DeepBrain AI's Jungwook (John) Son said limits on current technology meant that developing AI chatbots modelled on deceased people was not only uncomfortable but also required a lot of time and resources to get done right.
California-based generative AI startup DeepBrain AI released re;memory 2, a service to help grieving families recreate their deceased loved ones using generative AI, in June this year.
This was just the latest development in a fairly new area of generative AI, now being referred to as "grief tech". Earlier this year, the use of AI to recreate deceased family members gained prominence, with several Chinese companies making waves, including SenseTime who delivered an AI-generated video of their founder Tang Xiao'ou, who had died late last year, to their employees.
Shortly after this, families that had come out in favour of using AI modelled on their deceased family members were attacked for making use of the tech. While some said that it was an unethical use of AI, others said that it served to worsen the mourning period.
However, while many have made their minds up on generative AI used for grieving, the methods by which this is done has changed significantly, as companies realise the challenges in making chatbots modelled on deceased loved ones to converse with grieving families.
The timeline of re;memory's development reflects this realisation, as it's initial iteration was in the form of a conversational chatbot.
Speaking with AIM, DeepBrain AI global marketing manager Jungwook (John) Son said that current technology meant that developing AI chatbots modelled on deceased people was not only uncomfortable but also required a lot of time and resources to get done right.
"Our first version of Rememory was conversational and it was very limited. The actual target model had to be in our studio to be generated as an AI avatar, which meant pre-preparing ahead of time. Which is, I would say, a very uncomfortable moment," said Son.
Additionally, due to the sensitive nature of the product, having to get the actual likeness and manner of speaking down while having a back and forth was a monumental task.
In re;memory's case, the company had this realisation back in 2023, about a year after its first launch. "We decided to discontinue because we learned the difficulty of actually producing the product. We got some good feedback that's why we decided to pivot and create a video product rather than a conversational product. So, we relaunched it," he said.
Which seems to be something that other companies that have gone into the field of grief tech have realised as well. Many of the companies offering the services, usually offer a video-based product, rather than a conversational one.
Earlier this year, popular Taiwanese musician Bao Xiaobo had gone viral for generating a video of his deceased daughter using AI. During an interview on the subject, he said, "AI is a tool for expressing yearning, a way to express missing someone."
Similarly, while there are beliefs that AI is being used to recreate or even replace deceased loved ones...
Families choosing to use these services seem to be doing so with an understanding that it's more of a tool rather than a replacement for grief.
Deepbrain's model is developed in-house by the team, as their primary product is AI avatars, having provided their services to large-scale retailers, finance and banking companies, and even sports teams across the world.
In terms of the funeral industry, DeepBrain AI has partnered with one of South Korea's largest funeral service providers, Preedlife, to offer its re;memory services to grieving families.
The current iteration of re;memory generates a video of the family's deceased loved one, with inputs from the family at every step of the way, from the message they want delivered to the end product. This will be displayed during the funeral of said member, and will also be available to the family, much like looking through a photo album.
"It would be something that we discuss with the family. There will be an episode that they want to talk about and then we'll moderate it and generate the message using memorable scenes. We try to focus on capture unique facial expressions and voice intonations so that we can try to make it real as possible," Son said.
DeepBrain AI's shift from conversational chatbots to video-based memorials reflects a broader industry trend, driven by the realisation that creating lifelike digital avatars of deceased loved ones is riddles with difficulties and sensitivities, as well as other factors like limited reach as well.
"Because of the limited access, we are currently trying to reach out to more people. So this is why we currently decided to pivot because of the difficulty to create it as a product, whereas now since its a video, it's more reachable, it's more continuable," Son said.
Deepbrain AI's collaboration with Preedlife is the first such partnership to happen. However, with a growing acceptance of grief tech, this partnership could be replicated across the board, as families find a certain comfort in these snapshots of their deceased loved ones.
As Son said, "We sincerely think the technology will evolve. The feedback that we get is that this does help families emotionally grieve. Grieving is something that people take very seriously.
But with this type of supplement, I would say, it helps process things more smoother and healthier, and that's the basic approach that we're trying to have here as well."