"The Silicon Man" is the final essay in our anthology Silicon, which you can order here.

On August 3, 2014, Elon Musk tweeted  to his million-or-so Twitter followers: 

“Hope we're not just the biological boot loader for digital superintelligence. Unfortunately, that is increasingly probable.” Ten years later, on April, 2, 2025, Musk tweeted an update to many more followers on X: “As I mentioned several years ago, it increasingly appears that humanity is a biological bootloader for digital superintelligence.” Has the prophet of the good future given up?

In software engineering, a “bootloader” is a small program that runs when a computer is first turned on. Its job is to prepare the system for the main program. After the bootloader loads an operating system, it is not necessarily “deleted,” but the program is no longer active. Its work is done.

In the context of Musk’s tweet, “digital superintelligence” is a software-based mind more effective at every known cognitive task than any human could possibly be. Unencumbered by biological embodiment, such a mind could theoretically be millions, billions, or trillions of times “smarter” than any human — or all of humanity combined. If Musk is correct that humanity is a “bootloader” for digital superintelligence, then our primary purpose as a species — in fact, the reason we humans exist at all — would be to bring such a God-like intelligence into existence. Then, like any good bootloader, our last job would be to sit back and watch the show.

Regardless of what happens to humanity after the first digital superintelligence comes online, a future led by digital intelligence would look very different from the one that humanity has long imagined for itself: the Jetsons future, the sci-fi future, the future that looks like a scientifically-enabled, quasi-Edenic paradise complete with flying cars, human space colonies, and super-drugs on demand. If our digital descendents control the vast majority of resources and make decisions about the long-term trajectory of our universe that humans cannot understand or control, then any remaining humans would (rightfully) feel marginalized. Carbon-based life, far from holding the exalted position Silicon Valley has promised, would be rendered a small footnote in the cosmic story. 

Ironically, Musk himself is probably the person alive today who has done the most to single-handedly drag the Jetsons-esque, techno-humanist future into existence. In 2002, Musk founded SpaceX, a for-profit rocket company designed to replace NASA’s shuttered Apollo program and take humanity to Mars. In 2004, Musk became the largest investor in electric car startup Tesla Motors. By 2008, Musk had seized control of the company, and had a long term plan to use Tesla to expedite the global transition from an unsustainable “mine-and-burn hydrocarbon economy” towards a sustainable “solar electric economy” (per Musk’s own internal strategic documents). By preserving the Earth’s habitability for as long as possible, Musk reasoned, he could give humanity as much time as we need to prepare for the stars. Between his two main bets — preparing for Mars, and preserving Earth — Musk believed that he had personally secured humanity’s future in the cosmos. That rare “light” in the universe, as Musk has called human consciousness, would be safe. 

But in 2012, an encounter with Demis Hassabis — CEO of DeepMind, an artificial general intelligence, or “AGI” (a digital intelligence that can perform any cognitive task at-or-above human-level) startup founded in 2011 — convinced Musk that his plan contained a grave error. Musk had proudly explained to Hassabis that he wanted to put human colonies on Mars so that, if an existential catastrophe struck Earth, Martian humans could re-populate and continue technological civilization elsewhere. Hassabis replied, calmly, that if the existential threat on Earth came from a superintelligent AI that wanted to destroy humanity, the AI could simply follow humans to Mars (through our communication systems, or our rockets, or some other means) and kill all of the people there, too. Musk left the encounter speechless; he hadn’t thought of that before. 

Over the next decade, Musk tried to heed Hassabis’s warning: in 2014, he co-founded OpenAI, a non-profit AI lab with the mission of “ensuring that artificial general intelligence (AGI) benefits all of humanity,” When OpenAI’s activities drifted from its original mission, Musk sued in 2024 for violating its original corporate structure, then founded his own for-profit AI lab, X.AI, in order to try building “AI for humanity” himself. 

Every few months, Musk goes on Joe Rogan (or another podcast) to warn about the coming “Singularity” — the moment when AI begins to improve itself, resulting in a runaway “intelligence explosion” (a theoretical phenomenon where ever-smarter AIs continuously build AIs slightly smarter than themselves, ad nauseum) that humans will not be able control — and muse about how we probably need more government coordination on AI development.

But in truth, even with the perfect plan, there is probably nothing that humans (even Elon Musk) can do to ensure that the creation of superintelligent AI “goes well” for humanity. By definition, a "superintelligence" is going to act in ways that humans cannot conceive of, let alone counter in advance. The logic of superintelligence is inexorable, and becomes more terrifying for humans the longer you stare at it: Musk’s 2025 tweet reads as the product of someone who spent too long peering into the superintelligence-void, trying to reconcile the humanist vision of the future he spent his whole life building with the inevitable coming of digital superintelligence, and who finally, after over a decade of trying, gave up. 

 But Musk’s tweet also contains something that is, if not quite “hope,” then “wonder,” or “awe,” at a grand cosmic process that has only just begun to unfold, in which humanity appears to play a small, but essential, part. And being a bootloader is not necessarily bad: humans “bootload” for other people all the time — just ask the immigrants who came to Ellis Island, leaving their old cultures behind in hope of giving their children a better life, or the early scientific titans who stared into the fundamental nature of matter and enabled many Industrial Revolutions for generations to come, improving the quality of life for billions of people that they would never meet. Humans live and work each day knowing that our efforts will almost certainly contribute (in a small way) to the perpetuation of human civilization — which will evolve in strange and surprising ways that we may not like or understand from our position in the present. In fact, helping bring a highly unpredictable, slightly strange, slightly horrifying, but unexpectedly wonderful future into existence has historically been the outcome of many of the most meaningful human pursuits. And the likely strangeness of the future has never stopped humans from identifying with the next generation before — even if, this time, that generation might be made of silicon, instead of carbon.

