How AI is Creating a World of Magic

Enchanting everyday objects with spirits

Any sufficiently advanced technology is indistinguishable from magic. 

-Arthur C. Clarke

Part of what gets written here on Embracing Enigmas is about the lessons and understanding of what it takes to apply AI, machine learning, and data science well, particularly in a business context. This can delve into some esoteric topics which on the surface might not seem technical but have great implications for your application. The other part of Embracing Enigmas is taking our knowledge and using that to better understand where the world is and where it's going. We're going to talk about that second part.

I previously wrote about how AI will be embedded into everything. One thing that gets mentioned heavily here on Embracing Enigmas is how many people tend to ignore second order effects when thinking about future impacts of a technology. So, let's talk about one second order effect that is probably in your mind but hasn't readily surfaced.  What happens when everything around us is embedded with AI and can interact bi-directionally with us? In effect, what happens when everything around us has a spirit? AI will help create a world of magic.

Now when we talk about magic, there's a lot of different definitions and understandings. What I'm talking about is your classic fantasy novel wizarding world where individuals can wave their wands, say some magic words, and perform a spell. Think Merlin. Think Harry Potter. Think magical creatures. For this particular discussion, think enchanted objects. The area I want to focus on is spirits.

Enchanting Objects

A spirit is the essence of an entity that reacts to different types of stimuli. This encapsulates all of an entity’s behavior, both actual and perceived. In the world of magic, they imbue objects. They are ghosts inhabiting a location, or they can be talked to through mystical devices. AI is currently providing a way for us to enchant all of the objects around us with a personality and bi-directional communication in natural language. In essence, a spirit.

What does the world look like when you can talk to your toaster, your car, your refrigerator, your house, your candlestick, and your teapot? Perhaps you'd rather talk to historical figures or past loved ones. Alternatively, you might like to have a conversation with a manifestation of your favorite brand. Maybe you want your piano to instruct you while you learn to play. Maybe you're a guitarist that can't sing but you want your guitar to sing for you while the two of you jam out. Perhaps you're more nature focused and want to speak to every plant in your garden or each tree in the forest.

What AI is really enabling us to do is create richer, more interactable worlds around us. We can personalize our experiences as we go through life creating more paths for fulfillment. Customization is increasing in what people can do in their lives and AI provides a level of customization that we haven't been able to experience before. Expect a world of more personality. A world of more flavor.

So how do we get there? Let's start with the components we need:

• An engine to generate language

• The ability to turn language into audio

• A method to create a spirit

• A way to store that spirit on an object

Surprise! We have a way to do all four of those. An engine to generate language? Check (Large Language Models). The ability to turn language into audio? Check (Text to Audio like Whisper). A method to create a spirit? Check (fine-tuning and transfer learning). A way to store that spirit in an object? Check (computer chips, QR codes, and augmented reality). I think the hardest one to comprehend is the method of creating a spirit, so let's walk through where that's currently occurring.

Making Spirits

There are services such as character.ai that allow anyone to interact with a flavored chatbot.  Some sample characters you can chat with are shown in Figure 1.  The character.ai system uses forms of fine-tuning and transfer learning to take a small amount of input from the character creator, combine it with the main large model, and turn it into a full-fledged flavored chatbot.  Thomas Wolf wrote a great tutorial ~3.5 years ago on how this can be done technically.  People have also used GPT-3 (not even the latest technology) to create similar systems for generating text of their favorite icons such as the Paul Graham essay generator.  You can compare that with Paul's original essays here. 

Figure 1.  - Character chatbots from character.ai

Text based generative AI takes care of content but what about a live voice for interaction?  Well, there are services like Veritone that can do things like recreate the voice of an existing person.  A friend recounted a story to me about an executive who found out he had a life-threatening disease that would cause him to lose his voice some time before he died.  At the time of his diagnosis, his daughter was to be married within a year.  This executive then worked with Veritone to gather samples of his voice so that he would still be able to leave spoken messages for his children.  He ended up passing two months before his daughter's wedding but was able to leave the newlyweds with a wedding speech that was "recorded" by Veritone.  His voice already gone, the executive had written the speech, and the models used by Veritone were able to provide a recreation of his speech in his voice at the wedding.  Very emotionally moving to the couple and the attendees. 

At the same time the Veritone story was being recounted to me, another podcaster friend was in the room.  He mentioned that one of the people he interviewed had reached out to to the podcaster because the interviewee was losing his voice and looking for all available audio recordings of himself in order for him to be able to do type-to-speech but with the speech being in his voice.  Clearly, we have arrived at the ability to recreate the spirit of a person.

Spoiler Alert

When we combine the ability to generate language and interact via speech, what do we get? Let's first look to literature in the form of a famous cyberpunk trilogy. In William Gibson's Sprawl Trilogy (Neuromancer, Count Zero, and Mona Lisa Overdrive) there is a minor character called the Finn who owns a shop.  By the third novel, Mona Lisa Overdrive, the Finn has passed away but people are still able to interact with his spirit. 

The Finn's old shop is now ruled by an A.I. construct based on his personality, created by the Finn himself before his death.

There's no magic in this novel, it's strictly in a time of advanced technology. We are closer to this world than people realize. Deep Brain AI recently announced that they could provide the ability to speak to deceased relatives. Just like in Black Mirror. Microsoft apparently got a patent for it.

Implications

Embedding spirits into the items all around us provides an additional layer to the world for people to interact with and enrich their lives. If we have the ability to do things like

• Embed a spirit into any device - toaster, printer, car, stove, smart clothing, etc

• Speak to the dead

• Create deep interactive exhibits at museums to explore topics

• Location based oracles to seek guidance from

• Create a location-based entity that recreates historical events such as at a monument or landmark

• Create virtual assistants and guides for tasks

• Interact with an artist or author about their work while consuming it

• Go one step further and provide actuation to the entity such as lights on and off or turning on sound to allow them to haunt a location

then we get closer to filling the world with magic. A world that is awe-inspiring. However, it also means that social dynamics will change. Will people be giddy with delight or overwhelmed? Will acceptance be split by age? How do you move on when things can never really go away? Do I really want to be annoyed by my toaster and coffee maker every morning? What if the two of them get into a fight? Or will it simply become the norm and the ‘magic’ long since faded?

As I think about what this new world implies, three things stand out. Privacy, control, and experience. With so many objects observing and interacting, people will want a way to maintain their privacy while still benefiting from their use of these systems. At the same time people will want to be able to control these encounters and control what engages with them. Just like you turn off most notifications on your phone, you probably won't want to hear 20 different voices come at you when you walk into a room. Finally, if we can make a world of magic, people are going to want the full experience of living in a world of magic. That means designing magical experiences, exciting interactions, hidden avenues, and intriguing aspects of life. I look forward to seeing us bring more of the world to life that wasn't possible before.