My Answer to the Hamming Question

Hint

The Hamming Question: what are the most important problems in your field, and why aren’t you working on them? *

“We are as gods and might as well get good at it.”

— Stewart Brand

The god argument

TLDR. The most important problem I know is that our power to change the world is growing faster than our ability to direct that power well. We can intervene at greater scale, speed, reach, and irreversibility than we can reliably understand consequences, decide what should happen, coordinate action, observe results, and correct mistakes. I use steering below as shorthand for that whole chain.

Capability and steering develop inside the same causal system. Technologies require people and institutions to choose, distribute, and govern them. Those people and institutions are shaped by biology, culture, incentives, inherited tools, and material conditions. Each intervention alters some part of the system that will direct the next one.

Readers who prefer the claims and dependencies without the essay can jump to the structured argument.

External capability is necessary but not self-executing

Stewart Brand opened the first Whole Earth Catalog with “We are as gods and might as well get good at it.” The Catalog offered access to tools: not just machines, but books, maps, and ways for people to educate themselves and shape their environment.

I think the sentence is still unfinished. We are becoming gods externally while remaining fragile, irrational humans internally. We can split atoms, redirect rivers, edit genomes, and alter the climate; we remain vulnerable to distraction, manipulation, fear, status competition, and violence, and often fail to follow effects through the systems we change.

Technology has done things Buddhism cannot. A starving child needs food, not a better attitude; no amount of meditation substitutes for clean water, antibiotics, or a safe birth. Under-five deaths fell by more than half between 2000 and 2024, yet 4.9 million children still died in 2024, most from causes preventable with proven, low-cost interventions and access to quality care.¹

But production is not delivery. An adequate national or international food supply does not guarantee that a household can obtain food; availability, access, use, and stability are separate parts of food security.²

Inventory is not a system. Configuration is causal: enough food in aggregate is not access, because channels and rules determine where it can flow and who can use it.

A communication network can carry truth, propaganda, outrage, or noise. A weapon does not decide when restraint is justified. External progress and inner transformation depend on one another. Becoming gods internally would mean seeing more clearly, wanting better things, governing our impulses, working well with other people, and noticing when we are wrong. Changes to either side alter the conditions under which the other works.

The people and institutions doing the steering are inside the causal system

Knowledge, judgment, and self-command develop inside material and social conditions. Nutrition, toxins, illness, education, tools, culture, and institutions affect what people can notice, understand, want, and do. “Internally” describes the capacities under discussion. The interventions may be biological, environmental, cultural, or institutional.

Multicausal and multiscale only help when the causal paths remain identifiable. Severe iodine deficiency during pregnancy and early childhood impairs brain development. Lead can reach a child through paint, plumbing, contaminated dust and soil, mining, or battery recycling; it can permanently damage the brain, attention, and behavior, and no safe exposure level is known.³ The available interventions sit in product standards, housing, pipe replacement, soil cleanup, mining controls, and battery-recycling practices.

Salt iodization makes the path visible: public standard → producer and quality control → distribution → household consumption during pregnancy → protection against iodine-deficiency brain damage. A public standard and a factory process can protect the development of children not yet born.

Human capacity changes through learning too. Education raises scores on tests used to measure intelligence. Targeted debiasing can improve performance on the tasks it trains, though transfer to real decisions remains uncertain.⁴ People can deliberately improve their performance on particular tasks of attention or judgment; their ability to do so depends partly on development, health, tools, and social support.

Writing externalizes memory; mathematical notation makes otherwise impractical operations routine. Languages, notations, training, and roles extend what people can do with the capacities they have.

Collective execution requires more. For many tasks, capable people still need authority, equipment, supply chains, and procedures. A brilliant clinician without medicine or a delivery system cannot vaccinate a population; a sound doctrine that nobody can understand or execute under pressure is not a safeguard.

Culture is part of the machinery. A checklist without equipment, authority, and practiced use is a piece of paper. A civilization can preserve the description of a practice while losing the ability to perform it. Samo Burja’s “intellectual dark matter” names the tacit and inaccessible knowledge on which institutions depend.

A cultural artifact stays inert unless people can understand and enact it and material systems can support it. Population-scale technologies inherit the same dependency: institutions have to distribute, maintain, and repair them.

Do better people create better institutions, or do better institutions create better people? Both processes can occur at different times. An inherited institution changes exposure, education, incentives, and available roles. People develop and learn inside it. Some later staff, obey, resist, or redesign it. The institution facing the next cohort is partly an output of the previous one.