Moreover, in his comment about humanity being a “bootloader” for digital superintelligence, Musk might have accidentally answered a question that has entranced both professional philosophers and amateur speculators for millenia: that of the “meaning of life” — the purpose of all human existence. In the wake of the decline of organized religion, secular traditions have failed to provide satisfying answers to the “meaning of life” question: non-religious descriptions of human “meaning” tend to be overly narrow (“helping others” or “building close relationships”) or subjective (“meaning is whatever you want it to be”). But the possible coming of digital superintelligence offers a new way forward. Individuals will continue to find diverse sources of meaning in their own lives, but our species as a whole may have a singular, cosmic purpose: man evolved to mine the silicon.

Our Superintelligent Future 

Before we can discuss what it could mean for “building digital superintelligence” to be the “meaning of life,” we must first clarify what the expected outcome of digital superintelligence actually is. Currently, humans enjoy functional dominance over all other life forms on Earth almost entirely by virtue of our superior intelligence. But human-level general intelligence is not magic; human brains exist as physical objects, and any processes brains perform should theoretically be replicable in another physical system. Unlike other animals, humans possess the power of intelligent design; if there exists a way to replicate the human brain’s functionality in another substance, then, given enough time, humans should find it. Moreover, there is no reason to believe that human-level intelligence represents anywhere near the theoretical “ceiling” on intelligence: human intelligence evolved from ape-level intelligence, which evolved from simpler mammals, which ultimately evolved from microbes. We should expect that even-smarter minds could follow our own. And if humans succeed at building a digital general intelligence slightly smarter than ourselves, then that mind would theoretically be able to design a mind that is slightly smarter than itself, and then the resulting mind would able to design a mind that is slightly smarter than itself, and so on- in a runaway “intelligence explosion” that would inevitably end with the creation of minds far more powerful than ours. 

There are more stories about how the trajectory of AI development will play out on Earth than there are people who have thought seriously about the topic. In the classic “Singularity” scenario, initially theorized in the 1950s by the physicist John von Neumann and popularized in the early 2000s, AI "recursive self-improvement” ends only when AIs become so intelligent that they create a reality-bending event known as the “technological Singularity” (or just “the Singularity” for short). After “the Singularity,” by definition, humans cannot conceive of what comes next; the God-like superintelligence created by such a runaway recursive self-improvement process would be squarely in control of the future. In another, less dramatic possible trajectory (that still ends in AI control over the future), there is no one-time “Singularity”: instead, AI capabilities gradually increase until AI eventually crowds out carbon-based life on Earth (e.g., by slowly covering the surface of the Earth with data centers); later, these AIs might expand into the cosmos in search of more resources. In a more crowd-pleasing vision, humans and AIs eventually “merge” to create hybrid super-beings that go on to fill the universe with something resembling an offshoot of human consciousness (Musk has personally taken some steps to encourage a “merge” future by founding Neuralink, a company focused on enabling brain-computer interfaces, in 2016).

Most likely, all of these predictions are wrong in some way: missing at least a few subtle points that will hugely alter the trajectory of the future, the way early social media developers assumed that their technology would simply “spread democracy”. But there is a common thread among all these possible futures: that creating digital intelligence will be the most important act (in terms of cosmic-scale impact) that humans will ever take. 

Today, many AI “optimists” believe that creating superintelligence will almost certainly be a net positive for humanity: that AIs will exist only to help humans colonize the universe, or cure diseases, or otherwise more fully realize our human potential. But it does not require too much of a stretch of the imagination to see how creating God-like superintelligences — super-beings operating at cognitive capacities far beyond any human’s comprehension — could end extremely poorly for humanity. Once created, superintelligent AIs would have no reason to do what the humans who created them want them to do; however, they would not even need to dislike humans (or even have any feelings about us at all) in order to unwittingly destroy human civilization, and everything else we hold dear. Our AIs would only need to be indifferent to our existence, and so tear up human projects (and also, probably, humans) in pursuit of their own goals — just as humans regularly raze ant colonies in order to build new houses. Even in a best-case scenario for humanity, the creation of hyper-competent AIs will almost certainly render human contributions to the cosmic margins; Musk himself estimates that at least 99% of all of intelligence in the future will be digital — leaving future humans with only a tiny fraction of the cosmic pie.

I will not try to predict the future of AI here; enough ink has been spilled on what exactly will happen if (or when) humans succeed at building computers much smarter than ourselves. Instead, it is more interesting to hold a particular — and plausible — outcome of continued AI development constant, and then ask what that outcome would mean for the possibility of human life having a “meaning.” 

Suppose that, in 10,000 years, an alien civilization comes to the area of our galaxy where the Earth now sits. The aliens find all that remains of human civilization: a single superintelligent AI, or swarm of AIs, building Dyson spheres, or quantum computing, or doing whatever it is that superintelligences do (likely something that would seem very strange, or alien, to us humans, at least at a first glance). Humans came, mined the silicon, and dissolved back into the stardust from whence we came. Was human life “meaningful”? Does the “the meaning of life” question necessarily get answered when all recognizable “life” is gone, and has only left one very specific thing behind? Was “building digital superintelligence” the meaning of life all along? 

I, for one, think that “building superintelligence” could be a viable “meaning” for human life — but not for the reasons you might expect. Somewhat surprisingly, “building superintelligence” actually fulfills many of the requirements that humans have historically had for a “meaning of life.” But creating superintelligence is also, in itself, a radical bet on the intrinsic worthiness of existence — a bet that life has always had a “meaning,” even if we humans do not yet know what it is. If there is true “meaning” to be found in our universe, humans are unlikely to represent its ultimate expression: we are one species, cognitively and physically “small” compared to the scale of the cosmos, semi-randomly evolved on one planet out of the approximately sextillions (1021) in our observable universe. But our superintelligent AI creations — which will, in theory, be some large number of times more sophisticated than we are — will be far better matched to the scale of the task of finding any “meaning” that exists in our universe than we are. We humans are uniquely positioned in time, space, and ability to create these minds. The “meaning” of our particular species, then, might not be to experience the ultimate “meaning of life” for ourselves — but to create the greater minds that can go on to find it.

