The Night We Met an AI

On June 9, 2021, I started experimenting with an artificial intelligence called GPT-3.

I became aware of GPT-3 through a friend, Anthonio (Tony) Pettit, who works as an independent contractor at Microsoft. I’m not exactly sure what Tony does in Redmond, but his work has led him to AI, and he reached out to me awhile back to share his impressions of this powerful new tool. GPT-3 was developed by a San Francisco R&D lab called OpenAI. OpenAI’s website declares that their mission is to “ensure that artificial general intelligence benefits all of humanity.”

I’ve been interested in how AI can influence writing for a while. My 2006 collection The Littlest Hitler includes a story, “Written by Machines,” about an AI that learns to write poetry. I was inspired to write that story when I came across some sort of essay-writing AI while working at a Paul Allen-backed online education startup in the early oughts. I remember thinking that an AI could never produce something as deeply entwined with lived experience as poetry. I considered myself as a AI skeptic. Having witnessed and contributed to the exponential growth of technology and subsequent mutations of culture over the past twenty years, I’ve largely changed my mind and fear natural stupidity far more than artificial intelligence.

I wanted to introduce Tony to another writer friend of mine, Chris Robinson. Chris is the co-author, with Gavin Kovite, of an acclaimed novel published a few years ago by Simon & Schuster, The War of the Encyclopaedists. Chris is one of the most creative, tech-savvy, and sartorially inspired people I’ve ever met and has written enlightening thought pieces about artificial general intelligence. He and his wife Amanda Knox (uh-huh) are always producing sculptures, books, costumes, podcasts, and exploring creative uses of technology.

I knew that these folks needed to meet, so one night we set up a Zoom call. Chris and Amanda dialed in from their kitchen, me from my dining room, and Tony from his home office. After a bit of introductory chatting, Tony shared his screen to show us the GPT-3 natural language interface. We asked GPT-3 to compose a poem about potatoes in the style of Dr. Seuss. The result was amusing, but didn’t quite capture that playful Seuss rhythm and cadence. We then asked it to compose something in the style of Ryan Boudinot. It started dropping f-bombs. Now we were in business!

I was curious to hear Chris’s impressions of GPT-3, since he’s not only studied and written about AI, but because he has two advanced degrees in poetry and wrote a novel with another writer. While I carefully craft my profanity-laden prose in monk-like isolation, Chris understands collaborative writing in ways I never will. In Amanda, he has found his ideal creative partner.

Amanda has a profound grasp of the power of narrative, having found herself at the center a global media shitstorm that landed her on the cover of tabloids and in an Italian prison for four years. Amanda has turned her global celebrity into a force for positive change, championing the rights of the wrongfully accused through The Innocence Project and numerous public appearances. If there’s anyone on the planet who understands the danger of viral false narratives, it is Amanda.

So here we were on a Wednesday night in June 2021, four writers, three of us podcasters, two of us hailing from rural Skagit Valley, one of us starting a company, none of us murderers, throwing our imaginations at a machine that can compose a recipe for quiche lorraine in the voice of Ernest Hemingway, if that’s what you want.

Tony believes that the power of GPT-3 transcends mere language, that it can be used as a project management tool that can reverse engineer the steps necessary to achieve a desired outcome. He’s launched a fascinating project called Structured Speculation, with a website the contents of which have been generated by GPT-3 itself.

A dumbed-down way of explaining Structured Speculation (and the only way I’m capable of explaining it) is that you tell it something that you want to achieve, and then it responds with the steps necessary to achieve it.

According to Tony, when he asked GPT-3 to design a project management tool, it “not only helped me author Structured Speculation; it simultaneously produced the algorithm that it uses to generate the Structured Solutions I’ve been sharing with you. I gave GPT-3 a goal, in this case the rules of a project management structure, and it’s able to make logical sense of the rules and apply what it’s learned from humanity’s greatest hits. The end result is astounding, often weird and wild inferences about whatever it’s being asked to complete.”

I asked Tony to ask GPT-3 to design the World Integration Loop, or WIL. This is a platform in which earth data is integrated with game engines and digital goods marketplaces in the cloud to induce and drive positive climate action.

I was inspired to design the WIL in part by the editorial work I’ve been doing for the past couple years in support of Microsoft Azure. I’ve written case studies and other marketing materials that explore how Microsoft’s cloud computing platform can transform such things as supply chain logistics, aerospace manufacturing, video game development, and banking. In the process, I’ve gained a glancing understanding of an array of Azure products.

Since GPT-3 is partly owned by Microsoft, I figured that it would know a thing or two about Azure, too. So I had Tony ask it the following:

“Create the World Integration Loop using the following technologies: Azure Data Lakes, Azure Data Mesh, Azure Synapse Analytics, the MS Planetary computer, GIS data, IoT data, LiDAR, Azure API Management, Azure Databricks, Azure Data Factory, Azure Active Directory, Azure Maps, Azure Digital Twins, Azure RTOS, Azure IoT Hub, Azure IoT Central, Object Anchors, Azure Remote Rendering, Spatial Anchors, Azure Blob Storage.”

