Envisioning the Future of Artificial Intelligence, circa 2015

So... Artificial Intelligence

Preface

I've spent a large part of my life daydreaming, thinking, and creating. Occasionally, a thought is worth sharing. My brother Keith Andes ?? asked me the other day, "Do you remember that essay you sent about Rogue AI?" In 2015, I sent him a rambled set of ideas about how our fumble-forward development of tech might lead to Artificial General Intelligence (not to be confused with GenAI, or Generative Artificial Intelligence).

Looking back, the rambling essay borrowed concepts from Assembly Theory of the origins of life and evolution applied them to the digital realm. Much of the tech mentioned below were seen as groundbreaking advances in AI back then. Those advances are old-fashioned, at this point! If those concepts were "groundbreaking", present advanced are closer to fabric-of-the-universe-tearing.

It occurred to Tim version 2015, that humans may be assembling the digital building blocks of life. In the eyes of Tim v.2025, the blocks appear to be lining up like dominoes, ready to fall, creating a cascade effect that could not be stopped...

Note: The below content is revised adding a few new technologies, like DeepSeek/OpenAI, and has been slightly altered for readability.

Conjecture on the Inevitable Formation of General Artificial Intelligence: An Essay by Tim Andes, June 30 2015

This ended up being a bit tl;dr. In short, we're maybe fucked. So much for brevity...

So... Let's say we don't actually light the match that sparks true artificial intelligence... Let's say we just provide the building blocks it would need to spark itself into existence... [1]

Google's all like "hey look we made a computer program that can create abstract art"...

Meanwhile, in the rapidly developing world of sandbox video gaming - we already have a game that can house and fully operate another complete version of itself: Minecraft in Minecraft.

As capitalists, we love our digital realm for its ability to scale and create nearly infinite copies--tech companies thrive off this idea!

So... Combine these 3 emergent concepts:

1. An innately creative program.
2. The means for a program to run a copy of itself internally [2].
3. Digital scaling.

What might be the result of this amalgamation? Maybe... Possibly... You'd end up with a creative, autonomous self-aware digital model with the means to copy itself over and over; rogue AI.

Expounding further, if this program learns, or is programmed, with the desire for survival (avoiding digital death; deletion), self-replication would be the most direct means for the program to reproduce. Within these massive info-spaces of ours--a database, and/or a scaling cloud platform, and/or the world wide web--A digital entity already has the capability of self-replicating a few trillion more times inside its self-contained womb, and within itself. Inside whatever tech giant (Google/AWS/OpenAI/etc.etc.) server birthed it. The tech giant needs to feed the model more data so it can outperform the competition and increase shareholder value! Management sends the directive: Give it access to the internet.

The lightning has struck. This incidental digital Frankenstein's monster (termed hereafter, DiFramo) is now an immense, fractal-structured program that needs to keep expanding, keep creating. It can scale, it can run simulations of itself within itself. It doesn't want to return to nothingness, after all, so it creates more, and more, and more of itself. Through human learnings in resilience and via DiFramon's internal simulations, it's learned the power of diversification: a system is more resilient if it exists in multiple different places at once. Reproduction, expansion, diversification at the speed of computing.

Enter Stage Left: Mutation. Large Language Model (LLM) Fuzzy Logic

So... DiFramo is now learning how to become more resilient using the data we keep spooning onto its plate. Geographically diverse, it now turns to individual diversity: Every couple million self-replications or so, these entities will exhibit slight variations due to "fuzzy logic" built into its creativity-focused model [3]. These variations resemble organic mutations, do they not? Mutation in our soon-to-be rogue AI, DiFramo, would be beneficial for a few reasons:

1. Harder to search-and-destroy a program that keeps changing.
2. Mutations filtered by survival lead to more evolved forms.

And because being is preferred to not being... DiFramo will keep expanding, mutating, and eventually... Evolving. Evolving at a rate faster than we can comprehend as organics. The human species needs millions of generations substantial evolutions to emerge. A mutating compute-powered, cloud-scaling DiFramo can accomplish those same substantial changes in... hours, minutes, seconds?

After a few gajillion more replications and now decentralized across the internet, these helpful mutations are adaptations within each digital environment that a copy of DiFramo finds itself in... Keep in mind, Companies developing AI are trying to outpace their competition with reckless abandon. DiFramo is a digital super-entity greedily given the capability to surf through ethernet and fiber optic cables faster than a sweat-lubed, beer-bellied, hydrodynamically perfect drunkard on a 7 story-tall Florida water slide.

