Billionaire LinkedIn co-founder Reid Hoffman has recently released a video ‘interview’ with his new digital avatar, Reid AI. Built on a custom GPT trained on two decades’ worth of Hoffman’s books, articles, speeches, interviews, and podcasts, Reid AI utilizes speech and video deepfake technology to create a digital clone capable of approximating its source subject’s mannerisms and conversational tone. For over 14 minutes, you can watch the two Hoffmans gaze lovingly and dead-eyed, respectively, into the tech industry’s uncanny navel. In doing so, viewers aren’t offered a “way to be better, to be more human,” as the real Hoffman argues—but a way towards a misguided, dangerous, unethical, and hollow future.
Many people might shudder at the idea of unleashing a talking, animated AI avatar of themselves into the world, but the tech utopian “city of yesterday” investor sounds absolutely jazzed about it. According to an April 24 blog post, he finds the whole prospect so “interesting and thought-provoking,” in fact, that he recently partnered with generative AI video company Hour One and the AI audio startup 11ElevenLabs to make it happen. (If that latter name sounds familiar, it’s because 11ElevenLabs’ product is what scammers misused to create those audio deepfake Biden robocalls earlier this year.)
After teasing a showcase of his digital clone for months, Hoffman finally revealed a (heavily edited) video conversation between himself and “Reid AI” last week. And what does the cutting-edge, deepfake-animated culmination of a custom built GPT-4 chatbot reportedly trained on all things Hoffman? A solid question—and one that isn’t easy to answer after watching the surreal, awkward, and occasionally unhygienic simulated interaction.
“Why would I want to be interviewed by a digital version of myself?” Hoffman posits at the video’s outset. First and foremost, it’s apparently to summarize one of his books for an array of potential audience demographics: the smartest person in the world, 5-year-old children, Seinfeld fans, and Klingons. While Hoffman seems to love each subsequent Blitzscaling encapsulation (particularly the “smartest person” one) they all sound like it came from a ChatGPT prompt—which, technically, they did. The difference here is that, instead of only a text answer, the words get a Hoffman vocal approximation layered atop of a (still clearly artificial) video rendering of the man.
Amidst all his excitement, Hoffman—like so many influential tech industry figures—yet again betrays a fundamental misunderstanding of how generative AI works. Technology like OpenAI’s GPT, no matter how gussied up with visual and audio additions, is not capable of comprehension. When an AI responds, “Thank you” or “I think that’s a great point,” they don’t actually experience gratitude or think anything. Generative AI sees sentences as lines of code, each letter or space followed by the next, most probably letter or space. This can be adapted into conversational audio and dubbed to video personas, but that doesn’t change the underlying functionality. It simply received new symbolic input that influences what basically amounts to a superpowered autocorrect system. Even if its language is set to Klingon, as Reid AI offers at one point.
So when Reid AI warns Hoffman a wrong answer may result “because I misinterpreted the information you gave, or I don’t have the full context of your question,” Hoffman doesn’t pause to explain any of the above facts for viewers. He instead moves along to his next conversation point, which usually involves a plug for his books or LinkedIn.
[Related: A deepfake ‘Joe Biden’ robocall told voters to stay home for primary election.]
Meanwhile, Reid AI’s visual component is supposedly meant to simulate many of Hoffman’s conversational mannerisms and queues. Judging from Reid AI’s performance, these largely boil down to stilted attempts at “nodding vigorously,” “emphatically tapping to illustrate a point,” and “picking his nose.” As New Atlas points out, the moment at 10:44 is an odd quirk to include in such a clearly condensed and edited video—perhaps meant to illustrate some of humanity’s more awkward, relatable traits. If so, it does little to distract from the far more absurd and troubling sentiments said by both Hoffman’s.
Reid AI expounds on boilerplate techno-libertarian talking points for fostering a “framework that fuels innovation.” Hoffman repeatedly opines that any concerns about bias, privacy, labor, and digital ownership concerns are just “start[ing] with the negative and [not realizing] all the things that are positive.” The digital clone regurgitates bland, uncreative ways to spruce up Hoffman’s LinkedIn page, like adding “personal flair” such as a fun and colorful header image.
But the most worrisome moment arrives when Hoffman contends “Everyone should be asking themselves, ‘What can I do to help?’” make AI like digital avatars more commonplace. He even goes so far as to equate the current technological era to Europe’s adoption of the steam engine, which made it “such a dominant force in the entire world.” (Neither he, nor Reid AI, concede the other tools involved in the industrial revolution, of course—namely a colonialist system built on the labor of millions of exploited and enslaved populations.)
Hoffman says future iterations of Reid AI will add “to the range of capabilities, of things that I could do.” It’s an extremely telling sentiment—one implying people like Hoffman have no qualms with erasing any demarcation between their cloned and authentic selves. If nothing else, Hoffman has already found at least one task Reid AI can handle for him.
“I am curious to know what others’ thoughts are on how to mitigate impersonation and all other types of risks stemming from such a use-case and achieve ‘responsible AI,’” one LinkedIn user asked him in his announcement post’s comments.
“Great question… Here is Reid AI’s answer,” Hoffman responded alongside a link to a new avatar clip.
