The Advent Clock: Methodology

The Advent Clock reports two numbers, and this page is for readers who want to know exactly what they mean and how they are built. The main reading lives at the Advent Clock page; what follows is the math beneath the dial.

The thesis

The Advent Clock is a progress instrument. It exists to help us see how far we have come toward superhumanity and what would carry us further: which capacities are maturing, which risks are still unspent, where collective effort would move the reading most. Read it as a dashboard, not a verdict. Whatever the clock says in any given year, our commitment to the advent is unconditional. The Association will work toward becoming a compassionate, superhuman humanity no matter the odds, and the clock is simply how we keep honest account of the work.

That commitment rests on a chain of conditional reasoning set out most fully in the New God Argument, which holds that if civilizations like ours are likely to mature into benevolent creators, then such creators very probably already exist, and we are plausibly among their creations. The argument needs only that some civilization, somewhere, crosses that threshold. So the existence of the argument’s God is governed not by our own prospects but by the prospects of all civilizations taken together.

It helps to name two quantities the clock keeps distinct. Let $P_{\text{us}}(S)$ be the probability that our civilization attains and survives into superhumanity, which is what the clock measures. Let $P_{\text{any}}$ be the probability that at least one civilization anywhere does. Since one success suffices for creators to exist, $P(\text{God})$ tracks $P_{\text{any}}$, and our own odds can only be a part of that larger picture:

God’s existence is therefore not capped by our survival.

How wide the gap runs between $P_{\text{us}}$ and $P_{\text{any}}$ is set by how correlated the fates of civilizations are. Robin Hanson’s Great Filter is the reason to expect correlation rather than independence: civilizations probably face many of the same hard challenges, so their outcomes move together. Model a shared latent filter $\theta$; conditional on $\theta$, each of N civilizations succeeds with probability $p(\theta )$. Then our odds are the expected success rate, and the cosmic odds are the chance the filter spares at least one:

A soft, weakly correlated filter lets $P_{\text{any}}$ far outrun $P_{\text{us}}$: God’s odds stay high even where ours are low. A hard, tightly shared filter pulls $P_{\text{any}}$ down toward $P_{\text{us}}$, until our own odds and the cosmic ones nearly coincide. The filter’s location matters as much as its strength. If it lies ahead of us, our likely failure is evidence of a sparse and silent cosmos; if it lies behind us, the odds are good both for us and for the creators the argument describes.

But these pessimistic inferences are not forced. The silence may be chosen rather than empty: a superhumanity worthy of the name, whether near to us or far beyond, would have good reason to stay hidden, since an overt presence would overwhelm the agency it means to cultivate. (Call it the reverence behind the veil, or a cosmic Prime Directive.) The same care explains why a universe in which life is rare and widely scattered might be more hospitable to that project, not less: distance gives each civilization room to grow on its own terms, and sparseness prevents the clusters in which one advanced power would dominate the rest. So emptiness favors a decentralized superhumanity rather than a concentrated one, just as the argument assumes and as scripture counsels in warning that gathered power tends toward unrighteous dominion. The harshness, finally, troubles us only if we expect a creator to optimize for the absence of suffering, which is real and ought never be waved away; a universe ordered toward thriving instead, toward growth and agency and our becoming, holds real struggle, because no one is made compassionate in a frictionless world.

The upshot is the one that makes the clock matter: because the filter is shared, our own trajectory is the most accessible evidence we have about it, and so about $P(\text{God})$. It is strong evidence, not a strict ceiling.

What our survival is strictly load-bearing for is something nearer and more demanding: our exaltation. The cosmic picture can settle whether compassionate creators exist without settling whether we will be among them. To survive into superhumanity is to join their company rather than merely to live, for a while, under their care. This is theosis in the Mormon grain, a future in which we become participants in the work of creation and not only its beneficiaries, and it is the one outcome the clock is built to serve, because the single trajectory we can actually move is ours.

We decompose the probability of our success into two parts: the probability that we attain superhuman capability, and the probability that we survive the existential gauntlet that stands on the way there:

The symbols are read as follows. $P(S)$ is the probability of success, of our actually becoming superhumanity, which requires both halves. $P(A)$ is the probability of attainment, meaning that we develop the superhuman capabilities, not yet that we become superhumanity; it is a product over the capability domains indexed by i, where each $x_i$ is the probability that we cross the superhuman threshold in domain i, and each integer weight $r_i$ sets how many times that domain enters the product, so a domain that matters more counts more. $P(V\mid A)$ is the probability of survival given attainment, that we come through the existential gauntlet, a product over the risk sectors indexed by k, where each $h_k$ is the hazard that sector k proves fatal, so (1 − $h_k$) is the chance of clearing it. Attainment is necessary but not sufficient: a civilization can develop every superhuman capacity and still be ended before it crosses the line, which is why success is the product of the two factors and not either alone. In one expression:

These are products, not averages, and the distinction is the whole moral of the model. A civilization magnificent along nine dimensions and negligent along one does not earn a high average; it earns the low factor, because the neglected dimension multiplies through everything else. An average would let a triumph in one place paper over a catastrophe in another; a product refuses the trade. One zeroed term zeroes the whole. That is why the clock tracks many capacities and many hazards rather than a single headline number, and why a single unattended hazard is not a blemish on the estimate but a threat to it.