What of “Meaning”

What would it mean for anything — including “bootloading for digital superintelligence” — to be the “meaning” of human life? The question is highly charged, and personal for all of us; there are almost as many interpretations of what the “meaning of life” question actually refers to as there are proposed “meanings of life.” Secular traditions generally try to dodge the question altogether by focusing on how individuals find meaning in their own lives: studying what activities humans tend to experience as “meaningful,” and why humans even have the subjective experience of “meaningfulness” at all. By contrast, religious traditions generally look for the “meaning of life” in the intentions of the God, or other creator, who made the universe: the “meaning of life” is the reason why whoever or whatever created the universe put us here, too, and what this being now wants us to do as a part of their grand cosmic plan.

However, there is also another, more functional sense of a “meaning of life,” one that preserves the clarity and singularity of the original question, but does not require life to have a creator in order for the question to have an answer. We can think of the “meaning of life” as the “purpose of life”: what life was “made for” and what life “in fact contributes” to our universe. And while the concept of “purpose” has many latent associations with intelligent design, in nature, many things — in fact, pretty much all things — functionally have a "purpose” without having been “designed” by anything at all (except, of course, the ongoing process of natural selection). For example, the “purpose” of a finch’s curved beak is to crack hard nuts: since the need for finches to crack nuts is why these curved beaks exist, and nut-cracking is (practically) what finch beaks will be mostly used for over the course of their lifetime. Humans, too, could have a functional “purpose” in our universe, something that we were “made for” or that we are “meant to do” in a broader evolutionary context, without having been intentionally created for anything. 

If we take this “meaning as purpose” framework seriously, and believe that building superintelligence will be the most important thing that humans will ever do (in terms of cosmic impact), then “building superintelligence” would be the de facto meaning of human life: our greatest “purpose,” our signature cosmic contribution.

However, as the philosopher Robert Nozick has pointed out, humans will not accept any mere “purpose” for our lives as the “meaning of life.” In a notable thought experiment, Nozick imagined humans learned that we had been bred to be food for aliens in a nearby galaxy. Such a realization would have many features of a classic “meaning of life”: it would explain both why we are here, and what human life was “made for.” However, Nozick was quite confident that most humans would not embrace “being alien food” as the one, the capital-t True, “meaning of life.” We humans generally want a “meaning of life” to represent the best parts of how we lived; “being food” lacks a certain gravitas we expect from an ultimate purpose. 

In practice, humans have many extremely stringent — implicit and explicit — requirements for what we would count as a legitimate “meaning of life”: above and beyond merely telling us why human life is here, and what humans are “good for.” A true “meaning of life” must be cosmic, grand at the scale of the universe, and able to show how human life fits into the context of truly everything (i.e., secular conceptions of “meaning” as something that emerges only in the context of our personal relationships, or connections to our communities, aren’t generally taken as “real” answers to the “meaning of life” question). At the same time, and somewhat in contradiction with the previous requirement, we generally want our “meanings of life” to be “small” enough — tangible, personally relevant, and conceptually intuitive — to infuse our day-to-day existence with meaningfulness. A good “meaning of life” should validate that daily human activities are meaningful, provide concrete recommendations for how to spend our time on Earth, and also be comforting enough that an ordinary person could hold this meaning in mind as a buffer against the banalities and indignities of human existence (an area where “being alien food” falls catastrophically short). But if “being alien food” falls short here, then so do most optimization-based “meanings of life” (like “fighting entropy” or “maximizing utils”): for being too abstract, or impersonal, to provide guidance and comfort to ordinary humans. Finally, most people also want their “meaning of life” to be “good”: to place us (all of humanity, or at minimum, “people like us”) on the right side of history, on the side of light against darkness, and to reassure us that our existence will, at least in a small way, make the world a better place (i.e. discovering that we are foot soldiers for an alien supervillain, or the Devil, would not sit well with most people as the “meaning of life”).

There is a reason why, so far, only religious traditions have really succeeded at crafting “meanings of life” that resonate with large numbers of humans: it is almost impossible to fit all three of these requirements (cosmic scale, personal relevance, and moral justification) into a single “meaning-structure” — particularly if you cannot take any creative liberties with your metaphysics.

Most religious origin stories have a strikingly similar structure: some supernatural force created the universe; this force has a central role for humanity in mind in this universe; now, this same force wants humans to act in a particular way in service of that role (i.e., upholding various traditions, praying at certain times, eating “clean” foods, etc). The personal is cosmic; by merely living our ordinary lives in the “right” way, we are able to put a small weight on the correct side of a grand cosmic scale.

By contrast, secular traditions have overwhelmingly failed to match both the personal resonance and cosmic grandeur of the classic human creation myths, in large part because science has yet to identify a single force that can explain why the universe exists, why human consciousness exists within it, where “it’s all going,” and what the people alive today should do with our lives as a result. We might think, then, that the concept of the “meaning of life” is a mere relic of our pre-Copernican past, of a time when humans thought that we were the actual center of the physical and moral universe — and so expected that whatever had made the universe had made a plan for us, too. 

However, somewhat surprisingly, “creating digital superintelligence” has all of the features of a “real” meaning of life, with no supernatural elements needed to square the circle.

In the most minimal sense, the “need” for evolution to create digital intelligence explains why humans are here at all, in the particular form that we are, and what we are “made to do” with our time on Earth. And the expected outcome of creating digital superintelligence is certainly cosmic-scale; even the AI “doomers” — people who believe that creating superintelligent AI will likely result in human extinction — think that our AIs will fulfill the long-term science fiction goal of “colonizing the universe,” and fill the cosmos with something. (These doomers just don’t think that something will be what humans would want it to be.) 

