AI avatars for dead people raise ethical and security concerns

Grandma, what big AIs you have.

South Korean firm DeepBrain AI offers consumers the possibility of talking to their departed loved ones again through the company’s use of AI video avatars.

DeepBrain’s Re;memory vertical conducts seven-hour, in-person interviews with people with terminal illnesses, or others who want to make sure they’re not forgotten, to gather the information they will share with those they leave behind.

“Re;memory is more than just a memorial hall with photos and videos,” an introductory web page declares. “It’s a place where you can share memories with your passed [loved] ones through an actual conversation.”

Deathics. When one IT Brew reporter saw the Re;memory part of DeepBrain’s booth on the floor at CES ’23, he posted a picture of it to Twitter. Reactions to the tweet were uniformly hostile: responders referred to it as “dystopian,” and implied it was unethical.

In an interview conducted at CES, DeepBrain’s California business development manager Joe Murphy told IT Brew that any ethical concerns over Re;memory’s technology were effectively in the eye of the beholder, and that people would need to make their own decisions on what’s right or wrong.

“That’s a very personal decision,” Murphy said. “Different people, even different cultures, look at death differently.”

Brian Finch, co-chair of Pillsbury Law’s privacy and cybersecurity practice, largely agrees. He told IT Brew that integrating AI technology with end-of-life procedures is so new that there is no “hard-and-fast set of ethics associated with that as of yet.” A more pressing concern for Finch is whether or not the person whose avatar is being generated was capable of giving consent or coherent enough for that choice.

Necro-security. With such a potentially invasive technology comes concerns over security. Murphy said that Re;memory runs out of an AWS cloud, which allows it to take advantage of the security measures in place. When asked if someone could maliciously hack Re;memory to cause distress or pain—by, say, having your late family member say hateful things to you—Murphy assured IT Brew that there are measures in place to stop that from happening.

“We have all the security procedures, like any AWS implementation, but then we go a bit further where we have text filters, so you can lock out curse words and offensive terms,” Murphy said. “That can’t even be entered; the model will never utter those phrases.”

Finch takes the zero-trust view that adversaries are always going to find a way in and thus mitigation and defense at all levels is most important. Finch stressed that this was not a specific comment on DeepBrain’s security measures but rather a general comment on the state of security in the industry as a whole: “I assume somebody will break in just because everybody’s gonna be broken into at some point.”

Of further concern is that the information provided in the interviews could hypothetically be used for social engineering with possible applications to crack passwords or answer security questions.

“Hackers are very inventive in a malicious way, and they’re always looking for new sources of information—new ways to gather data that would help them actually go ahead and execute some more of their criminal schemes,” Finch told IT Brew. “I think this would definitely be part of the universe of information sets that they would look into.”—EH

Do you work in IT or have information about your IT department you want to share? Email [email protected].