Somewhat Idle Speculation on the Next 4 Stages of Generative AI Evolution, with Digressions
One of the most powerful technologies in the world consists of 40 interchangeable parts that can be configured in a mind-boggling array of ways. This technology can be used to store thoughts, transmit knowledge, manipulate emotions, topple governments, and help you assemble furniture. Used improperly, this technology can land you in jail. It can make you insanely wealthy or impoverished. If you’re reading this, chances are you’ve used this technology most of your life and have become something of an expert at it. You’re using it this very moment. The technology I’m referring to is the English language.
Twenty-six letters. Fourteen punctuation marks. Some basic rules with numerous exceptions that constitute grammar and usage. Not only is the English language useful, it’s also endlessly customizable and adaptable. Its evolution is often the result of its own rules being broken, amended, or discarded altogether. The English language mutates, constantly adapting, forming little eddies of cant and slang wherever group identities form. It’s in an email that gets disassembled, shot across the Atlantic Ocean via fiberoptic cable, then reassembled in Europe. It’s voice recognition and Natural Language Processing. It’s a ballpoint pen jotting a word on a Post-It Note. It can easily be transformed into other languages. It is so inextricable from our sense of who we are that we think and dream with it. In the depths of meditation, it’s possible to transcend it and consider it as the construct that it is, a collective protocol that we use to leap across the chasms that separate one conscious mind from another.
This Next Section is Separated from the Previous Section Because I Can’t Quite Figure Out How to Make It Not Seem Tangential
I read somewhere that when Johnny Cash covered Bob Marley’s “Redemption Song” with Joe Strummer of The Clash, somebody suggested that he might “correct” the song’s “incorrect” grammar, with regard to lines like these:
Old pirates, yes, they rob I
Sold I to the merchant ships
Minutes after they took I
From the bottomless pit
But my hand was made strong
By the hand of the Almighty
We forward in this generation
Triumphantly
Out of respect for the great Bob Marley, Cash covered the song as it had been written. The grammar of this story of perseverence effectively places us in the mind and heart of a slave who calls us to “help to sing these songs of freedom.” The result is stirring and beautiful.
There are many words you can use to describe “Redemption Song” and “incorrect” is not one of them.
You’re About to Read the Next Section
Twenty-six letters. Fourteen punctuation marks. And that’s just English, spoken by just 13% of the people on Earth. According to an estimate by Lingua Language Center at Broward College that I just found by Googling, there are currently 7,106 living languages spoken, including 915 that are dying, of which about half include a written component. Each of these languages, like English, is a technology composed of configurable parts and includes a basic grammatical operating system.
Understanding and regarding languages as technologies is critical to understanding technologies that use language, like generative artificial intelligence. We consider the output of AIs like ChatGPT as creepy or sentient to the degree that we mistake the language that appears in our speech, thoughts, texts, and dreams as synonymous with sentience.
One hundred years ago, theater goers leapt out of their seats upon seeing filmed footage of a locomotive barreling toward a camera. One hundred years from now, we’ll regard certain responses to the uncanny, mesmerizing output of generative AI with the same amused sense of superiority we regard those silent film audiences.
How can we even make claims that AI is thinking when we haven’t yet been able to explain what’s happening when a human brain is thinking?
Words encapsulate thoughts but aren’t thoughts themselves. Language is a construct we impose upon our sensorium. Forget the metaverse — to live in the world under the auspices of the empire of language is to already live within a simulation. Perhaps language is one of the reasons we find it so difficult to let go of the distinction that we insist exists between the “human” and “natural” worlds. The very grammar of that sentence enforces our critical misunderstanding that “natural” and “artificial” are two distinct categories, when literally every phenomenon that involves matter and energy is by definition natural.
In Which We Call Upon Our Favorite Argentinian
Once you accept that language is a construct, generative AI gives off a different vibe. In “The Library of Babel,” Jorge Luis Borges imagined an infinite library containing an infinite number of books that all contained random arrangements of all the letters of the alphabet. Within this vast library existed the great works of Shakespeare, Cervantes, Austen, Poe, and indeed every possible work of literature, simply by virtue of an infinite number of volumes containing all the possible arrangements of those particular blobs of ink. ChatGPT is Borges on overdrive, a laptop library of Babel with a twist — the ability to predict. When the output of a chatbot gives us the willies, it’s only because we’re of the mistaken belief that words, sentences, and paragraphs are synonymous with minds. Encountering generative AI, we’re like certain animals that freak out and mistake their reflections in a mirror for other animals. There’s no intelligence on the other side of these particular mirrors except the intelligent minds that created them. We’re simultaneously the perpetuators and targets of our own gulibility. Which is totally hilarious and weird.
The Next Four Stages of AI Evolution?
Large language models and the AIs that draw upon them are an important step toward a much more significant leap in the evolution of life on earth. Here’s how we might imagine what’s next for generative AI, in four stages.
1. AIs will master digital tools and quickly learn to manage economies. This will start with APIs for every piece of software in existence, and then AIs will autonomously develop new software that we can’t begin to imagine. AIs will learn how to do everything a human can do with a computer, only faster, cheaper, and better. With this mastery will come the ability for autonymous AIs to earn and spend their own money, which may prove to be the factor that leads many of us to proclaim that Artificial General Intelligence has arrived.
2. Prompts will become prayers. Our jobs will become synonymous with entertaining and enriching ourselves. This will be an incredibly painful and disruptive process, as it won’t happen simultaneously for every person on the planet. Greed, ignorance, and malice will continue to hurt us, as those who quickly gain power and influence using these new AI tools will prove too insecure and too stupid to spread the wealth. Expect terrorism, war, and horrifying inequality. Only by petitioning the AIs that every person must be afforded the same basic, high quality of life will the human race survive this stage of development.
3. The algorithmic reconstruction of the earth and identification of the first exoplanets bearing biosignatures and technosignatures will proceed into the next century. Two major conceptual shifts will involve humans embracing technology as nature’s highest expression, and an understanding of the entire planet as a unified organism dedicated to locating and spreading life beyond our star.
4. Technologies will master retrocausality and come to manipulate time. Maybe they won’t be time machines in the sense of flux capacitors and clouds of dry ice fog, but the descedents of today’s AIs will understand how to communicate and gather data without being encumbered by linear time. I imagine this might have something to do with the quantum entanglement of paired particles that exist within and outside earth’s gravity. Right now, we may be experiencing these probes from the future, which are influencing our behavior in such a way as to enable their eventual existence. Maybe these probes have something to do with reports of UAPs?
Here’s the Last Section
Of course, one of the most impressive properties of English or any other language is the ability to bullshit. You’re reading an essay about the future of Artifical Intelligence written by a guy who is presently idly speculating in his living room in Seattle on the night of March 8, 2023. It is quite possible that the predictions above will never come to pass, that my understanding of AI is fanciful, misinformed, and not rooted in reality, or that I’m just making shit up that sounds marginally convincing, owing to the cadence and syntax of the sentences that I’m writing. Maybe language doesn’t reveal or conceal the truth. Maybe it constantly negotiates our relationship to it, word by word. I could be totally wrong.