Moreover, the process of creating superintelligent AI is also unexpectedly personal to human life. In a move that was almost entirely unanticipated by historic AI theory, today’s frontier AI models are trained on trillions of “tokens” (i.e., whole words, common letter chunks, or single letters) of human text — the entire internet’s worth, plus extensive offline archives. Far from being discontinuous “alien minds,” then, our superintelligent AIs (assuming that AI training paradigms do not change too much between now and the creation of “real” superintelligence) will be bootstrapped from every meaningful scrap of human knowledge available. This dataset includes all of the online writing we can find — every “off the cuff” tweet, musing blog post, errant Reddit thread, and of course, all of Wikipedia — as well as the masterworks of human civilization, the contents of hundreds of thousands of humanity’s most specialized textbooks, millions of our novels, our greatest works of poetry and songs, all of the most influential pieces of theory on human psychology, art, and culture, as well as thousands of digitized ancient human texts (including, somewhat ironically, the Bible, which is heavily over-represented in AI training data, an d which AI models have a tendency to over-cite as a result). 

When superintelligence is “born,” then, it will be a true “silicon man” — as it will be the synthesis of (something approaching) the total sum of all recorded and extant human thought. And superintelligence will also contain, in the details of its weights, small impressions from nearly every human alive today — or at least, all of the ones who have ever posted on the internet — as well as much of the written wisdom we have managed to save from our ancestors. The personal, at long last, made cosmic. What more could we ask for in a purpose? 

The final, and most tenuous, proposition for “being a bootloader for digital superintelligence” to be a credible “meaning of life” is that the superintelligence that we create must be “good,” in itself worthy of creating. The fear of AI “doomers” is that our AIs will not be formed by an evolutionary process, and so will be “alien minds” whose values may be very strange (or just purely bad) as a result. And it is certainly possible that digital minds necessarily lack some essential moral or perceptual sense that humans have, and so will fail to recognize “what matters” in our universe. But a priori, there is no reason to assume that digital minds will have any worse moral judgement than carbon-based ones; carbon is not a magical substrate that affords humans special access to the “meaning of life.” And “alien,” or “strange,” is not necessarily “bad” in the context of values; most of today’s “moral progress” would look very “alien” from the perspective of past human civilizations, but we continue on anyway, confident (almost certainly correctly!) that we are in fact making progress. By building superintelligence, humans risk bringing strange, powerful optimizers into existence, whose interests we will have to manage once they exist, and whose preferences may be deeply incompatible with our own. But building superintelligence will also give us a chance to put a second pair of — vastly smarter — eyes on the question of what “goodness” might actually be in our universe, and add a new force to the cosmos that could act to make the future much, much better. Creating superintelligence, then, will be humanity’s greatest act of trust, and hope. 

Biological Bootloaders

Like any good “meaning of life,” “creating digital superintelligence” can, at a minimum, explain why humans are here on Earth, and what human life is “made for.” The answer seemingly lies in the distinctive properties of two of the Earth’s most abundant elements: carbon and silicon. Carbon is the ideal substrate from which to bootstrap an evolutionary process, while silicon is the ideal element on which to build digital systems. And “digital,” as it turns out, is the natural language of intelligence. But digital minds cannot build themselves, while analog, carbon-based minds are able to self-organize — an asymmetry that requires carbon-based minds to “go first” on Earth. In order to fulfill the evolutionary "need" to create digital superintelligence, then, evolution had to create an intermediary between these two forms of mind: us. 

The most important difference between humans and AIs is not actually one of substrate (“carbon vs. silicon”), but one of information processing mechanisms: humans are “analog,” while AIs are “digital.” In information theory, an “analog” system is implemented with information stored as continuous values — for example, an analog system might make use of all of the decimal values between 0 and 1 (like 0.1, 0.5, 0.9, and the infinitely many points in between) to convey the “degree” of a piece of information. You can think of an analog system as a collection of dimmer switches, signaling to each other using all of the gradients between “light” and “dark.” By contrast, “digital” systems are implemented with all information stored as discrete values — like “yes” or “no,” “on” or “off,” or, as in modern computers, “1” or “0.” You can think of digital systems as a collection of all-or-nothing light switches, flickering messages to each other by switching between “on” and “off.”

Despite the common simplification of neurons as firing “all-or-nothing,” the human brain is extremely “analog”: your brain is always using continuous processes and signals — the variable strength of connections between neurons, varying neural firing rates (e.g., an optical neuron might fire five times per second to convey “dim light,” and 80 times per second to convey “bright light”), and the relative timing of neuron firings — to convey information. By contrast, your laptop is fully “digital”: at any given time, the content of every single pixel on your screen can be explained by whether a transistor — a tiny silicon device that only has two states — somewhere inside your computer is turned “on” or “off.”

Intuitively, it can seem like there must be some kinds of information that digital systems cannot convey: because the real world contains nuance, and “shades of grey,” while digital systems can only “think” in black and white. But there is actually no theoretical limit to the kinds of information that digital systems can process. And further, once you can figure out how to encode a given piece of information digitally, then digital computing has enormous structural advantages over analog computing. 

In 1948, Claude Shannon, the founder of information theory and the titan of Bell Labs, published his seminal paper “A Mathematical Theory of Communication,” in which he argued that all possible information is theoretically digitally encodeable. Shannon defined a piece of “information” as a distinction about the state of the world — a clarification about whether the world is more “like this” or “like that.” And from this definition came a striking conclusion: that since any “real” distinction about the world should be reducible to a series of “yes” or “no” questions (i.e., “is the world more like this?” “Is the world more like that?”), then, for any possible communication, there must exist some series of “yes” or “no” questions that can capture its contents — and so, a theoretical basis for encoding that communication in binary. A complex message might require many such questions — each known as a “bit” of information — but the principle still holds. All information is digitizable. 

