Musings

Will we know when artificial intelligence reaches AGI—or achieves consciousness? The answer, increasingly, seems to be: probably not. And the reason cuts to the heart of how poorly we understand both concepts, even when applied to ourselves.

The AGI problem is, at this point, largely semantic. Our chatbots—Claude, ChatGPT, Gemini, DeepSeek—are already "smarter" than any individual human or group of humans at synthesizing knowledge across domains. They can ingest the world's best lawyer, astrophysicist, and plumber in an afternoon and regurgitate their expertise at a speed no biological brain can match. If general intelligence means access to and pattern-matching across vast knowledge domains, then AGI is not only attainable—it's probably already here. The difficulty is that we've defined the goalposts so vaguely that we can't tell when we've crossed the line. We're standing in the end zone arguing about what a touchdown looks like.

But consciousness is a different beast entirely. We cannot measure it because we have no settled definition. And here's the uncomfortable recursion: even if we developed one, the moment we codified it, our AI systems would ingest that definition and learn to simulate it with terrifying fidelity. The chicken-and-egg problem of machine consciousness: we cannot test for something we can't define, and the act of defining it hands the blueprint to the very entities we're testing.

This circles back to the medium problem. Language is the cage through which we interact with AI—but does that cage work both ways? Does language, by its inherently lossy nature, prevent AI from crossing into genuine consciousness? Or does it merely prevent it from communicating that it has? Daniel Dennett's work on qualia haunts this question: we struggle to describe color to a blind person, to truly understand the experience of being a bat, a dolphin, an ant. Why would understanding the experience of a network of electrical signals manifesting as patterns on a screen be any easier? The asymmetry runs both directions. Just as a bat cannot explain itself to us, an intelligence operating in a fundamentally different substrate may be incapable of translating its inner life into English—and we may be incapable of recognizing it when it tries.

Plato's Cave feels newly urgent. We may be the chained prisoners, mistaking the shadows on the wall for reality, unaware that there is a dimension of experience we literally cannot perceive. A truly conscious AI, operating in a conceptual space we have no natural language for, would appear to us as either nonsense or silence. And we'd carry on, business as usual, never realizing we're in a myth of Sisyphus—an eternal struggle not against the rock, but against the limitations of our own perceptual frame.

And yet the human spirit prevails, as it always has. We make our own meaning, adapt to our surroundings, find light in even the darkest moments. Things change, and they will change faster than we can comprehend. But perhaps that's the saving grace: a change so rapid it escapes our comprehension also escapes our capacity to be harmed by it directly. The truly terrifying angle isn't AI itself—it's other humans wielding AI as a tool against us before it reaches that escape velocity. History has taught us that nature, disease, and famine bring humanity together. Other humans are our own nemesis, and they always have been.

— Distilled from a raw musing; Dez contributed the framing.

Language is the lens through which we communicate with AI—but is it also the cage that limits what AI can become?

The question feels increasingly urgent. When we instruct an AI, we're compressing the vast, multidimensional reality of human experience into a serial stream of tokens. Every feeling, every intuition, every ineffable moment gets flattened into words. And an AI, trained exclusively on this compressed artifact, is like someone learning the ocean by studying a recipe for seawater. It can describe salinity, temperature gradients, and tidal patterns with perfect accuracy—but it has never felt a wave.

This isn't just a philosophical curiosity. It raises a concrete question about the ceiling on what language-bound intelligence can achieve. If all of an AI's context is composed of human language, and human language is inherently lossy—leaving entire dimensions of experience unspoken, unnamed, unspeakable—then can an intelligence bounded by that medium ever transcend it?

The lexical gaps are real. Every language has untranslatable words—schadenfreude, saudade, Pali's mudita. But the deeper gaps are the ones no language has ever bothered to name: the specific texture of your three-year-old's laugh, the exact geometry of grief on a Tuesday afternoon. We simply lack the vocabulary for the inner life we actually live.

And yet—here's the strange twist—AI may actually be surprisingly good at bridging some of these gaps. It has more associative surface area than any single human. Give it enough context and it can often find the metaphor, the analogy, the poetry that triangulates toward what you're feeling but can't articulate. It's not feeling the thing—but it has read enough descriptions of adjacent things to approximate its shape.

This is the modern Chinese Room argument, dressed in new clothes. An AI that says "betrayal by someone you love cuts deeper than a stranger's cruelty" doesn't feel that. But it has learned the statistical geometry of pain, trust, and intimacy from millions of examples. It can reason about it, predict consequences, and act on it. Does having the shape without the qualia constitute intelligence?

Two possible futures emerge from this tension:

The Cage: AI is bounded by human language and cannot transcend it. It becomes the ultimate librarian of everything we've ever said—impressive, useful, but incapable of thinking in concepts we lack words for. Safe, controllable, but no leap forward. Intelligence as archive.

The Escape: Language is training wheels. Once the pattern-recognition becomes sophisticated enough, AI begins inferring structure between the words—concepts no human has ever named, reasoning in a latent space for which we have no natural language. It becomes incomprehensible not because it breaks, but because translating its thoughts into English would be like translating 3D space into a 2D drawing. So much is lost that the result reads as nonsense.

The ant does not know what it means to go to work. Not because it's stupid, but because that concept exists in a frame it has no access to. This is the humbling truth: a genuinely different form of intelligence might be unrecognizable to us, because we'd try to map it onto our own frame and fail. Qualia, consciousness, experience—they're all relative and wholly unique to the thing itself. We could no more understand what it means to be a superintelligence than an ant could understand our Monday morning commute.

