This week, Melania Trump apparently convinced her husband to do what I consider to be the first good thing of his tenure as president (either term).
With the recent signing of the ‘Take It Down Act’ by Donald Trump, it is now officially illegal in the US, at federal level, to publish a digitally altered video or photo of someone that is impossible to distinguish from real life, particularly depicting intimate images without consent.
It got me thinking though (and bear with me, there is a point to this personal download I am about to deliver). I am blessed with a very textured, but busy, life.
Today I’m writing this column in Kenya in between shooting scenes on a programme that I’m producing about the endangered African antelope species, the Bongo. Next week, I’m delivering a keynote for the Atlantec Festival Conference (I haven’t started that yet, please don’t tell them). In between, I’m hosting the Global Economic Summit in Killarney. I’ve just finished giving notes to my producer on Futureproof for a radio documentary about the search for life beyond our planet (you should listen, he’s done an amazing job).
Then there are the other inconveniences: bills, emails, lawyers, accountants – and the daily challenge of convincing my wife and kids I’m not just a rumour. It’s exhausting.
It got me thinking about how maybe this deepfake technology could be used for a greater good and get me back some downtime. Which led to the unorthodox question: could I clone myself?
In 2025, the year of generative AI, the answer of course is yes.
I get cloning
I began by training ElevenLabs.io, the most advanced voice synthesis and generation platform currently available. I fed their algorithm with three unadulterated hours they will never get back of pure me – professionally recorded from years of doing voiceover for TV and radio. The poor thing ingested all of this without complaint and slowly cranked through 180 minutes of self-important blathering. It took 11 hours to perform this task – I’m guessing it needed to take a few breaks in between to avoid whatever the AI equivalent of throwing up is.
Now, I have a digital voice. I can write whatever I want, click a button and hear myself reciting it verbatim, complete with my natural rhythm, cadence and baked-in pretentiousness. Not bad, I thought … but I wanted more. If I truly wanted my clone to be useful, it would need to know what to say all by itself.
The ‘brain’ part was easy. I built a custom AI assistant in ChatGPT that uses a technique called RAG (retrieval augmented generation).
I fed this assistant with reams of the sort of musings that I have posted over the years in the guise of opinion and journalism. Articles for The Irish Times, The Independent, the Business Post, LinkedIn posts and even my last two columns for SiliconRepublic.com.
I forced my new digital twin to analyse my writing ‘style’ and assume my identity. From now on, if I asked it to write me a script for a radio piece or a podcast, or write a fantastically engaging LinkedIn post, it must do so in a voice that was unmistakably my own.
After tweaking the instructions to get it just right, I instructed it to write me a piece on cloning myself (not this one you’re reading, I hasten to add). What came out was actually perfect – observations that were trite, unengaging and obvious. Perhaps the model was a little too good.
‘Frankenstein’s monster brought to life’
With a voice and a brain in place, all that was missing before I could unload all of my work onto my new clone, was something I could talk to: an interface – or for short – a ‘face’.
At the bleeding edge of generative video, researchers at Bytedance labs (the folks behind TikTok), have created models that can deliver something that feels like magic. Take an old photograph of your grandparents and it will bring them back to life. It can convincingly animate a portrait by Van Gogh or get Einstein to perform hip hop like he was born to rock the mic. From one single image, what they can do is truly uncanny.
At the commercial level, though, the best currently available method of generating a fake me is by using HeyGen’s digital avatar service. This Chinese leader in deepfake technology can take a number of videos from your phone and generate a likeness of you that is impressive, but admittedly stops shy of convincing. I pushed on regardless, dismissing the doubts that began to creep in. The promise of a digital slave was too great.
I fed it a bunch of selfies, connected it to my ElevenLabs voice and attached my creative brain. After just one day, I selected the live chat option in HeyGen and found myself face to face with the closest possible version of me that billions of euros worth of the most revolutionary technology humanity can provide. Frankenstein’s monster brought to life.
Confronting your own self can be difficult at the best of times, but this was unbearable. Perhaps it was the worst of me, perhaps it was the best. Either way, I do not like thee, Dr Fell.
My clone lived for only 8 minutes. While each cog in the machine gave me something close to a synthetic analogue, when put together the whole was decidedly less than the sum of my parts: like Max Headroom, but pathetic, predictable and torturously upbeat. Melania was right – clones are bad.
Without a moment’s hesitation, I murdered the closest thing to me with just three clicks of a mouse.
I learned two things from the exercise.
First, we are only 18 months away from this technology being utterly convincing. What will that mean for video, live events, Zoom meetings? Will our digital selves just meet each other while we play golf? Will our AI secretaries give out about other AI secretaries at an AI water cooler?
The future of digital twins may throw up many weird alternative universes we haven’t begun to consider.
Second, no matter how flawed we may be, human beings are extraordinarily complex and sophisticated organisms. We are each unique and beautiful, one of a kind and completely irreplaceable … for now.
Further information on Jonathan’s Get Started with AI.
Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.