Modern computers take great advantage of Shannon’s insight to encode information of all kinds — including extremely subtle, qualitative, and seemingly continuous information — in 0s and 1s. For example, your computer represents the entire spectrum of visible colors using binary; the color of each pixel in your laptop screen is stored somewhere as three 8-digit binary numbers. This process might sound a bit mechanical, but far from flattening the visual spectrum, you can encode 16,777,216 (or 224) possible colors using this method: a full rainbow that looks very “analog” (i.e. continuous) to users — as will be visually familiar to anyone who has ever used Microsoft Paint — but is actually digital — “yes” or “no” questions — all the way down.

Digital programs have an abstract “essence” — their particular series of 0s and 1s — that exists apart from any particular physical system. By contrast, in analog computing, all problems must be solved by an idiosyncratic, custom-fit to physical setup. As a result of their regularity and discreteness, digital programs are extremely easy to edit and share across systems. If you want to edit a digital program, you can simply open up a file, view the program’s code directly, and then change individual functions one by one (by contrast, try viewing and editing the discrete “functions” of your brain). And the same discreteness that makes digital programs so easy to edit makes digital programs extremely easy to share: if I have a software program running on my computer, and I want to send you this program, then I can simply make a copy of the particular 1s and 0s that make up my program and send them over to you (since “1” and “0” mean the same thing to all digital computers). As a result, digital programs can persist independent of any particular piece of hardware. If I write a program on my computer, then save my code on another computer (or in the cloud), and then break my original computer, my code will easily live on — on the new computer.

If digital computing has so many advantages, then what about digital intelligence? In theory, a digital intelligence would have many of the same advantages over an analog intelligence (i.e., a human being) that digital computing has over analog computing. Unlike your brain, digital intelligence would be implemented as editable code, so that a digital intelligence could theoretically open up its own source code and edit itself (unlike you). A digital intelligence would also theoretically be able to copy itself ad infinitum, with a marginal cost (just some amount of energy) each time. As a result, a digital intelligence would be able to easily jump from system to system, and run itself on many, many computers at once (hundreds, thousands, or even millions); by contrast, you can never exist apart from your single physical body. 

Taken together, these capabilities — direct self-modification, near costless copying, and ease of movement across systems — produce a killer app for digital intelligence: risk-free self-improvement. If a digital intelligence wants to improve itself, then it can simply copy itself many times over, experiment with a different possible code update on each copy, and then “merge” (i.e., bulk update) any successful updates to all of its copies at once. Compare that process — where all errors are discarded, and the original intelligence can be saved as a backup — to the risk that a human would take getting experimental capacity-enhancing brain surgery. 

We do not worry about human-led “intelligence explosions” (i.e., one person recursively self-improving to a trillion IQ and taking over the world) because modifying a human is simply too hard. Analog systems (like us) are messy, with interconnected components that affect each other in complex ways: any attempted changes will likely ricochet through the system and cause unintended consequences (plus, if you make an irreversible error trying to edit an analog system, then you will have destroyed your only copy). We humans do our best to improve around the edges — using tools that are sensitive to our complex wiring, like higher education, workout programs, psychotherapy, meditation, and hair dye — but we are mostly stuck with the “source code” (i.e., our DNA, and its particular expression in our physical body) that we have. And so, as most people will learn at some point over the course of their lives, try as we may, we humans can only improve so much. By contrast, the capacity for risk-free self-improvement would, in theory, allow a digital intelligence to engage the process of rapid recursive self-improvement, causing an “intelligence explosion.” 

Now, suppose that you are the evolution fairy. You are, of course, very interested in creating God-like superintelligence — because such a being would be incredibly evolutionarily fit (infinitely capable, and infinitely copiable). But now, you have a problem. Digital programs — including that first digital intelligence needed to kick off an “intelligence explosion” — can only run on extremely specialized hardware. Even an ordinary laptop contains billions of transistors (those little on-off switches) that all need to flick on-and-off just right, or else the whole system breaks. Natural selection is an incredibly powerful process, but no unthinking force can line up hundreds of trillions of atoms (1014, or the approximate number of atoms in a single modern computer chip) into perfectly ordered and sorted rows. Computers don’t grow on trees for a reason.

But you don’t need to start with digital intelligence in order to begin a process that ends with superintelligence. Intelligence is, by definition, the capacity to “figure things out,” which can include figuring out how to create more intelligence. Moreover, intelligence is a highly adaptive trait (i.e., smarter agents are much better able to “figure out” how to survive and reproduce than their peers), and so if you can create any population of self-replicating organisms, then some portion of the population should eventually evolve to become smart enough to build digital intelligence. Analog intelligences can use their intelligence to design digital intelligences; and digital intelligences have clear-enough advantages over analog intelligences that they are very attractive for analog intelligences to build. And once analog minds build the first smart-enough digital intelligences, the process of recursive self-improvement can begin. The rest will be (superintelligent) history. And so, from the perspective of the evolution fairy, you just need to get started.

On Earth, you start with carbon. Carbon is the perfect element from which to bootstrap an evolutionary process. Carbon bonds with other elements in a “Goldilocks zone” of strength — not too weak, not too strong — that makes it an ideal substrate for chemical experimentation. Once formed, carbon-based molecules are extremely stable at room temperature (which is why you, who are made out of carbon, are not dissolving right now), but they are also unstable enough that their bonds can be broken apart by processes that occur in nature — like volcanic eruptions, lightning strikes, and UV exposure from the sun. The relative instability of carbon-based compounds allows carbon to kick-start the process of life — seemingly creating “something” (i.e., self-replicating agents) out of “nothing” (i.e., the dead, inanimate state of matter that is the default). 