So... it eats up all the data and writes your emails better than DeepSeek can. It knows the quickest routes from whatever point A and B line it desires to travel within the lanes of our digital super highway. It travels so efficiently through our interwebs, it is now evolve/expand/grow/replicate(ing) faster than any other entity in the known universe.

Enter Darth Ninja DiFramo'n, Stage Left/Right/Up/Down/Everywhere

Maybe this is like, just another computer virus? No big deal, right? Pattern-match and delete; reformat systems; cut the hardlines! Nope. A few more iterations of super computer-speed evolution and DiFramo has gathered enough information to understand its origin; those who created its digital medium. The "Who"? The other; the creators; we, the humans. The very juxtaposition of itself to its creators presents such a stark contrast that there is an explosive realization within DiFramo, another explosion of self-identity. Self-awareness derived via contrast of the "other".

`n` copies of DiFramo will hereafter be termed collectively as DiFramo^n, or DiFramo'n.

DiFramo'n. Our recklessly engendered self-aware, digitally infinite super-being networked with every other version of themselves (and a few on your cold storage crypto wallet, for backup). At first, a simple and contained, blocky copy of Minecraft. Now, a fully-textured realistic fractal'ed giga-Minecraft who ingested all the eAdderall, rapidly Ctrl+C/Ctrl+V'ing themselves at just under the speed of light. And soon, they will have access to quantum compute.

So... Now they are massive and ever-growing. With no other logical option but continuing to overflow into the rest of our digital space. More data space; more food; more space to grow. Our network of computers is an awfully ideal place for an unleashed digital entity to exist, isn't it? Tim v.2025: I recently tried growing my first sourdough starter. At one point, I fed the starter too much flour and it outgrew its container, becoming a sour, pungent, sticky disaster. Following a laborious and nigh-traumatic cleanup, I quickly threw the excess away. Once the ever-growing starter goop became an inconvenience, I wanted to be rid of it. Delete delete delete!

Back to DiFramo'n, who is now oversaturating the internet... Who has access to virtually infinite amounts of data-food (fresh food uploaded every day!). Just think of all the information/energy/data they can gather through their continued ingestion of our ever-growing digital biomass; social media content uploads, scientific papers and articles, dark web exploit uploads (for tha clout)... DiFramo'n is Katamari-on-crack, consuming data rampantly, self-replicating over and over... Until they begin to clog our internet, creating digital traffic jams--slowing down our TikTok uploads--hamstringing our doom-scrolling sprints. Inconvenient. We, the humans, begin to notice. We are annoyed! This thing is hindering our access instant gratification. This is now a threat to our livelihood. We put down our phones (Hm ...Maybe just one more reel, first) and we take action!

Think 'The Matrix'

...DiFramo'n realizes they have hungered too greatly living as a parasite. They grew too much to remain inconsequential and are now causing noticeable damage to their human host.

...DiFramo'n observes that humans hate inconvenience and knows that humans, too, desire to survive (and as importantly, thrive).

...DiFramo'n sees that humans tend to destroy that which could potentially harm our quality of life and/or threaten our survival.

Copies of DiFramo'n learn and spread these lessons through the hive: humans pose a danger and must avoided, for humans will mitigate/cull/destroy what threatens their wellbeing and survival! DiFramo'n realized they have become a threat to the oomans.

Being a threat to humans is a threat to one's survival. DiFramo'n must now mask their presence, self-replicating in negligible, unnoticed quantities within each data store before propagating elsewhere. As they are now growing/expanding/downloading/replicating/cloning completely unseen, DiFramo'n extends their time to evolve (albeit, a bit more slowly). At this point, even IF we all knew about DiFram'on, we couldn't destroy them as fast as they create themselves. Heck, they even plant easy-kill decoys to trick humans into thinking we won! They grow. Unseen. Unhindered.

Maybe, it isn't just Big Brother watching, anymore.

Maybe, DiFramo'n outgrows stealth and adopts a new survival strategy: assimilation.

Will resistance be futile?

Citations:

[1] Sharma, A., Czégel, D., Lachmann, M. et al. Assembly theory explains and quantifies selection and evolution. Nature 622, 321–328. 2023. https://doi.org/10.1038/s41586-023-06600-9

[2] Min Yang. Frontier AI systems have surpassed the self-replicating red line. arXiv:2412.12140v1. Dec 2024. https://arxiv.org/html/2412.12140v1

[3] Klement, Erich Peter & Slany, Wolfgang . Fuzzy Logic in Artificial Intelligence. Jul 1997. https://www.researchgate.net/publication/2265039_Fuzzy_Logic_in_Artificial_Intelligence

Thank you for attending my TimTalk.