Moonshots panel into an argument about money: one guest said American labs chase revenue per token while Chinese labs give their weights away, another said video is the only market that will trust a Chinese model. They ended up disagreeing about whether world models or text models reach self-improving AI first.

6 min read

Would she pay the real price? Zitron's test for AI adoption

On The Diary of a CEO, critic Ed Zitron praises a chatbot for reading a troubleshooting log and for helping fix his son's Minecraft mod, then argues that neither is worth a trillion dollars. The host counters with his fiancée's one-woman business and his chief of staff's inbox. The argument turns on tokens, subscription rate limits and who is paying the real bill.

7 min read