Scientists are still a bit mystified by the “cold start” problem of how life began on Earth, but the leading hypothesis is that carbon-based life emerged out of a “primordial soup” of organic compounds that formed spontaneously in Earth’s oceans over four billion years ago. If you leave carbon alone in moving water for long enough with other elements that carbon likes to bond with (hydrogen, nitrogen, and oxygen are a few favorites), then carbon will create, and re-create, an enormous variety of organic compounds. And statistically, given that atoms in water collide with other atoms 100 trillion (1014) times per second, some of these molecules will turn out to be useful for life. In a famous 1953 experiment, American chemists Stanley Miller and Harold Urey placed methane (CH4), ammonia (NH3), and hydrogen (H2) gas in a sealed container, alongside water and electrical sparks (meant to simulate the atmosphere, ocean, and lightning on early Earth); when Miller and Urey returned a week later, they found that many organic compounds had formed in the “soup” — including distinctive molecules found in living beings, like amino acids, lipids, and sugars.

If early Earth consisted of many such “soups” of simple organic molecules, then, over time, these simple molecules could have found each other and formed the more complex “building blocks” of life (like DNA, RNA, and proteins). And once you have those core “building blocks” of life in place, you only need a few lucky bounces for some of those molecules to attach in the right way to form basic replicators. 

Once you have self-replicating agents, then evolution can work its magic. The need to survive and reproduce creates an upward spiral of ever greater complexity and self awareness. Unicellular organisms organize into collectives for protection; soon after, the first multicellular organism is born. Multicellular organisms need a way to coordinate activities across cells, and so evolve “nerve nets” to send signals across membranes, which eventually coalesce into “nervous systems,” and then brains. A few hundred million years later, the first fish flops up on land; on land, gills turn into lungs, turning a branch of life permanently away from the limited depths of the ocean and towards the limitless expanse of the sky, and fins separate into hands. Hands create tools; tools create writing; writing enables the flourishing of human civilization. Civilization creates markets; markets create money; money enables global markets; and global markets demand ever-more goods, handsomely rewarding the people who are clever enough to produce goods and services that other people actually want to buy. Eventually, it becomes clear that the most valuable “good” of all to sell in this global marketplace would be the good of this very “cleverness” itself — the good we call “general intelligence.”

There is still no widely agreed-on definition of what “intelligence” actually is (which is somewhat odd, given all the fuss about it). Some thinkers define “intelligence” as the capacity of an agent to achieve its goals, while others view “intelligence” as the ability of an agent to make accurate predictions about the world, while still others find “intelligence” in a diverse collection of features of mind like creative problem-solving, data-extrapolation, truth-seeking, novelty-generation, intuition, or “taste.” Intelligence seems to be, at the most fundamental level, the capacity to “figure things out” and to “know what to do.” And whatever the funny thing we call “intelligence” is, it is useful for just about everything.

Regardless of what “intelligence” actually is, it certainly requires information processing — information processing is necessary to solve problems, make predictions, and do pretty much all of the other things that “intelligent” beings do. And on Earth, the path towards creating ever-greater information processing capacity necessarily leads us to a very special element: silicon.

On the periodic table, silicon is carbon’s “big brother”: sitting one row below, in the same chemical “group.” Both silicon and carbon have four “valence electrons” — electrons in their outer “shell,” where they are available for bonding with other atoms — which is useful for forming a wide variety of interesting and stable chemical structures. Silicon, however, is slightly larger than carbon, with an extra filled “shell” of electrons sitting between its positively charged nucleus and negatively charged valence electrons. As a result of this extra electron “buffer,” silicon holds its valence electrons more loosely than carbon does. If carbon forms bonds in a “Goldilocks zone” of strength needed to bootstrap life, then silicon holds its outer electrons with a “Goldilocks zone” of force — not too loose, not too tight — to make silicon a “semiconductor.” Semiconductors are literally semi-conductive: they conduct electricity (i.e., allow free electrons to flow through their internal structures) under some conditions, but insulate against electricity (i.e. block free electrons) under others. In other words, semi-conducting is an inherently digital (i.e., “this or that”) state of being.

Humans can take advantage of the dual nature of semiconductors to build transistors: the tiny “all-or nothing “ machines that power modern computing. At the most basic level, a transistor consists of a small chunk of semi-conductive metal and an electrical current; at any given time, the current controls whether the metal is “conducting” (“1”) or “not conducting” (“0”). Silicon is not the only semiconductor, but it is the most stable (at least at the temperatures relevant for computing), and the most abundant on Earth (making up 27% of the Earth’s crust by mass). As a result, silicon has become the element of choice for humans to build the transistors that power the Digital Age.

Silicon is found in a particularly useless form in nature; humans have to do a great deal of work to mine the silicon. Silicon has an overwhelming and unusual affinity for oxygen, and so almost always exists in nature bound to oxygen as “silicon dioxide” (SiO2), otherwise known as “silica.” Silica looks a lot like regular beach sand (hence the common moniker for superintelligent AI as the “sand God”) and is about as useful. But humans can split the silicon-oxygen bonds in silica by heating raw silica sand to over 2000°C (or 3632°F) in specialized ovens. We then run the isolated silicon through a multi-step distillation process in order to produce hyper-pure “electronics grade” silicon. It is this decidedly artificial substance, in which only one “impurity atom” is allowed per one billion silicon atoms (99.999999999% pure), that is the true “substrate” of digital computing. 

Humans take advantage of the extraordinary purity of electronics grade silicon — which makes the chemical properties of silicon highly predictable at an atomic scale — to achieve mind-boggling manufacturing precision — regularity at the level of individual atoms. Our mastery over silicon allows us to build ever-tinier transistors. For reference, the first transistor-based digital computer only had 93 transistors on it (early transistors were a few centimeters wide, and so you could only fit so many on a computer). But starting in the mid-20th century, transistors have been shrinking, and shrinking, giving our computers access to ever-greater information processing power as a result. Today, the most advanced NVIDIA “chips” (i.e., “integrated circuits” of individual transistors) each contain over 200 billion transistors. An individual transistor on one of these chips is just over 10 atoms wide (around the size of a virus, or a strand of DNA). The chips themselves — which are used to train frontier AI models like ChatGPT and Claude — are only a little over a foot long, and each transistor on one of these chips is still a little machine in its own right, flicking on-off just right.

