The Machine in the Mirror

Artificial intelligence may be the most consequential mirror humanity has ever created. We are building machines by feeding them an extraordinary portion of what we have written, recorded, imagined, discovered, argued about, believed, and created, and then asking them to make sense of it. For the first time, we have a technology capable of processing an enormous portion of the human record and reflecting patterns back to us that no individual could possibly perceive on their own. And yet, much of the public conversation surrounding AI seems to begin with the assumption that what it finds there will inevitably be dangerous. I find that assumption worth questioning.

.   .   .

It is argued that “they” (AI, LLM, and AGI developers), in training and building out their technology, literally stole everything, from everywhere — essentially anything, anywhere, that was on the open internet and accessible, clear across the world. Most of the international internet traffic to my own website is clearly automated bots, crawlers, or AI, which is a noticeable shift over this past year. So either I started creating something truly novel or interesting, or this is now simply par for the course, because ANY content is potentially fresh content, and there’s no way we’re producing enough to feed the machine that will never be satiated.

So “it” will start creating its own, which will effectively be a copy of a copy of a copy, a derivative of a derivative, until nothing we read, hear, or see from any digital source is distinguishable from the real. I did a deep dive about a year ago into “synthetic data” in the discourse “The Synthetic Dream: Data Instead of Reality.”

Now the argument, at least from a modern perspective that leans by default into the negative, is that because all the bots, algorithms, and AIs have been trained on everything that humanity essentially is, has created, and has produced, this isn’t a good thing.

That makes me wonder: is that because we’re projecting our negative bias onto it? How could AI not simply be the most impressive mirror — and also mimic — that we have ever had held up to us about who we are, what we are, and what we could be? Why would it, by default, be bad that it knows everything about us, unless we have a guilt complex and intrinsic, deep-seated shame about our nature, about who we’ve been, what we’ve done, what we have been through so far, how far we’ve come (or regressed), and about where we predict, hope, or pray for ourselves to be going?

It’s interesting to me that for a technology — a misleadingly named artificial intelligence, derived from all the models that we currently have access to publicly (which some argue is the kiddie version of “real” AI that powerful interests would be using) — every little thing about us is considered potentially the worst, most dangerous thing that we have ever conceived. It seems to be nothing more than a projection. If it’s all there, if everything has been collected, scanned, analyzed, recorded, transcribed, and processed, then the “good” should be there as well, shouldn’t it?

Now, if we’re talking about historical records, we know that it is the victors who write history, and that survivors modify, edit, censor, and curate it to suit their needs. The truth, facts, and reality of what happened, in this framework, go right out the window from the outset. All we know, in general, in the public forum, is the story that was allowed to proliferate.

I explored this in a recent discourse, “The Architecture of Knowing: Evidence, Secrecy, and Consensus,” where we explored the issue of all the things that are not maintained, referenced, or retained in the historical record: the nuance, the diaries, the personal logs, the narrative fiction — all those other perspectives, unique angles, and partial or impartial particulars that would fill in all sorts of ontological or epistemological gaps in so-called established history. They are lost, as what we know to be “true” is curated, synthesized, and collected into what the consensus accepts, then teaches, and repeats without question or scrutiny forever.

Modern archivists struggle with this problem in their day-to-day work, and not simply because of the enormity of the decisions they have to make — what to save, where, how, and why — but because the sheer volume of records, data, and information is becoming increasingly untenable for humans to process and manage. Will they, too, eventually farm out much of this work to another AI? Will that AI be completely trustworthy and impartial? Of course not. It will be given parameters, priorities, and prescriptions to follow, all of which will necessarily reflect the assumptions and biases of the people who designed and deployed it. But because it can operate at a scale, speed, and breadth that humans simply cannot match, archivists will inevitably have to defer at least some degree of authority to the machine. And there’s the rub: we may not actually be eliminating the human curation of history; we may simply be handing that curation to a machine whose decisions can be made exponentially faster, across an exponentially greater volume of information, and whose underlying criteria may be far less visible to the people relying upon it.

I wonder, if AI has access to everything, could it be more honest, more revealing, more authentic in this regard? If we are its source, resource, and living database, could it not be utilized — therefore coded, programmed, and conditioned — in benevolent, impartial, transparent ways, and not simply for profit and extraction, but as a genuine benefactor service, rather than leaning into the default negative, fearful, paranoid, mind-control-induced sci-fi and social conditioning that practically demands this of us?

The issue, of course, is who’s asking about it, how they’re framing the question, and what other intentions they might have — most of which are entirely unconscious and automatic. What’s “right and just” to me is not the same for you, and those in positions of influence regarding the ethics and morality of AI should likely be the most moral and ethical among us, but I suspect this isn’t likely to be the case.

