AI creates fictional interview between Joe Rogan and Steve Jobs
Joe Rogan finally got his chance to interview Steve Jobs. No, Jobs did not come back from the dead and talked to Rogan. Neither did we discover something to communicate with those residing beyond the realm of the living. Instead, the Rogan-Jobs rendezvous is a creation of artificial intelligence. The fictional podcast has been developed by a Dubai-based voice synthesis company.
Why does this story matter?
- Every day, we hear something new about artificial intelligence. Due to these innovations, the gap between reality and AI diminishes with every passing day.
- The Rogan-Jobs podcast is an example of what AI can do with the leftover transcripts of a dead person. It's far from perfection, but once it achieves perfection, where will it end?
- That's a question for another day.
Rogan called Jobs 'a memory from the past'
The first episode on Podcast.ai starts with Joe Rogan inviting his "weird and brilliant friend." He described the guest as "a memory from the past." Then we hear Steve Jobs's voice. The conversation then pans to the Apple co-founder's time at Reed College, his interest in spirituality, and Buddhism. We hear Jobs speaking passionately about reincarnation, Macintosh, and LSD, among other topics.
The Rogan-Jobs podcast is a creation of a text-to-voice generator
We know what you're thinking and you're right. That was fake Rogan and fake Jobs. But if you listen to the podcast, it would be quite difficult to see through it. Podcast.ai is a creation of Play.ht, which sells voice synthesis services. The interview was created by the company's AI-based text-to-voice generator. The company claims that interview's transcript was "generated with fine-tuned language models."
However, Jobs's voice didn't sound natural enough
The interview was not without its flaws. For the most part, Jobs sounded robotic and most of what he said seemed to be taken out of old interviews. It is not surprising considering Jobs's voice is generated from his recording found online. However, despite the flaws, the similarities in the voice created by AI and that of earlier recordings, are still remarkable and eerie.
The AI imitates tones, emotions in a self-supervised manner
Play.ht uses language models based on similar concepts such as DALL-E and OpenAI's GPT. It imitates human tones and emotions in a "self-supervised manner." Syed Hammad, the founder of Play.ht, said Jobs's voice was created "with just a few minutes of his speech from his videos on stage or interview that had a lot of noise, hence the glitches."