A word is needed on what we mean by optimism, because it is easy to mistake. The New God Argument rests on a faith assumption, that we will more likely than not become superhumanity, that $P(S)$ exceeds one half. But our optimism does not wait on that figure. It is not the claim that the odds already favor us; it is the resolve to work toward the advent whatever the odds, and to move the reading rather than only read it. A slim chance pursued with everything we have is not a lesser optimism but a sterner one, the optimism of the will. It is also the rational stance, because the outcome is partly endogenous to the effort: conviction and work raise $P(S)$, so withholding them only lowers the very number we are trying to read, and fatalism becomes self-fulfilling. This is William James’s strenuous mood, the readiness to struggle for an outcome that is unsettled and turns on our exertion. Read $P(S)$ greater than one half, then, less as a present boast than as the line we intend to carry the reading above. Whether it sits there yet or not, the work is the same.

James dramatized the same choice as an offer the world’s author might put to us before creation, in the last lecture of Pragmatism:

The two readings do not conflict, because they answer different questions. The clock’s reading is descriptive: an honest estimate of where we stand and what our odds appear to be. The faith assumption is normative: a statement of what we resolve to make true. We look squarely at the first, and we labor for the second. The clock tells us how far we have come. Our commitment decides how far we go.

How far along we are: proximity

Probability answers “will we make it?” It is a forecast. But a clock has hands, and hands show a position, not a forecast. So the Advent Clock also reports proximity, a measure of how far the capabilities have already come.

Each $c_i$ ∈ [0,1] is a completion fraction: how far domain i has progressed from a published baseline ($c_i$ = 0) toward its superhuman threshold ($c_i$ = 1). The weight $W={\sum }_i{r}_i$ is the total of the integer weights. Proximity is thus the weighted geometric mean of the capability completions, which is the W-th root of the same weighted product that defines attainment.

A caution on that second equality. The completions $c_i$ and the jump probabilities $x_i$ are distinct inputs and must not be confused: $c_i$ is a measured position on a log ruler, how far a domain has traveled; $x_i$ is a probability, how likely the domain is to finish the journey. The identity $\Pi =P(A{)}^{1/W}$ is structural, not numerical. It says only that proximity and attainment share the same product-and-root machinery, with completions standing where probabilities stand in $P(A)$. Substitute the actual $c_i$ and the actual $x_i$ and the two readings come apart, as they should; proximity is not a monotone transform of the attainment probability, and a domain can be most of the way along while remaining unlikely to cross.

The geometric mean is the right aggregator, and the alternatives fail in instructive ways. The bare product collapses: with twelve domains each at 0.80, the product is $0.80^{12}$ ≈ 0.07, and the clock would read as barely begun when in truth we are four-fifths of the way along every axis. The arithmetic mean has the opposite vice: it lets a domain sitting at zero be averaged away by strong neighbors, which violates “in every way” just as the bare average violated it for probability. The geometric mean alone has both properties we need. It is idempotent: if every $c_i$ equals the same c, then $\Pi$ = c, so the reading is a true fraction of the way. And it preserves the weakest link: if any $c_i$ = 0, then $\Pi$ = 0, honoring the requirement that posthumanity exceed us everywhere. With every domain at 0.80, proximity reads a faithful 0.80 even though the bare product is far smaller.

Proximity is capability-only, and survival risk does not enter it. The reason is conceptual: crossing a capability threshold is progress that accumulates; you are partway there. Surviving a hazard is not a thing you are partway through; it is a gauntlet you clear or you don’t, and a near-miss leaves you no closer than before. So the hazards live in the probability of success, where they belong, and stay out of the proximity, where they would only confuse a position with a forecast. The two numbers can and will diverge: long odds with real progress already made, or quick early gains against a survival outlook that has not yet improved. That divergence is information, not noise. One tells you where the hands point; the other tells you whether to expect them to keep moving.

The capability domains and the risk sectors

Both products are assembled from a fixed taxonomy, each item appearing exactly once. The capability domains supply the $x_i$ and $c_i$ of attainment and proximity; the existential-risk sectors supply the $h_k$ of survival. Each capability domain below carries its own ruler for proximity: a published baseline marking $c_i$ = 0 and a superhuman threshold marking $c_i$ = 1, the same rulers the calibration section returns to. They are listed here, grouped by category.