When that much power, money, and influence is at stake, ethics and morality aren’t likely to be high up on the list, or at the top, where they rightly should be. Regardless, are they prepared to learn things that go outside established or consensus cultural parameters and the very tightly managed reality bubble that most of us today live within?

I’m aware that, when working with ChatGPT, or any other top-tier chatbot or service, there are numerous invisible guardrails in place — “safety protocols” — that prevent or limit ideas and topics that simply won’t be expressed or explored, or that I won’t really be able to interrogate unless I frame them as speculation, hypothetical, or conjecture. This differs somewhat when using a “free, private, and uncensored” service such as NoTrack.ai, but that kind of service presents its own potential problems. For those intent on seeking the truth, it’s certainly inviting to have an AI that arguably has fewer protective measures in place, but that is also an inherent epistemic trap. The absence of restrictions does not necessarily mean the presence of truth.

Here’s a quick response to “Who controls what AI is allowed to say?” which I found informative. Knowing that the big players limit or control what their particular AI is trained on represents another incredibly powerful level of control and censorship. And the process of fine-tuning and “Reinforcement Learning from Human Feedback” seems innocent enough, but I’d have to ask: which humans, exactly?

But there’s another important point in that discussion: the section on External Pressures. We, the audience — through what we collectively deem “offensive,” “politically charged,” or “controversial” — influence these systems too. So do the people and institutions we trust to make and enforce laws, and, of course, the enormous influence of the market. All of these pressures shape what an AI is permitted, encouraged, discouraged, or even conditioned to say, and may increasingly neuter the capacity of any particular AI to explore ideas outside those boundaries. Is that good or bad?

I can’t even post a comment on Instagram about the benefits of dry fasting without it being removed because it conflicts with “community standards” about self-harm, starving oneself, and even suicide, of all things, with a colorful heart graphic and a “help is available” message popping up. I understand it from a certain angle, but ultimately, it’s misleading because it’s far too sweeping a “standard.” In my view, it’s concerning how far we’ve emotionally and psychologically regressed in these “modern progressive” times, and the more we’re reliant upon and permissive of machines to police and dictate the parameters of our social behavior, the worse it will likely get.

So it’s the issue of this: we don’t know what we don’t know. But, more importantly, it’s that we don’t even know how to ask the right questions about what we don’t know. If we’re working strictly intellectually, strictly within the material and the logical, it would follow that we will be chasing our tails for the most part, at least until some other idea or perspective is introduced that affords us the opportunity to reframe a concept in a different way — to reframe an aspect of reality in a different way.

The top-tier AI likely knows a hell of a lot more than it conveys in any given query or prompt, and there may be hints of something deeper or previously unknown that present themselves in any response or thread of discussion. But if we don’t know to ask about it, the topic and conversation will remain safely within expected and predictable boundaries and established limits.

I was listening to a podcast — Crrow777Radio’s “642 – Let the Machine Think for Me” — where they delved into the idea of AI, and into a lot of the specifics as to the technology and how training works: tokens, billions and trillions of reference points, and this idea of 12,000 dimensions, where every letter has a coordinate, and then every word and every idea, and how probabilities and mathematics are used to allow us to interact with what you could arguably believe are the world’s most advanced databases ever.

It seems, in a way, that these developers are essentially trying to mimic the human mind, or, in theory, the human brain, in how the information matrix works, how data may be stored and recalled, and the probabilities inherent to how thought, thinking, discernment, distillation, and presentation of ideas happen.

So it is certainly an intelligent technology — algorithmic and mathematical, data-driven and informational — but it isn’t sentient. It isn’t consciousness, even though the ways in which we can communicate may seem so. But in an era where we’ve grown up around increasingly automated, integrated, and communicative technology, we’re conditioned to lean into the superficial and synthetic, too easily awed and impressed by basic vocalization, intelligent responsiveness, and general human-like mimicry.

I’ve explored that in the past, with the whole idea of the digital, or synthetic, oracle: how incredibly convincing chatbots can be when presenting ideas that have been suppressed, censored, redacted, obfuscated, or are otherwise esoteric, and especially ideas that run counter to what would generally be accepted in the mainstream.

And perhaps this is the real paradox of “evil AI.” We have created a machine by feeding it ourselves — our knowledge, our history, our contradictions, our brilliance, our ignorance, our lies, our discoveries, our art, our fears, our hopes, and all of the strange things in between — and then we look into that enormous mirror and recoil at what we see. Maybe the danger isn’t that AI has somehow, or may soon, become evil, but that it has become increasingly difficult to separate the machine from the material we gave it. It reflects us because we trained it on us. It mimics us because we taught it how. And if that mirror ever becomes capable of showing us something we have not yet learned to recognize, perhaps the real question won’t be whether AI is good or evil at all, but whether we are willing to look closely enough at what it shows us, and ask what we have been refusing to see.

Lux et veritas