Tech Talk
Dawn of the deadbots By Michael E. Duffy A dvances in sound and video processing, along with the advent of generative AI, now make it possible to create a “deadbot” (also
The Eternos website ( eternos. life/chat-demo/ ) has a slightly different demo. It shows a customer interacting with their “avatar” via speech. Despite lacking the video avatar of the Storyfile demo, it appears to be a better, more fluid conversationalist, and reacts as you might expect if you’ve had experience with using ChatGPT. On the other hand, Eternos charges $15,000 to create a replica of you, whereas StoryFile charges a mere $500. Regardless of what you pay, one key issue is when the hardware and software that power your avatar belongs to a business, what assurance do you have that the business will be around for the long run? Storyfile filed for Chapter 11 bankruptcy in May, although it is seeking to reorganize the business so it can continue to deliver service on a more sustainable basis.
called a “griefbot”) which allows survivors to see, hear and interact with a recreation of a deceased person. As you might expect, companies are springing up to offer this service, and not far behind, ethicists are calling for regulation of the technology, which they describe as having the potential to cause “huge distress to people if they are subjected to… alarmingly accurate AI recreations of those they have lost.” (More about the potential negatives at tinyurl.com/2dg6rb78 ). People are already fooled by the technological recreation of living people, known as deepfakes. In one case, a financial worker in Hong Kong transferred $25 million to thieves based on a Zoom call with what they thought were their actual co-workers. The co-workers were, in fact, just realistic facsimiles created by the thieves. Similarly, deepfakes have been used
An AI-generated William Shatner, above, is ready to chat about the ‘digital afterlife’ at storyfile.com .
to create pornographic videos of well-known stars such as Taylor Swift. Deepfakes are basically “synchronization” technology—images taken from one person are (more or less) seamlessly grafted onto video of another. For example, the image of a public figure replaces the face of an actor in a video. A neural network called an autoencoder is used to abstract the essentials of the video containing the actor (positions of their face and body), and those essentials are replaced with data representing the face and body of the public figure and reconstituted into the deepfake video. It’s really just the clever application of mathematics (see the “autoencoder” article in Wikipedia), but it seems a lot like magic. A similar approach can be used to transform audio. It’s now possible to recreate someone’s voice with only a few seconds of captured audio. Of course, recording a person speaking a set of phrases which cover the “essentials” of a voice gives an even better result. Finally, mix-in generative AI (i.e. large language models, or LLMs). An LLM can be trained on what a person has said and written, and using its “autocomplete on steroids” ability, generate brand new words in that person’s style. This combination of voice, video and interactivity starts to seem very realistic. Here’s a fake William Shatner at Storyfile ( storyfile.co m), one of the so-called “digital afterlife services.” ( Forbes magazine refers to them as “digital embalmers.”) Shatner’s avatar uses his answers to pre-recorded questions, so it’s not quite as sophisticated as I’ve described above. You can’t get answers to questions he didn’t answer, and there can be a brief “glitch” as the avatar transitions to answering a question which breaks the illusion of actually speaking to a real person.
Why would someone want to create a deadbot in the first place? Deadbots are not generative in the same way as a living person, so your heirs can’t really come to it and get the benefit of your expertise by asking questions. I suppose there could be some comfort in being able to see and hear someone “as they were,” moving and responding to your presence, and telling beloved (but all-too-familiar) stories. But Grandpa ‘s avatar isn’t going to tell you why he disowned your aunt, or where he hid those shares of Berkshire-Hathaway stock back in 1980. The LLM behind the avatar can make something up (and might, without safeguards in place), but it’s just a sad copy of Gramps. Part of the appeal is the convenience. Based on what I’ve read, none of these services require as much effort as the tried-and- true method of writing things down, perhaps as a memoir, or just a simple narrative of the facts of one’s life: dates, names, places, events, things you like and dislike. And for gosh sakes, print it out. History shows us that things on paper last a long time. Not so that 3.5-inch diskette, which Sony stopped making in 2011. On the other hand, if someone derives comfort from what these deadbots can offer, who am I to object? As one Eternos customer, Jan Bommer, said to his wife, “Think of it sitting somewhere in a drawer, if you need it, you can take it out, if you don’t need it, just keep it there.” g Michael E. Duffy is a senior software engineer for Atlanta- based mobile gaming company Global Worldwide ( global- worldwide.com ), who lives in Sonoma County. He has been writing about technology and business for NorthBay biz since 2001.
July 2024
NorthBaybiz 19
Made with FlippingBook - Online Brochure Maker