The Arc Project Update #3 – Recommended Reading

During my research, I have found some super informative and thought provoking articles that have informed The Arc Project. I wanted to share my favorites in this blog post because these podcasts and books and articles have formed my decision making and thought processes as The Arc project takes shape. Below are some discussions between some of the leading thinkers in AI, but many of their topics engage with philosophical issues, scientific questions, and thought experiments that illustrate the state of AI research today.

Lewis Mitchel was on my radar early on. He gave a talk in Australia in 2021 talking about his experiences and observations with writing through AI. He believes there is still a human touch required to make something meaningful in a script and I agree. But I also wanted to challenge it. I was directed to a great podcast by Boost –  this simple and honest portrait from This American Life of a woman who uses AI to write after the loss of a loved one. The intimate back and forth with the AI model that is captured here is not only moving but fun, and reveals the limitations and wonder of working with an AI to express something so human. Loss. But it has great touches of humor as it honestly as it recounts how much personal engineering and re-writing is required to make the AI work for you as a collaborator.

These days, of course, podcasts are a great method to get detailed info on AI. In fact, podcasts lead the way for me as a way method to understand where AI is and how it can be used to tell a story on film. As AI is composed of so many strands, schools of thought and influences, I ensured I kept my information diet varied and allowed influences from many podcasters to help me understand the potential directions of The Arc. One of the best ones came from Lex Fridman and his 150 minute interview with Yann LeCun, the Chief AI scientist at Meta. I found this podcast after digging around and reading this article about self supervised learning. This is LeCun’s world.  LeCun is a tinkerer, he likes to combine mechanical and electronic ‘stuff’, hence his fascination with AI. Here the two cover a deep range of topics connected to the challenges and future of AI and deep learning and debate the very nature of learning. I reveled in LeCun’s razor sharp observations. He ponders how humans evolved not knowing how to drive cars, but can learn to drive a car in about 20 hours of practice. An AI takes many more hours of practice and a lot of data to learn what, in some ways, comes to humans instinctively.

It made me see how far away we are from general AI intelligence. General AI It is not impossible. But let’s not start talking about consciousness just yet. Cats, humans, and almost all thinking creatures seem to have an understanding of basic physics. How else can an animal understand that, for example, it should not approach a cliff too closely or it will fall. An AI would need to fall off that cliff many times in order to learn that it should avoid a cliff. But that same lesson of falling might not transfer when it comes to jumping off of a house or out of a plane. Transferable knowledge represents one of  the many great challenges of AI right now, and how LeCan talks about intersectional data and his relationship to data is enlightening. My favorite part is when the discussion touches on another motivation for learning, one that strikes at the heart of creativity: our particular relationship to our finite selves. LeCan reveals he has had a near-death experience and is therefore ‘fine’ with death. Fridman explains that everything we do as humans is to avoid confronting our eventual death. It is the great human motivator. So how can AI ever be attuned to this or to create something moving if it is so far from one of the foundations of human inspiration? If an AI can be taught to learn, it can be taught to loss.  It is a great conversation. And since Fridman is a computer scientist himself, the sparring is deep and relevant though a bit technical at times.

I learned a lot from Kent Bye, an American philosopher, journalist and immersive reality commentator. I met Kent in 2017 and have followed his Voices of VR podcast ever since. I will be publishing a conversation between Kent and I on this blog in the coming weeks. But in the lead up to our chat Kent asked me to check out some of his interviews, namely The Ultimate Potential of XR & AI: Immersive Storytelling and Experimental Design and this incredibly deep and thought provoking interview that touches on the intersection of AI and Astrology, consciousness and philosophy of mind. I have learned that too much focus on consciousness and AI should be treated carefully since researchers are far away from understanding the phenomenon of consciousness in humans, let alone in a machine. But Kent brings so many perspectives to his arguments including mathematics, Jungian philosophy and game theory, that it becomes a deliciously mind expanding trip to listen to.

In this discussion, Kent breaks down his theories of presence within VR and XR experiences in understandable and relevant detail. It is around the 51 minute mark where he describes his encounters with AI-led creations where the question arises: compared to AI without context, AI that operates within a context of a story world can be much more convincing of its breadth and power. It can make the user feel that AI is more advanced and even autonomous than it actually is. Something that I keep in mind as we prepare our next stages of The Arc Project. Kent explains the importance of language, and how we use language to get into the mind of another person. Language alone engages our suspension of disbelief in storytelling. Kent speaks about how the choices creators must make to communicate ideas to the audience: either with words, or in architecture or music, and this made me think about how we can make our AI script explore other ways of working with symbols in order to create a story world. There are many kinds of AI out there. Some make visual scenes specifically for VR, some use language, some work with music. (In fact, check out this article from February this year announcing Apple’s purchase of an AI music startup that creates music for your workout depending on your heart rate. Or this from DJ Magazine about how AI is penetrating the DJ booth). If I can combine various modes of AI together with language based models, it might make our story richer and help engage our minds to suspend disbelief, even for a moment.

Closer to home, I am a big fan of the Medium blog from André Hedetoft. He is the co-founder of Stjärnstoft Studios in Malmö, and has been working in the AI/storytelling space for several years now. Topics range from gaming to how AI is shaking up the Hollywood system. André is taking on one of the hardest quests in AI storytelling, and that is to crack your own code and make AI models that operate in multiple mediums, whether it’s gaming, tv or film.  André is a passionate storyteller and has made several narrative films in the past, so his transition to AI content creation fascinated me. I have an interview with André coming up, so be sure to get acquainted with his curiosity and passion for AI by dipping into his blog.

While film is my focus for The Arc, I found out the Young Vic in the UK used AI to create a stage play just last year. I am really curious to find out what this play is like and how their creative process was in order to reach the level of a fully fledged play. As I chased up information about this play, a coincidence happened and I was approached by Hazni Demir, who lives here in Helsingborg. He was super interested in The Arc Project and directed me to this project called TheAlter, which also harnesses the power of AI and theater. Check out this epic discussion here with the leads of this project. I will be interviewing some of the members in the coming weeks to understand how they worked with actors to bring their ideas to life, and hopefully add some insight to the next stages of The Arc. Hanzi will be featuring more in this blog over the coming weeks as research stories and articles together to enhance The Arc Project.

Finally, for some speculative sources of information, I have ordered myself a copy of Human Compatible by Stuart Russell. This is a foundational modern text that delves into societal and human relations to AI, addressing major fears around AI’s role in the near future. It covers the major historical turning points and modern developments in AI technology, like the chess winning AI program that beat some of the best chess masters in the world. I hope to read this soon to reflect on some of the concerns creators have when discussing collaboration with AI. I can empathize with the legitimate concerns of screenwriters who question the use of AI as a creative tool. They understand better than most  that societies are built around stories. Whatever affects one, the reverberations are felt in the other.