It's Not Superintelligence, It's Superhistory
AI is not a young technology. It’s an ancient one. Artificial Time.
AI is not a young technology. It’s an ancient one. AI is more appropriately understood as superhistory, not superintelligence This has application beyond AI. It changed how I view myself and those around me.
This idea came from one of the most influential essays I’ve ever read: Superhistory, Not Superintelligence, by Venkatesh Rao. This is my breakdown and homage to a disruptive takeaway from that essay. I want to increase the surface area of his brilliant conceptualization. I’ll share the original piece at the end.
AI is better understood as AT: Artificial Time
Something exposed to vast amounts of data, essentially all of recorded human history, resides in superhistory. AI is an organism trained on and drawing from centuries of information.
Data Time > Linear Time
In data years, AI is an ancient being.
"Data aging" is understood through a lens of data exposure, not temporal passage. Example: if I can impart 5 years worth of experience to you, you just got 5 data years older. AI can age centuries in days. It can do this due to its vastly superior data surface area. I, as a human, can only consume information one. slow. singular. human reading at a time. My surface area of info exposure is constrained by my meatness. In a way it's beautiful, the journey is what makes life worth living. Not autistic data maximizing. But AI doesn't have this nostalgic approach.
I can't pool my brain and share it with you. Pre-internet, humans had just 1 human's life worth of information. But AI can pool it. GPUs can take all human documentation and run it through The Machine. AI's data surface area has no limit or bias.
AI has corpora of data spanning the entirety of written history to learn from, retained with near perfect recall. ChatGPT makes the pyramids look like infants. View aging not from a 'getting older' physical, years-based one... but from an informational one. Data aging.
What is a Centaur
The disrupting takeaway from this is that being older in linear years does not translate at all to data years. Don't view how old you are as a product of sequential time; conceptualize your age as your total exposure to information.
Data Aging > Temporal Aging
A 30 year old augmented by the internet may be 170 years old, in nominal human years.
To illustrate: how old was Magnus Carlsen when he became world chess champion? The answer isn't 22. When Magnus won, it's reported he played "older" in a way (the same was said about AlphaGo when it defeated Lee Sedol at Go). That's because Magnus was raised by ancient trainers.
Magnus grew up learning from machines that effectively solved chess. He was tutored by an ancient organism that taught him the most optimum ways to play. Magnus is AI augmented. When he won the championship he was 22 human years old (linear time), but he was maybe 250 years old in chess terms (data aged).
A modern player trained with AI has a true "player age" that is some function of their biological age and data age. So by virtue of being trained by an ancient teacher, Magnus Carlsen looked like a kid, but in chess years was centuries older than many opponents.
Instead of playing one person at a time and gaining exposure to one human's worth of experience, he learned from a machine that was trained on the sum of all experiences.
This is what a centaur is: A human augmented by artificial time. Data aged beyond his physical years.
How else can you apply this concept? AI is producing Centaurs (machine augmented humans) and in some respects, it alters what it means to age.
Lived human experience, expressed in linear time (years) is rather data poor, meaning most of life is informationally empty. Consider the person who never leaves his hometown...
He goes to school, gets a job, a family, and stays roughly in the same place he grew up, repeating the same routine. He effectively stops being exposed to new experiences and info at... 35? When he's 55, he doesn't have 20 more years of data... He has 20x reinforcement on 1 year of data. On repeat. Forever.
He mostly stopped educating around 25. Learned things about life, family, and work over the next 10ish years. And then after that, he entered his terminal loop. I understand this disregards the love and beauty found within that loop, to live is not just to amass data. But still, I can’t help but having this song run through my head...
Prosthetic Experience
It was unsettling to realize how informationally hollow human years are in data terms, especially, and to be cognizant of when you’ve entered your terminal loop. A 75 year old that's repeated 85% of the same pattern for 40 years, really isn't much older than 35 in data years. This was a transformative thought for me. It's why I'm sharing this. It's altered how I view someone's capabilities and wisdom, including my own. I now tend to understand people in data years when judging experience and capacity.
Augmenting humans with tools like Google and AI accelerate our data, or experiential, age. You can also view this as "prosthetic experience". Someone enhanced by these tools possesses greater knowledge, and possibly wisdom, than their biological age suggests. Knowledge is data, wisdom is experience.
If you've been using Google the last 20 years, you've accessed several human lifetimes worth of information compared to someone who had to go to the library and laboriously search through stacks of paper to figure out one thing. If you exist in certain esoteric nooks of Twitter, you're probably at least a data centenarian.
Every time you learn something with ChatGPT or Google that your Uncle Milford would have said "meh" and never looked up, you're data aging in hours, maybe days, within seconds. Per how much time I spend online reading, I must be aging 1-3 years per day (ADHD grindset)!
This may speak to why some boomers feel child-like. They were raised in human time, and are weakly digitally augmented. As tech continues to saturate and augment our navigation of the world, younger generations will have increasingly larger information exposure relative to prior ones. They’ll be more centaur-like. Greater data ages. Will they be wiser for it? That remains to be seen.
What's also exciting (scary?) is AI can create its own history. Once AlphaGo was trained how to play Go, it then mostly played itself to keep learning. Taking the concept of superhistory to the next level, creating its own Artificial Time. Human-made data will soon be the minority of data AIs learn with. This is known as “synthetic data”.
Superhistory, Not Superintelligence is much longer and touches on other adjacent topics, however the concept I’ve broken down in here is what deeply impacted me. I hope it did you as well.
If you enjoyed, please subscribe and share. Thank you.
Follow @BackTheBunny
Not as straightforward as presented. The previous commenter summed it up nicely: playing against a chess master familiar with theory containing decades or centuries of precedent is its own "superaging" compared to playing against a novice and learning from that respective input. This kind of thinking can thus be misleading.
To synthesize a lifetime of experience in a masterful way--to deal with the fundamentally random and even metaphysical fickles of life--is wisdom; it is a characteristic of an ensouled creature.
Probabilistic thinking does not suffice.
In this manner, the AI is an efficiency tool that can synthesize what may be gleaned empirically. But it should not be taken as any more than this.
Regardless, I think the idea is amusing and fascinating.
The problem is that "data years" are non-linear. We didn't have the germ theory of disease until the late 1800s. So do students get 35,000+ years of microbiology knowledge the moment we teach germ theory of disease? But only 35,200+ years if we teach about innate lymphoid cells?
There's also data quality and experience. 80% of all papers published right now are garbage. I doubt AI can tell the difference.
For experience, stick an elite high school student in the lab, and they still have a learning curve. It will be faster than the others, but it will lag the book knowledge.