Impossible interviews
Steve Jobs passed away in 2011, just two years after Joe Rogan started his podcast "The Joe Rogan Experience." So it is chronologically possible for Joe to have interviewed Steve for the show... but no, it never happened. So what is this recording of Joe Rogan interviewing Steve Jobs? Well, it's Fake Joe Rogan interviewing Fake Steve Jobs...
In order to produce this (19 minute) recording a company first took recordings of Joe and Steve and created synthesized voices. Both have a quite distinct sound to their voice and are instantly recognizable in this recording, if a bit choppy at times. Then the company used a large language model trained on respectively Joe's interviews and Job's speeches in order to generate the dialog that these two synthesized voices would then speak during the episode.
The result is amusing while a bit weird at times (at one point Steve is talking about being a fan of Joe's and says "It's nice to sit back in the car and listen to you rant"). But what is perhaps more important is the demonstration of what is possible today -- any voice with sufficient recorded examples can be simulated, anybody with sufficient textual content can have a simulation of their views generated as a conversation.
领英推荐
Ray Kurzweil for example used a large language model with all of the written texts from his late father to construct an "avatar" with whom he could interact . And "The Jessica Simulation" brought a man's dead girlfriend back to life as a chatbot. And Bruce Willis appeared in a mobile phone advertisement - entirely as a generated image (although his agent states that Willis has not agreed to additional appearances despite reports that he is the first celebrity to agree to license his image for deep fakes).
Where we are: Artificial intelligence can already today simulate appearance, voice, and the kinds of things one says given enough source data to train the model. As with any technology this can be used for good (talk again with a grandparent, be entertained by a celebrity) or for evil (putting words into an opponents mouth). This technology will certainly continue to improve and we will soon see so many fakes that it will become impossible to believe the authenticity of any image, voice, or written text. But this will also open new possibilities for entertainment and advertising. As Fake Steve Jobs says "I think that we're right on the edge, technology is a double edged sword. I'm not saying it's bad, its just that it's double edged."
SVP Head of Digital Product and High-Tech Vertical
2 年Thought provoking to say the least! Enjoying the series Ted.
What’s next?
2 年Well that's a bit disconcerting. When I first heard about deep fakes I recall feeling a bit alarmed. As I read your explanation about how it works, though, and consider the fragile political environment we're currently in here in the US, it really is worrying. I wonder if since AI is used to create these, AI could likewise we trained to spot deep fakes as well as validate authenticity of @real whomever it is? (Would that be sort of a reverse Turing est??! :-)
Client Relationship and Business Development
2 年Another great one Ted. I have started looking forward to your daily musings! To the double edged nature of technology you highlight in this article, I am reminded of the web series "Black Mirror" that become available on Netflix a few years back. A must watch in my mind. It explores the "grays" on the edges of this technological change and what it can enable. A sort of social experiment. It is both enigmatic and scary to see what could be.