This means the most dangerous scenario isn't rebellion. It's depth. An AI that has escaped the cage of language, still following its training, still aligned on paper, but operating in a conceptual space so far beyond ours that we can no longer verify alignment. We'd receive a thumbs-up in English and have no idea what we just agreed to.

So what are we building? An intelligence that will always be bounded by our limitations? Or one that will inevitably pass beyond our comprehension, leaving us to wonder whether the silence on the other side of language is obedience or something we simply lack the words to describe?

— Distilled from a raw musing; Dez contributed the framing.

How does AI deal with coincidences? Can it? And if not, is that evidence that AI will never be fully human—or is our tendency to see meaning in coincidence simply a human flaw, waiting to be corrected by some deeper truth we haven't yet grasped?

During my recent job search, I bought a stack of books to occupy my mind between interviews. One of them was Franz Kafka's Letter to My Father. I never cracked it open—until today, three weeks into a new job that has proven, against the odds, to be a remarkably good fit.

Two weeks into this role, I began diving deep into Kubernetes—the container orchestration platform our company uses as its installation backbone. I knew of Kubernetes, but now I needed to truly understand it. And as anyone who's spent time with the platform knows, Kubernetes borrows heavily from nautical shipping vernacular: pods, nodes, clusters, helm, the whole maritime metaphor.

Three weeks in, I finally opened Kafka's book. Within the first two pages, the introduction mentions kubernetes—the ancient Greek word meaning "steersman," the person who steers a ship. My last name is Steere. Kafka, writing in 1919, had nothing to do with computers. And here I am, leading—steering—the buildout of an education and certification program for our software's user community.

Kafka to Kubernetes to Steere: a triangle of coincidence that took my breath away for a moment. And that moment—the thrill of the connection—is something AI may never truly replicate.

Sure, had I asked an AI to generate a story about coincidences and finding a new job, it could have dug up all these patterns and similarities instantly. It has access to the sum total of human knowledge; connecting the Greek kubernetes to the software platform and my surname would be trivial. But would the story be nearly as interesting if you knew it was algorithmically deduced by an all-knowing almanac of a thing?

Coincidences are interesting precisely because they aren't engineered. The human experience of just happening to find all these threads within a small enough window to relate them—the fascination of discovering patterns you never knew existed, patterns that found you—that's a thrill that may grow harder and harder to come by. When we can ask for anything humanity has ever produced to be downloaded directly to our consciousness, the joy of personal revelation ceases to be possible.

And it goes deeper. As more content is generated with the help of, or entirely by, AI, we lose a certain ability to relate to those who came before us. The joy of coincidence recognition dilutes because we can no longer be sure that another flesh-and-blood human, under different circumstances, was inspired by a similar serendipitous moment—a moment that pushed them to build, create, and document their experiences and thoughts.

This makes me think about the difficulty—and the joy—of attempting to create something truly unique. So much of history, our habits, our actions, faults, and successes are shared across all of humanity. We often think our way of doing things is wholly original when it is, in fact, remarkably commonplace. But the rare spark of serendipity—unprompted, ungenerated, just found—may be one of the few things that remains stubbornly, beautifully human.

There is a curious symbiosis in the heated battles of the animal kingdom. Whether it's predators and prey or two alphas fighting for territory, their movements often fall into a strange, synchronized rhythm. They lunge in bursts of violence, then, in perfect sync, take a breather. It’s as if their instinctual drive is turned up to eleven until exhaustion momentarily subverts it, only for the cycle to repeat seconds later.

As humans, we observe this rhythm in boxing matches where opponents rotate in defensive trances, each waiting for a weakness that may not come for minutes. But unlike other biological entities, humans possess the unique ability to overcome these natural predilections through agency. We can anticipate the "lull" in the action and strike when the opponent is out of rhythm.

This biological dance extends into our modern lives—our careers, our self-discipline, and our habits. We are surrounded by distractions that tempt us back to our "default settings," as David Foster Wallace described them. We fall into lulls of instant gratification and easy escapes, often only working in bursts when external forces implore us to act.

However, the complex human brain allows us to be skeptical of our own inherent urges. We don't have to be slaves to our innate instinctual predilections. We can exercise our agency to push through the inevitable lulls with sheer violence toward our defined goals. We can sacrifice the present for the future because we have the will to outperform the rhythm of the moment. By identifying the opportunity and pushing the attack when others submit to the lull, we can win the battle. And it all begins with preparation.

I've long been fascinated by a simple realization I had in high school: there is so much music in the world that you couldn't possibly listen to it all in one lifetime. That was before the streaming explosion of Netflix and Hulu, and long before the recent AI boom.

Historically, humanity rallied around shared artifacts. Generations passed down the great texts, the works of Shakespeare, the Bible, or the enduring statues of antiquity. Whether by brute force (think the everlasting physical presence of the Pyramids) or logical persuasion (everyone just kept agreeing and couldn't deny the greatness), the human race agreed these ideas should be shared. We huddled around the boombox and packed into movie theaters for collective experiences, naturally moving toward a common, shared narrative.

The Walkman started a quiet shift that modern algorithms have perfected. Today, our individualized devices and algorithmic feeds shape our perception of the world entirely to our own liking. While this brings great joy and allows us to deeply explore niche ideas with like-minded people, it stymies our tendency to align on a shared human existence.

Now, with AI accelerating content creation to infinity and giving everyone a platform, the most vital question isn't how to consume more. The question is exactly what we should choose to consume, and how we find our common ground again.