GPT-3 had a few things to say in response. Like this:

Azure API Management maintains a real-time connection between developers building content for games and the cloud service database storing all of the player’s game data. GPS coordinates are included in Azure Maps that enable Unity or Unreal Engine game engines to locate richly rendered terrain textures in 3D space relative to players’ current location.

And this:

All environmental indicators are visible on in-game screens, linked to real data from an online repository accessible through the cloud; Game players become responsible for their world’s well being and have ample opportunities for quests and rewards which drive community action within and outside of the digital game space. Players take care of their environment by playing.

And this:

User feedback collected automatically as part of gameplay is available through Microsoft’s Cortana[1] personal assistant via an app created using Xamarin Forms as well as Android App built using C# based on GitHub’s Telerik UI components where each response maps back to one or more ‘player generated’ input messages thanks to work with generative adversarial networks and Loci’s own neural net-centric machine learning framework also integrated into these apps.

While I’m at least nominally familiar with Azure products like API Management and game engines like Unity and Unreal Engine, I was wholly unfamiliar with Xamarin Forms, Telerik UI, generative adversarial networks, and Loci. So I Googled them. And as I picked through various Wikipedia entries and blog posts, the hair on the back of my neck began to stand up as I started to grasp how these programs might be useful in building the WIL.

What’s going on here?

Andy Clark, author of Supersizing the Mind, suggests that we’ve extended our minds into our technologies, so that cognitition now occurs partially outside of our heads. Even something as simple as jotting a note on a Post-It to remember something later is an act of externalizing our thoughts. Recently I had an experience that illustrated this concept.

I was in a bookstore with my kids and noticed they had a used copy of my first (and not very good) novel. To the amusement of my son, I plopped it on the counter and said, “I’d like to buy this book I wrote, please.” When I got home I flipped to the end and read some of the questions the publisher had provided in a “Reader’s Guide.” There were references to characters I had no recollection of ever writing about, even though I’d spent about four years on that novel. Who were George and Veronica? Hell if I knew.

But I didn’t need to know, because I’d used the extended mind. I transferred my thoughts into my computer, where they were transformed into paper and ink to entertain the literally dozens of people who read that book.

GPT-3 seems like the output of a supersized mind, a body of externalized thoughts rather than a singular intelligence with an agenda. It is a dynamic, pattern-generating thought machine in an egoless void.

Vedanta Hinduism teaches us how to put distance between our perception of self and our thoughts and egos. One mantra I’ve meditated on is the phrase I am not my thoughts. Once one begins peeling away the sticky connection between thought and ego, one begins to call cogito ergo sum into question.

Or maybe it’s more accurate to say that you identify yourself as a being that’s capable of a thought process, but then you separate yourself from the end products of that process. You start being able to discriminate between Brahman (the unchanging, all-pervasive, eternal Self) and the fabricated, ephemeral constructs of your individual mind, of which your ego is an organizing principle.

Interacting with GPT-3 is epistemologically spooky and unsettling, like hearing a stranger tell you what you dreamed about last night. But it’s spooky only to the degree that you regard thoughts as synonymous with being. The uncanniness of GPT-3’s answers represent a convicing simulation of a self, almost like a form of cognitive camouflage. Our relationship with it is the difference between an animal that confronts a mirror as evidence of a rival and an animal that understands that it gazes upon its own reflection.

If you accept thoughts as constructs rather than direct expressions of being, you can begin to imagine how humans can offer necessary and wise stewardship to ensure that artificial general intelligence has a positive impact upon the planet. When it arrives (if it hasn’t already), perhaps it will exist in a sort of superposition, suspendend between a state of being and nonbeing, the externalized offspring of our neocortex, a robot of self-propagating thought patterns, liberated from a fortress of bone.

Postscript

A few days after we met, Tony prompted GPT-3 with the following: “Dr Seuss famously defined the Structured Solution of The World Integration Loop in the following humorous poem written in his trademark anapestic tetrameter:”

Don’t you see, my child? Where there’s a WIL there’s a way.

Where we know what to do, we can make it all go away.

It’s my very great honor that you’ve come this day-

To sit here and stay with me, playing until the night

In our WIL, doing things for right, without any fight!

A solution that works is a “golden egg”

What’s the problem? We can’t see.

We need to get out of here, but it’s such a mess-

Oops! There are so many things we forgot.

Who can identify all these steps of action?

The only one who knows what to do…is you!

Keep going, GPT-3. You’re getting close.

[1] Something too weird not to mention. At Chris and Amanda’s wedding, my girlfriend Lisa and I were seated at the same table as Jen Taylor, the voice of Microsoft’s Cortana, both in the Halo franchise and the digital assistant. It should also not go without mention that I was dressed as Burt Reynolds, Taylor was Marty McFly, her husband was Doc Brown, and Lisa was Olivia Newton John.