I recently listened to an audio clip of Joe Rogan declaring that he’s going to start sponsoring a hockey team made up entirely of chimps! He’s tired of people thinking that chimps cannot compete with humans on an athletic level. And he’s got these chimps working hard; they’re throwing kettle bells and they’re on a strict diet of bone broth and elk meat! Despite this audio clip hitting a good handful of Rogan buzz words, the recording its self is a forgery. It was created using an AI model developed by the company Dessa (blog post with audio clips). They created an algorithm and feed it not elk meat, but the past episodes of Joe’s podcast, in order to replicate his voice and delivery. The script, created by a human, was then given to the model, where it took that text and outputted the Joe Rogan impersonation to an audio file. As far as a computer voice goes, it’s good, not perfect, but really pretty good. Knowing that it’s fake, you can definitely hear parts that sound somewhat off, almost like an audio version of the uncanny valley.
,,A History of AI Mimicry
This is the latest in a long string of computer algorithms that aim to mimic a human’s appearance, voice, or artistic aesthetic. It’s nothing new, in fact you could say this is a subset of the Turing Test, which goes back to 1950. Below is a timeline of some of the more prominent examples of this technology, I’m sure there are many more.
2012
• An algorithm called Mathgen was able to create a fictitious academic paper and have is accepted to the journal of Advances in Pure Mathematics.
My own math paper created by Mathgen
2015
• TensorFlow debuted and one of its early demos was its style transfer capabilities. Enabling a person to take an image from Van Gogh and have that style applied to any other image. Here’s a demo from the TensorFlow blog.
2017
• An AI model from Botnik wrote a new Harry Potter chapter.
• Both President Obama and Trump were mimicked by this Chinese AI. They even have Trump speaking in Mandarin!
• Dadabots used deep neurula networks to create Black Metal music and remake some Beatles songs.
2018
• Deep Fake, a deep learning model, takes videos of one person and transfers their face to someone else in a different video. Dave Lee has a great review clip showing off its capabilities.
,,Capabilities and Rights
So, this is where we are, we can mimic the content of an artist or writer, we can mimic their voice, and with deep fake, we can even mimic their appearance. Wish you could have lived in the 19th century and had Van Gogh do your portrait? No problem. Bummed that your favorite band broke up? No big deal, we’ll generate new music in their style. Feel like a film would have been better with Nicolas Cage cast in the main role? Someone’s already done that for you. AI is giving us the immortality of the artist. No book series has to end, no painting has to be the last.
This of course brings up the question “what do artist have the right to?” If their music was used as the input to create a model, do they have a right to the output. Or can we simply say the machine was inspired by the music the same way a band like Greta Van Fleet is inspired by Led Zeppelin. Is inspiration simply mimicry with a high amount of variation? Time will only bring more discussion on the topic artists and AI, but a more interesting question is how people will apply this technology to their personal lives.
,,Enter Black Mirror
The best way to dive into this topic is by way of Black Mirror. The 2013 episode Be Right Backgave us the story of a young woman, Martha, whose boyfriend is suddenly killed in a car crash. Distraught by the sudden loss, she turns to a service that offers to replicate his personality. It does this by reading in all the digital data that the boyfriend had shared on social media. The service starts as a simple chat bot, then moves to voice, and finally even replicates his appearance. What was Science Fiction in 2013 is now completely capable in 2019.
Wish you could have had more conversations with a close friend? Make a chatbot. Do you have any audio of them? Even better, now that bot can reply in that voice. Wish your father could have been there to walk you down the aisle? Not a problem, just have someone stand in, and now when you go back to watch the wedding video it will be him (at least look like him).
Like any new technology it brings up questions. What are the ethics of using the likeness of someone deceased? Could this be therapeutic or harmful? Would this service be preying on the emotionally wounded; a digital psychic that claims to allow you to commune with the dead? It opens up a lot of questions and the genie is not going back into the bottle. Someone recently told me that technology follows Moore’s Law, but ethics and philosophy do not. It’s going to take time to for us to experiment and discover what the right balance is for letting this technology into our lives. Be Right Backmainly focuses on Martha’s struggle with this very question. In the end she does finds a balance that allows her to preserve her boyfriend’s memory but also not loose site that it is only a digital representation of him. It’s actually one of the happier endings that Black Mirror has. Hopefully we can have our happy ending too.