An attention-funded platform encounters people with existing desires for status, belonging, novelty, and outrage. Its interface and business model reward some expressions of those desires. Users and producers learn what succeeds; the successful behavior supplies the platform with attention and revenue; those results preserve or intensify the rules that selected it. The platform helps train and select which behaviors its users express.

When an institution rewards extraction with money or authority, the people and practices suited to extraction can gain influence, while competitors imitate them to survive. Nobody needs to choose the final result. Moloch is my shorthand for this selection problem. A better system has to persist at the scale its purpose requires and, where wider effects require it, spread without losing what made it better (Make beautiful things scale).

The boundary around a person or organization excludes some of its causes. Daniel Schmachtenberger asks, “I am not that tree, but what would I be without trees?”⁵ The tree is outside my body but not outside the system keeping me alive. Most modern lives likewise depend on farms, grids, institutions, other people, and inherited knowledge. A person or organization can be a meaningful unit without being a self-sufficient causal unit.

Scale can change the verdict. A company can improve its accounts by moving a cost into a river. A state can improve its apparent security while provoking an arms race. Large institutions can move resources far and fast while failing at local administration; local projects can fit a place precisely while lacking the channels to spread. Scale changes what a system can do, not just how much of it exists.

Coordination has a boundary problem. Vitalik Buterin points out that a group can solve a coordination problem among its members while imposing the cost on a larger group: sellers can coordinate as a cartel; a country can coordinate an aggressive war (Coordination, Good and Bad). The design problem is to decide who can coordinate, toward what, under which checks, and whether people outside the group can detect and answer what it does.

The correct boundary depends on the outcome being explained. Child survival may require tracing a path from international production through a state, clinic, household, and body. Climate requires following costs beyond the company or country reporting them. Choosing the boundary determines which effects, dependencies, and people remain outside the analysis.

Practical work begins with a causal path: which causes matter to the outcome; through which links, scales, and times; which link can be changed now; and what feedback would show whether the change propagated.

Greater leverage raises the cost of every weak link

Some technologies let a small number of decisions propagate quickly, affect many people, and cause damage that is difficult to reverse. Each final decision inherits an upstream chain of dependencies. Leverage raises the stakes of failures anywhere in that chain.

In the United States, one President can authorize nuclear use. That decision sits inside warning sensors, attack assessment, authentication, communications, command procedures, doctrine, personal judgment, and—in some scenarios—minutes of uncertainty.⁶ Ask why nuclear use was or was not authorized and the answer may run from a person’s cognition to an interface, an institution, an alliance, and the history that produced all four. An account of that authorization decision that excludes these dependencies would be incomplete.

In some domains, external leverage has grown inside the same multiscale system that produces the judgment, motives, organizations, and feedback governing its use. More power does not wait for the rest of that system to become adequate.

A general problem becomes actionable through a particular causal path

Power without steering appears differently in a food system, a school, a reef, a software company, and a nuclear command system. Each fails through different mechanisms. In each case I need to know what is limiting the desired outcome now, where an intervention could change the system’s behavior, who or what can act there, and what prevents that action.

These are different questions:

• A binding constraint is what currently limits the outcome.
• A leverage point is a place where an intervention could change the system’s behavior.
• An actuator is a person, institution, technology, or mechanism able to make that intervention.
• A blocker is what prevents the actuator from doing so effectively: missing information, authority, capital, engineering, distribution, trust, coordination, or something else.

Donella Meadows’s places to intervene in a system range from parameters and stocks and flows to information channels, rules, self-organization, goals, and assumptions. A visible parameter may have less leverage than the information, rule, or goal producing it.

For a coral reef, warming may be the dominant constraint on survival, beyond the authority of any local program. Where sewage is a significant local stressor, one tractable path is: a healthier, more resilient reef (desired outcome) → untreated discharge (binding constraint) → the treatment process or outfall (leverage point) → the utility and regulator (actuator) → missing monitoring, capital, or authority (blocker) → treatment or enforcement (action) → lower pollution and a measured reef response (observed result). Another reef may be constrained by sediment or the loss of herbivores. The general problem is made of partial paths like these (The answer to all of the problems is all of the solutions).

How I am working on it

I cannot work on this entire problem. The part I keep returning to is the gap between seeing an important constraint and enabling an effective intervention. That means tracing a real causal path closely enough to find the people or mechanisms able to change it, then finding what prevents them from doing so.

