This story first appeared in China Report, MIT Technology Review’s newsletter about technology in China. Sign up to receive it in your inbox every Tuesday.
If you could talk again to someone you love who has passed away, would you? For a long time, this has been a hypothetical question. No longer.
Deepfake technologies have evolved to the point where it’s now easy and affordable to clone people’s looks and voices with AI. Meanwhile, large language models mean it’s more feasible than ever before to conduct full conversations with AI chatbots.
I just published a story today about the burgeoning market in China for applying these advances to re-create deceased family members. Thousands of grieving individuals have started turning to dead relatives’ digital avatars for conversations and comfort.
It’s a modern twist on a cultural tradition of talking to the dead, whether at their tombs, during funeral rituals, or in front of their memorial portraits. Chinese people have always liked to tell lost loved ones what has happened since they passed away. But what if the dead could talk back? This is the proposition of at least half a dozen Chinese companies offering “AI resurrection” services. The products, costing a few hundred to a few thousand dollars, are lifelike avatars, accessed in an app or on a tablet, that let people interact with the dead as if they were still alive.
I talked to two Chinese companies that, combined, have provided this service for over 2,000 clients. They describe a growing market of people accepting the technology. Their customers usually look to the products to help them process their grief.
To read more about how these products work and the potential implications of the technology, go here.
However, what I didn’t get into in the story is that the same technology used to clone the dead has also been used in other interesting ways.
For one, this process is being applied not just to private individuals, but also to public figures. Sima Huapeng, CEO and cofounder of the Chinese company Silicon Intelligence, tells me that about one-third of the “AI resurrection” cases he has worked on involve making avatars of dead Chinese writers, thinkers, celebrities, and religious leaders. The generated product is not intended for personal mourning but more for public education or memorial purposes.
Last year, Silicon Intelligence replicated Mei Lanfang, a renowned Peking opera singer born in 1894. The avatar of Mei was commissioned to address a 2023 Peking opera festival held in his hometown, Taizhou. Mei talked about seeing how drastically Taizhou had changed through modern urban development, even though the real artist died in 1961.
But an even more interesting use of this technology is that people are using it to clone themselves while they are still alive, to preserve their memories and leave a legacy.
Sima said this is becoming more popular among successful families that feel the need to pass on their stories. He showed me a video of an avatar the company created for a 92-year-old Chinese entrepreneur, which was displayed on a big vertical monitor screen. The entrepreneur wrote a book documenting his life, and the company only had to feed the whole book to a large language model for it to start role-playing him. “This grandpa cloned himself so he could pass on the stories of his life to the whole family. Even when he dies, he can still talk to his descendants like this,” says Sima.
Sun Kai, another cofounder of Silicon Intelligence, is also featured in my story because he made a replica of his mom, who passed away in 2019. One of his regrets is that he didn’t have enough video recordings of his mom that he could use to train her avatar to be more like her. That inspired him to start recording voice memos of his life and working on his own digital “twin,” even though, in his 40s, death still seems far away.
He compares the process to a complicated version of a photo shoot, but a digital avatar that has his looks, voice, and knowledge can preserve much more information than photographs do.
And there’s still another use: Just as parents can spend money on an expensive photo shoot to capture their children at a specific age, they can also choose to create an AI avatar for the same purpose. “The parents tell us no matter how many photos or videos they took of their 12-year-old kid, it always felt like something was lacking. But once we digitized this kid, they could talk to the 12-year-old version of them anytime, anywhere,” Sun says.
At the end of the day, the deepfake technologies used to clone both the living and the deceased are the same. And seeing that there’s already a market in China for such services, I’m sure these companies will keep on developing more use cases for it.
But what’s also certain is that we’d have to answer a lot more questions about the ethical challenges of these applications, from the issue of consent to violations of copyright.
Would you make a replica of yourself if given the chance? Tell me your thoughts at zeyi@technologyreview.com.
Now read the rest of China Report
Catch up with China
1. Zhang Yongzhen, the first Chinese scientist to publish a sequence of the covid-19 virus, staged a protest last week over being locked out of his lab—likely a result of the Chinese government’s efforts to discourage research on covid origins. (Associated Press $)
2. Chinese president Xi Jinping is visiting Europe for five days. Half of the trip will be spent in Hungary and Serbia, the only two European countries that are welcoming Chinese investment and manufacturing. Xi is expected to announce an electric-vehicle manufacturing deal in Hungary while he’s there. (Associated Press)
3. China launched a new moon-exploring rover on Friday. It will collect samples near the moon’s south pole, an area where the US and China are competing to build permanent bases. Maybe the Netflix comedy series Space Force will look like a documentary soon. (Wall Street Journal $)
4. Huawei is secretly funding an optics research competition in the US. The act likely isn’t illegal, but it’s deceptive, since university participants, some of whom had vowed to not work with the company, didn’t know the source of the funding. (Bloomberg $)
5. China is quickly catching up on brain-computer interfaces, and there’s strong interest in using the technology for non-medical cognitive improvement. (Wired $)
6. Taiwan has been rocked by frequent earthquakes this year, and developers are racing to make earthquake warning apps that might save lives. One such app has seen user numbers increase from 3,000 to 370,000. (Reuters $)
7. Prestigious Chinese media publications, which still publish hard-hitting stories at times, are being forced to distance themselves from the highest-profile journalism award in Asia to avoid being accused by the government of “colluding with foreign forces.” (Nikkei Asia $)
Lost in translation
While generative AI companies have taken the spotlight during the current AI frenzy, China’s older “AI Four Dragons”—four companies that rose to market prominence because of their technological lead in computer vision and facial recognition—are grappling with profit setbacks and commercialization hurdles, reports the Chinese publication Guiji Yanjiushi.
In response to these challenges, the “Dragons” have chosen different strategies. Yitu leaned further into security cameras; Megvii focused on applying computer vision in logistics and the Internet of Things; CloudWalk prioritized AI assistants; and SenseTime, the largest of them all, ventured into generative AI with its self-developed LLMs. Even though they are not as trendy as the startups, some experts believe these established players, having accumulated more computing power and AI talent over the years, may prove to be more resilient in the end.
One more thing
During this year’s Met Gala, fans were struggling to discern real photos of celebrities from AI-generated ones. To add to the confusion, some social media accounts were running real photos in AI-powered enhancement apps, which slightly distorted the images and made it even harder to tell the difference.
One of the most widely used such apps is called Remini, but few people know that it was actually developed by a Chinese company called Caldron and later acquired by an Italian software company. Remini now has over 20 million users and is extremely profitable. Still, it seems its AI enhancement tools have a long way to go.