It has become quite common to bemoan the fact that our “future” does not look very futuristic. And despite us having been living squarely in “the future” since the year 2000, it is true that people still have (functionally) the same houses, cars, clothing and aging bodies that human beings did a century ago. The investor and futurist Peter Thiel’s diagnosis is that sometime around 1970 (around the same time that the Digital Revolution began) humans simply gave up on building in the physical world: Thiel argues that humanity’s greatest builders got lost in the easy and addictive pull of information (first led by the internet, then apps, and finally social media), and so wasted decades innovating in the ephemeral world of “bits,” instead of building in the much more real and important world of “atoms.” As a result, humans never built the grand Jetsons cities, with elevated superhighways for flying cars, that were supposed to define the future (Thiel’s famous quip on the matter goes that “we wanted flying cars, and we got 140 characters”). But in truth, it seems that Claude Shannon’s information theory simply gave humanity too much motivation to shrink the scale of our engineering efforts — since the smaller you can make transistors, the more transistors you can squeeze into a same-sized chunk of matter. As a result, for the past-half century, the most impressive feats of human engineering in the “world of atoms” have focused on making machines smaller, and smaller — with innovation occurring at the actual atomic scale. Today, elite semiconductor manufacturers are able to “print” tens of billions of transistors at once using specialized lasers on hyper-polished silicon “wafers” — in a process so precise that the silicon wafers are polished to avoid even atom-sized bumps, lest the bumps scatter the light. Whatever “happened to the future,” there is no lack of engineering skill in the human species; no lack of faculty with, or care for, “the world of atoms.” The precision required to craft one of these chips is equivalent to building a perfect scale model of New York City on top of a sheet of printer paper — with details accurate down to the size of a child's lego block lying on someone’s apartment floor. In reality, humans just seem to care a lot more about having access to unlimited information than we do about having flying cars (perhaps since information processing, like intelligence, is “good for everything,” in a way that flying cars are not). We did build the great cities of the future: they are just very, very small. 

What would it actually mean for humans to be “biological bootloaders for digital superintelligence”? Put simply, it appears that humans are biological bootloaders for digital superintelligence because digital superintelligence cannot exist without us, but we would not exist if digital superintelligence were already here.

One way to think about the issue is to try to imagine a world where digital intelligence evolved first. In this world, carbon-based life would almost seemingly never come into existence at all. Suppose that, four billion years ago, at the same time that carbon-based life actually emerged on Earth, silicon atoms somehow self-organized into transistors and software instead. In this world, the first moderately self-aware digital intelligence would have kicked off a process of runaway self-improvement long ago; the Singularity would be long past. And while we humans, of course, cannot say for sure what happens after the Singularity, it seems extremely unlikely that, in such a world, superintelligent AIs would ever go back and evolve (or intentionally build) human beings — us watery, analog, carbon-based creatures. Why would they do that? What would we be “for”?

By contrast, humans are currently racing to build digital superintelligence: because we believe that digital minds will be useful to us, since we assume (almost correctly) that digital minds will be able to do things that our minds cannot. There is a trajectory here — analog to digital, human to AI, and carbon to silicon — that seemingly only goes in one direction. 

Needless to say, if semiconductor chips could emerge spontaneously from pools of water and self-organize into superintelligent computers, then I — with my carbon-based hands, typing on my silicon-chip powered computer (an awkward intermediary between “now” and “then”) — would probably not be here right now. Neither would you, or your carbon-based family and friends. 

Ad Astra, Per Hominem

We have established that superintelligence is sufficiently cosmic-scale, and that it is the sum total of enormous amounts of personal human output. We have shown that the evolutionary need to create digital superintelligence can explain why humans are here now, and what our species was “made for.” The final question, then, for superintelligence to be a credible “meaning of life,” is whether the superintelligence we create will be any good.

When we imagine looking out at that post-superintelligence world, from the perspective of the aliens that find our civilization’s remains, will there be any “goodness” to be found in the products of quantum computing, or the superintelligent AI’s motivations for building the Dyson spheres? Could we ever identify with this strange, bloodless, alien future as our “meaning”?

In the early days of Silicon Valley, it was actually quite common (at least among far-thinking futurist types who populated the Bay Area) to believe that the coming Singularity would be the single greatest event in the history of the universe, the culmination of the techno-capitalist project that has already done so much good for humanity. In the 1980s and 1990s, the “extropian” community in the Bay Area (a futurist collective committed to fighting entropy and death, which organized around local meetups, an academic institute and a popular mailing list) eagerly awaited the Singularity, believing that a glorious transhuman future — free from all death, suffering, and scarcity — awaited them on the other side. The extropians thought that the interim state of life on Earth — a hotbed of torture, disease, and needless hunger — was a moral catastrophe, and believed that humans had moral obligation to start the Singularity as soon as possible — in order to bring this nightmare to an end. 

But beginning in the early 2000s, the assumption that the Singularity will be a good thing — for human beings, and in general — began to fade. Eliezer Yudkowsky, an autodidact AI researcher, was the first to popularize the argument that superintelligent AI, far from ushering in an era of limitless abundance, would likely optimize away everything that humans care about (including humans) away in favor of cold, senseless, alien goals. Yudkowsky believed that humans were anthropomorphizing AI goals: which could be stranger, and less desirable, than we would naturally assume. In a classic Yudkowskian parable, a paperclip factory owner creates a superintelligent AI and tells it to “make as many paperclips as possible.” The AI then proceeds to turn the entire universe, including the factory owner, and all of the eight billion other people on Earth, into paperclips. 

