The patent at a glance
| Detail | Value |
|---|---|
| Number | US 10,853,717 B2 |
| Title | Creating a conversational chat bot of a specific person |
| Inventors | Dustin I. Abramson and Joseph Johnson, Jr. |
| Owner | Microsoft Technology Licensing, LLC |
| Filed | 11 April 2017 |
| Application published | 11 October 2018 (as US 2018/0293483 A1) |
| Granted | 1 December 2020 |
| Status | Listed as active on Google Patents, with an anticipated expiry in 2039 |
One detail is often missed: the application was public from October 2018. It only became news when the patent was granted, more than two years later.
What does the patent actually describe?
The patent describes a system that collects “social data” about one person and uses it to build a chatbot that converses in that person’s personality. The listed sources include images, voice data, social media posts, electronic messages and written letters.
It is broad about who the person could be. The text says the chatbot may represent “a past or present entity (or a version thereof)”, such as a friend, a relative, an acquaintance, a celebrity, a fictional character or a historical figure. It also mentions a version of yourself.
- A personality index. The person’s data is organised so the bot can answer in their style and with their knowledge.
- Filling the gaps. When the person’s own data has no answer, the system may draw on crowd-sourced conversation data from other people.
- A deceased relative example. The patent describes a bot of a user’s late relative that should not know about events after that person’s lifetime, and that may ask the user questions to fill in missing periods.
- Voice and image. It suggests building a “voice font” from recordings and a 2D image or 3D model of the person from photos.
The gap-filling idea matters. Any system that fills missing knowledge with other people’s words can produce answers the real person never gave. Modern generative chatbots face the same problem.
Why it made headlines in January 2021
In late January 2021, the grant was picked up widely online and by outlets including CNN. Many people compared it to the Black Mirror episode “Be Right Back”, in which a young woman talks to an AI built from her late partner’s online messages. Our page on how much of “Be Right Back” is real now covers that comparison.
The timing also mattered. Large language models had just made fluent imitation much easier. A few months earlier, in September 2020, a man had used one to simulate his late fiancée, a story that became widely known later in 2021. It is told in our page on Project December.
What Microsoft said
Tim O’Brien, then Microsoft’s general manager of AI programs, responded on Twitter in January 2021. As reported by CNN, he said he had confirmed there was no plan to build it, that the application from April 2017 predated the AI ethics reviews the company now carried out, and, agreeing with many commenters: “yes, it’s disturbing”.
Microsoft has not announced any product based on the patent. CNN noted that the company had since created an Office of Responsible AI and an internal ethics committee to oversee such work.
A patent is not a product
It helps to understand what a patent does. It gives its owner the right, for a limited time, to stop others from using the specific method described in its claims. It does not mean the owner has built anything, or ever will. Large companies file many patents to protect ideas their engineers explore, including ideas they later abandon.
A patent also does not stop other companies from building different systems for similar goals. Whether a particular product would infringe depends on the exact claims. Laws differ by country; this is general information, not legal advice.
What has changed since 2017
| Idea in the patent | Where things stand in 2026 |
|---|---|
| Chatbot trained on one person’s messages | Widely possible with large language models, from hobby projects to commercial apps |
| Voice font from recordings | Voice cloning exists, and responsible providers now ask for the speaker’s active participation and consent |
| 2D image or 3D model | Video avatars and recorded-answer systems exist; reporting in 2024 described a growing market for AI avatars of the dead in China |
| Filling gaps with crowd data | Built into how generative models work, which is why they can invent things |
| Clear rules for disclosure | The EU AI Act requires chatbots to tell people they are interacting with an AI, with these duties applying from August 2026 |
Microsoft itself tightened its approach. In June 2022 it published its Responsible AI Standard and explained that its Custom Neural Voice service was restricted to approved customers, with safeguards to ensure the speaker actively takes part when a synthetic voice is created.
What it means if you are considering an AI clone
The patent is a useful map of the questions any “chatbot of a specific person” raises: whose data, whose consent, what happens when the data runs out, and how clearly the bot is labelled. Kurzweil’s attempt to recreate his father, described in our page on the Fredbot, shows a family answering the same questions privately.
Memory Clone is not affiliated with Microsoft or this patent. It is a Windows app, currently in alpha, that builds an AI simulation from chats you export yourself and learns style separately for each relationship. It is text only: it does not clone voices or build 3D models. Every clone is presented as AI, and cloning another living person requires their permission under our Acceptable Use Policy. You can find out more on the download page.
Domande frequenti
Did Microsoft build a chatbot of dead people?
No product has been announced. In January 2021 Microsoft’s general manager of AI programs said there was no plan to build it.
Does the patent mention people who have died?
Yes. Among other examples, it describes a chatbot of a user’s deceased relative and how to handle periods with missing data. It also covers living people, celebrities, fictional characters and yourself.
When does the patent expire?
Google Patents lists an anticipated expiry in 2039. Patent terms can end early, for example if maintenance fees are not paid.
Can Microsoft stop other griefbot apps with this patent?
Only products that use the specific methods in its claims could be affected, and that is a legal question for each case. This is general information, not legal advice.
Why did the idea seem disturbing to many people?
It raised questions about consent, about bots inventing words a person never said, and about grief. These are the same concerns researchers raise about griefbots today.
Fonti
Verificato il . I prodotti di terze parti e le pagine di assistenza cambiano spesso; segui i link per i dettagli aggiornati.
- US10853717B2: Creating a conversational chat bot of a specific person (Abramson & Johnson, granted 2020) (Google Patents)
- Microsoft patented a chatbot that would let you talk to dead people. It was too disturbing for production (Clare Duffy, 2021) (CNN)
- Did Microsoft Patent a Chatbot That Lets Users Talk to the Dead? (Snopes)
- Microsoft’s framework for building AI systems responsibly (Natasha Crampton, 2022) (Microsoft)
- Regulation (EU) 2024/1689 (Artificial Intelligence Act) (EUR-Lex)
- Deepfakes of your dead loved ones are a booming Chinese business (Zeyi Yang, 2024) (MIT Technology Review)
This article is general information, not medical, legal or financial advice. Memory Clone creates an AI simulation from the material you provide; it is not the person.