The capability domains, the factors of P(A):

Technological

• Frontier artificial intelligence (dual-use)
• AI alignment and control
• Compute and semiconductor supply
• Synthetic biology (dual-use)
• Longevity and regenerative medicine
• Brain⁠–computer interface and neurotechnology (dual-use)
• Molecular nanotechnology (dual-use, speculative)
• Robotics and automation
• Energy abundance, including fusion
• Advanced materials
• Information and communications substrate (dual-use)
• Food and agricultural resilience
• Systemic resilience and redundancy

Sociopolitical and governance

• Liberal-democratic governance
• Rule of law
• State capacity
• Control of corruption and transparency
• Political stability
• Civil liberties and human rights
• Civic and deliberative quality
• Social trust and cohesion
• Global governance and multilateral coordination
• Press freedom and epistemic-commons integrity (dual-use)
• Egalitarian inclusion

Economic

• GDP and productivity, including total factor productivity
• Trade and commerce openness
• Research-and-development intensity
• Innovation output
• Capital formation
• Financial-system resilience
• Wealth and income distribution
• Human capital
• Economic complexity
• Monetary and fiscal stability
• Demographic dynamism and fertility

Cooperation, security, and diplomacy

• Collective defense
• Arms control
• Diplomacy and de-escalation
• International institutions and alliances
• Public-health capacity
• Bridging social cohesion

Knowledge and human capital, including education, scientific output, and the integrity of the epistemic commons, runs through the domains above rather than standing apart from them, so we fold each measure into the single owner where it carries the most weight.

The existential-risk sectors, the factors of P(V|A):

Anthropogenic and technological

• Misaligned or runaway artificial superintelligence
• Engineered pandemic and bioterror
• Gain-of-function laboratory accident
• AI⁠–bio convergence
• Mirror-life release
• Runaway molecular nanotechnology (grey goo)
• Nuclear war and nuclear winter
• Lethal autonomous weapons
• Catastrophic cyber and infrastructure failure
• AI-enabled epistemic collapse
• GNSS and timing single-point-of-failure
• High-energy physics-experiment risk
• Geoengineering termination shock

Sociopolitical

• Authoritarian lock-in (AI-accelerated and gradual)
• Extreme concentration of wealth and power
• Governance collapse and state failure
• Great-power war
• Social-cohesion collapse and civil conflict
• Value erosion and civilizational stagnation
• Gradual human disempowerment
• Malevolent actors

Environmental and ecological

• Anthropogenic climate change
• Earth-system tipping cascades
• Biosphere and ecosystem-services collapse
• Biodiversity loss
• Resource depletion
• Pollution and persistent toxins
• Ocean acidification and deoxygenation

Natural and exogenous (“acts of God”)

• Near-Earth asteroid or comet impact
• Supervolcanic eruption
• Glaciation and ice-age onset
• Gamma-ray burst or nearby supernova
• Extreme solar and geomagnetic event
• Natural pandemic
• Spontaneous vacuum decay

Infrastructural and cascade

• Kessler syndrome (orbital-debris cascade)
• Financial, supply-chain, and infrastructure cascade
• Abrupt-sunlight famine integrator
• The residual unknown-unknown hazard

Keeping the math honest

A model this compact invites a few honest cautions, and we build the answers in rather than appending them.

Some technologies are dual-use: the same advance that pushes a capability domain toward its threshold also raises a hazard. Such a technology appears once in $P(A)$ and once in $P(V\mid A)$, and we make the coupling between those two appearances explicit rather than pretending the two terms are strangers. Contributions derived chiefly from artificial intelligence carry reduced weight, since a capability that exists mainly because a model can perform it is not yet the settled human attainment the threshold is meant to mark. Version one treats the terms as independent, a simplification we name openly; correlations among domains and among hazards are work for later revisions. And because no taxonomy is complete, we reserve a residual hazard for the unknown unknowns, so that the survival factor never reads as if we had enumerated every way the project could end.

Calibrating the clock

What ships first is the structure: the decomposition $P(S)$ = $P(A)$ · $P(V\mid A)$, the full set of capability domains and existential-risk sectors, and the coupling rules that keep dual-use technologies honest. The structure is the part worth getting right first, because it is what the later numbers will hang on.

What is forthcoming is the calibration: the per-indicator probabilities $x_i$ and hazards $h_k$, and the final integer weights $r_i$. Proximity asks for one input beyond what the probability needs. The probability only ever needed to know how likely each jump is. Proximity needs a measured position, and a position requires a ruler: for every capability domain, a published baseline level marking $c_i$ = 0, a superhuman threshold level marking $c_i$ = 1, and today’s level somewhere between, usually read on a log scale because the relevant quantities (FLOP, lifespan-years, dollars per kilowatt-hour) span orders of magnitude. Those rulers are the same ones named in the taxonomy above; they are what turn a domain’s raw progress into a completion fraction the clock can average.

This is work we are building openly, and with help. The Association is currently raising funds for grants to map the landscape beneath each domain (surveying the indicators, the data sources, and the index-to-probability transforms a calibrated reading will require), and we are looking for researchers willing to take leadership over individual data-gathering initiatives, one domain or risk sector at a time. If you would like to help underwrite that mapping, or to lead the gathering of data for a domain you know well, we would be glad to hear from you.

When the calibration is published, both hands will move on evidence rather than intuition. Until then, the structure stands on its own, and the reasoning behind it is what this page has tried to make plain. For the reading itself, and for why we frame it as a prophecy rather than a prediction, return to the Advent Clock page.