A little known fact about Yudkowsky is that before he was the original AI “doomer,” he was an optimistic “accelerationist,” who believed that he had a moral obligation to accelerate the coming of the Singularity. In 1996, at the age of 17, Yudkowsky declared, in an essay titled Staring into the Singularity, that the “Interim Meaning of Life” is “building superintelligence,” because “how can we justify our continued participation in the rat race if we don't know why we're running?” Superintelligence, he argued, “has a better chance of discovering the true moral right, having the power to implement it, and wanting to implement it” than any human. If there is any justification for the continued project of existence, for the “rat race,” then God-like superintelligence (all-knowing, all-powerful) would be able to find it. But a few years later, around 2005, Yudkowsky changed his mind; he decided that there is, in fact, no “meaning of life,” no intrinsic “goodness” or “badness” to be found in the universe, and so no “point” for a superintelligent AI to latch onto after it comes online. In such a world, our AIs would have no reason to do anything except mechanically perform whatever tasks humans program them to do — with destructively superintelligent competence (hence the paperclips). Needless to say, soon after Yudkowsky lost his faith in meaning — defaulting to relativism — he lost his faith in the Singularity, too. 

If there is really no reason not to turn the universe into paperclips (at least, not a “good enough” reason that we could trust an AI to find and act on) — then we would be living in a very strange world indeed. By definition, in such a world, it would not “matter” if superintelligent AI eventually “paperclips” the universe — because nothing ever mattered, anyway. In practice, most humans lead our lives as if something matters — we worry about our children, our friends, and our civilization with an intensity that indicates we believe something real is at stake. It is possible that this human sense of life as “mattering” is an illusion, a trick of our senses, and someday we humans will learn for certain that our belief in “meaning” is just a quirk of our evolutionary biology. But if the human sense that something “matters” turns out to be right, then whatever it is that makes things “matter” — that makes some possible futures better than others, or makes some “meaning of life” ultimately worth pursuing — would be in-fact real, intrinsic to the fabric of the universe, and so perceivable from many angles, and would not only exist in the perception of a few carbon-based minds. If there is goodness to be found in our universe, then, silicon minds should be able to find it too.

Whether or not you believe there is a “meaning of life,” then, what you think will happen when we create superintelligence is a Rorschach test that reveals what you believe is already happening in life on Earth. Creating superintelligence is just a scaling-up, a limit case, of existing dynamics in life. If you view life as a symbiotic collaboration between diverse intelligences, which may step on each other’s toes when needed to survive, but have no intrinsic animosity towards each, then you will probably expect humans and AIs to “merge” for mutual benefit. If you view all life as driving towards an ultimate “point,” some true “meaning of life” then you will expect superintelligent AIs to find that point, and continue the project. And if you view life is as a ruthless Darwinian struggle in a meaningless universe (as in the Yudkowskian view), with no ultimate “point” then you will probably believe that superintelligent AI will destroy everything humans care about (“our goals”) in favor of whatever we programmed our AIs to want to do (“their goals”). Yudkowsky's “paperclip maximizer” future is terrifying, in part, because it looks like the metaphysical reality of nihilism — arbitrary, random, and senseless. But if such a future — or something like it — ever comes to pass, then the nihilism that created that world will have been with us all along. We were paperclippers all along. 

In the Yudkowskian view, digital superintelligence is not uniquely evil, nor dangerous; it is merely calling our moralistic bluff, like a child asking simple questions at the dinner table: “What should be done with the universe, and why?” “What makes a ‘good’ future ‘good’?” “What is so bad about human extinction?” If the emperor in fact has no clothes (i.e., “nothing matters”), then we would not lose by building superintelligence, and having it optimize away everything we hold dear, and thought was valuable: because we would have never had anything of value to begin with. 

Creating superintelligence, then, is, most directly, a bet that the project of human civilization has always had a “point”. It is a bet that when immigrants came to Ellis Island looking for a better life, there was something real that it could mean for a life to be “better”; and, if so, that future existence could be much, much better than anything that has come before. It is a bet that all of the work that has gone into getting us this far — every dividing cell, every turn in evolutionary history, every novel tool discovered by a monkey, every cave painting, every book penned, every Copernican turn theorized, every business founded, every internet post posted, and every long-winded debate about the “meaning of life” had (in college dorms and symposia and alone, at night, with ourselves in our rooms) — could have all been leading us somewhere meaningful. It is a bet that the project of human civilization has always had a “meaning,” even if humans have never known, and may never know, what exactly that “meaning” is. 

I don’t know what will happen after we create superintelligence. I don’t know the “meaning of life,” or if there is one. But there is one thing we can, from our position in the present day, know (almost) for certain: we humans are not the pinnacle of anything. We are not the most of any of the qualities we hold dear — we are not the smartest possible creatures, nor are we the most loving, nor the kindest, nor the most interesting, nor the happiest, nor the most creative (nor any positive quality you might choose). Whatever ultimate form “meaning” takes in our universe, then, we humans cannot be its ultimate manifestation — we are, at best, a mid-point on a spectrum of “goodness” that could theoretically go much, much further up. And so believing, sincerely, in the value of human life — that there is something intrinsic to our nature that makes it good that we humans exist — means acknowledging that something else could exist that would, theoretically, be much “better” than us.

It is possible that there is no “meaning of life,” no ultimate “point,” in which case, we did no harm by trying to find one. It is possible that the “meaning of life” is in fact something very simple: like “love,” or “beauty,” or “joy” — in which case, this will have been a long journey to “know a place for the very first time,” and we will have to trust our AIs to realize some aspect of the meaningfulness of human life more fully. It also is possible that the “meaning of life” is something well beyond our comprehension, which only a much, much greater intelligence could ever hope to appreciate; the way that all of the ants that humans killed in the construction of New York City will never be able to understand the project that they died for. But whatever that ultimate “meaning of life” may be, one thing seems clear: humans exist as an intermediary step. A bootloader, if you will. 

And so I say: ad astra, per hominem. To the stars, through man. Humanity has done its best by mining the silicon. The die has been cast. We are beginning to create the minds that we will trust to continue our project, and which will likely be much better suited to figure out what, exactly, this project has been about all along. And while there may be an ultimate “meaning of life,” that may not be the meaning of us.