Reid Hoffman’s AI clone: So ‘human’ that it wipes a booger on the table

What would you ask if you found yourself sitting across the screen from an interactive, digital version of… Well, yourself?

Well-known Silicon Valley figure Reid Hoffman – co-founder of LinkedIn, former board member of Paypal, OpenAI, and Co-founder of Inflection AI – recently found himself in exactly this situation, interviewing an AI avatar of himself created using generative AIs from Hour One (for video) and 11ElevenLabs (for audio).

These AI models were trained specifically on the last 20 years of “in real life” (IRL) Reid’s writings, speeches, interviews and podcasts. So they’re full of his own personal ideas, turns of phrase, tone of voice, and indeed physical mannerisms – and designed to recreate how Hoffman himself would listen, react, answer and converse with another individual.

The idea is for public figures to be able to offer one-on-one personal interactions, for purposes ranging from coaching to simple fan interactions, to far more people, and at far lower cost than they could ever schedule in the flesh.

It’s a pretty incredible concept, which we first encountered in a chat with Motion Capture master Remington Scott of Hyperreal, a company that creates digital twins of celebrities, musicians, authors – Whether they’re currently alive or not.

“We’re working with a best-selling author,” Scott told us. “He’s had more than 20 books on the New York Times list. Talented guy. And he wants to create a conversational AI – you’ll talk to him, he’ll talk to you. So the training model we’ve used for his conversational brain is all his work. It’s not cross-pollinated with other things like ChatGPT is, this is his own words. So you’re speaking directly to the words that he says, and the thoughts and feelings he’s collected. We do tests, asking him and the AI the same questions. And he can’t see the AI’s answers, but they’ll both answer almost exactly the same. It’s as authentic as it’s gonna get.”

So we knew this was coming, but this is the first time we’ve someone interviewing their own digital copy. “I think these avatars,” writes Hoffman, “when built thoughtfully, have the potential to act as mirrors to ourselves – ones that reflect back our own ideas and personality to examine and consider.”

Reid posted the following 14 minute-long Reid vs Reid interview to YouTube. The conversation appears completely seamless and flowing, but we’d take that with a grain of salt. There’s typically an awkward processing pause between question and answer with most conversational AIs we’ve encountered, and there’s a jump-cut in the video every time AI Hoffman goes to speak – so there’s clearly some editing magic involved here.

But one of the most striking things is the body language – even if it’s probably been adjusted afterward for timing. AI Reid appears to be in a constant state of active listening, and seems downright playful at times whilst listening to IRL Reid speak – nodding its head, smiling, shrugging and in general being very much on-vibe when it’s not talking.

It almost looks like too much, honestly – until you flip through a couple of Masters of Scales podcast videos and see how enthusiastically IRL Hoffman speaks with his guests… I’m now convinced AI Reid is giving this exactly the right amount of sauce.

At first glance while watching AI Reid, you can tell that there’s something not quite right. Something synthetic. But that’s also because you know what you’re looking at – and either way, this technology is relatively new; at the current rate of AI progress, it’s likely to get astronomically better long before you’d expect. Within a few months, or years at most, there’s every chance you’ll struggle to tell a real person from an AI avatar in a video chat.

Just the sheer… Humanity of this thing can be astounding at times. When AI Reid speaks, you can audibly hear each sharp inhalation breath between sentences, the same as IRL Reid does when he’s excited and explaining something. Small details like these make AI Reid remarkably believable.

There are moments when AI Reid fidgets impatiently, waiting for IRL Reid to finish talking as if it already knows the answer to his questions. It replicates IRL Hoffman almost disturbingly well in movement and in timing – and while the timing might be down to editing, the movement certainly isn’t.

There are other moments that highlight why you might want to be very careful in front of cameras, if the vision’s going to train up an AI model. Our favorite, below, is around 11:44 in the video, when AI Reid wipes its runny nose as it listens and nods, then contemplatively wipes an AI-generated booger on the table.

The idea of having a digital twin of yourself can create plenty of exciting, creative, educational and helpful opportunities, as Hoffman outlines in his AI-self interview. Doctors having the ability to “see” dozens or more patients simultaniously. Teachers teaching around the world, tailoring their communications personally to thousands of students learning at different levels concurrently.

Indeed, Digital AI twins have already started taking people’s place for stressful situations like, say, video interviews for remote jobs:

It won’t take long for them to get good enough to win contracts away from meat-space competitors – and possibly even do quite decent work, now that I think about it, as long as they can avoid going to the Christmas party.

And that’s about the least of our problems; this also has the potential get very dark; interactive clones can now be harnessed for all sorts of purposes.

People are already being scammed by the likeness of Taylor Swift and Joe Rogan through clever deepfake videos – AI counterparts offer an entirely new level of manipulative coercion. Your elderly relatives already fall for phone scams – imagine how much more effective these could be if it was a fully interactive video call using your face, your voice, your speech patterns… Or Tom Hanks. Or Donald Trump.

Assuming you’ll have the choice, would you allow everything you’ve ever done, written, and recorded to be databased and turned into a digital copy of yourself that can interact freely, without you being there to audit its responses? How would you use it? I’d love to hear your thoughts.

Source: Reid Hoffman