Most of my work so far has been on instruments for seeing. At CivLab I built the first complete data model of San Francisco’s government. Tweetscope turns an exported Twitter archive into a labeled map rather than another feed. I worked as a design engineer at Roam Research, where linked ideas could be rearranged without losing their references. These projects taught me how much a representation can reveal. They also exposed its limit: the map has to enter a decision made by someone with the authority and means to act.

I once described the role I want as becoming the eye for the storms that will reshape the landscape in the right places. The eye finds the operative constraint, follows it across scales, and identifies where force could change the result. The storms are the people, institutions, technologies, and efforts able to supply that force through authority, money, engineering, distribution, or organized labor. An eye can remain a spectator. Its value is tested when an actor makes a better decision, an important blocker is removed, or the result shows that the model was wrong.

That test requires access to practitioners and responsibility for something that can succeed or fail. Becoming a dangerous professional is my shorthand for learning an important system well enough to work with its practitioners, identify its operational constraints, and make something useful. Formal descriptions help me enter a system; prolonged contact reveals which descriptions matter, which are false, and which constraints people inside it have learned to work around.

The largest gap in my work is continuity through the rest of the path. I have repeatedly mapped and prototyped; I have less often stayed with one consequential outcome through institutional access, implementation, measurement, and revision. I also default to working alone or asynchronously, while many of the missing capabilities—operating, selling, raising money, organizing people, and sustaining an intervention—are irreducibly collaborative. This is the practical reason to work with people who are stronger at execution, and to enter systems where someone owns the decision and its result.

Open Loops holds the problems and current models; Efforts tracks people and organizations already acting on them. My Quests for 2025 is a dated record of an earlier plan to learn important systems, build shareable artifacts, and collaborate more productively. Those pages can change as the work changes. The durable commitment here is narrower: take one important outcome, trace one causal path, find the blocked actuator, contribute what the path actually lacks, and let the observed result revise the model.

If all I produce is a better catalogue of important things, I have failed.

Structured argument

Note

This is a working map. Its purpose is to expose the claims and their dependencies before the prose makes them sound smoother than they are.

Main claim. Humanity is acquiring greater leverage over the external world without a commensurate ability to understand, govern, and improve the people and systems directing that leverage. Getting good at being gods therefore requires work on external capability and on individual and collective steering capacity as one coupled project.

Here, external leverage means the scale, speed, reach, and irreversibility of the changes we can cause. Steering capacity contains at least four different abilities: understand what is happening and what an intervention is likely to cause; decide what matters; choose an action; and coordinate its execution, observe the result, and correct error. Daniel Schmachtenberger distinguishes the first two as sense-making and meaning-making, which inform choice-making (Building Better Sensemaking). I use the ordinary questions here because each should eventually produce different methods and tests. The language of gods refers to leverage and responsibility; it does not imply omniscience or complete control.

1. Some important constraints are irreducibly material. A starving child needs food; an infection may need an antibiotic; contaminated water needs treatment. Knowledge, character, and spiritual practice can affect how those things are produced and distributed, but cannot substitute for them. This defeats an inner-only account of progress.

2. Material capability is not self-executing. A stock of food does not specify who can reach it. A communication network does not specify whether it will produce understanding, manipulation, or noise. A weapon does not select its target or decide when restraint is justified. Configuration, judgment, incentives, culture, and institutions stand between a capability and its effects. This defeats a technology-only account of progress.

3. The humans making those decisions are not fixed inputs. Nutrition, toxins, disease, sleep, stress, development, education, drugs, tools, training, and social conditions can improve or degrade cognition and behavior. Evidence exists for preventing impairment, improving tested cognitive abilities through education, changing particular failures of judgment through training, and extending performance with cultural tools. These effects differ sharply in size, persistence, transfer, and cost. Distinguishing biological, learned, and tool-dependent changes is useful; treating only one of them as “real” augmentation is not.

4. Culture and institutions are causal mechanisms, not commentary added after biology and technology. They preserve knowledge, train perception, create roles, assign authority, set defaults, reward some actions, and make coordinated performance possible. But they operate only through people able and willing to enact them, and through material systems able to support them.

5. The causal directions are reciprocal and cross scales. Material and social conditions shape development; developed people reproduce or change cultures and institutions; those institutions select and distribute technologies; those technologies alter the next set of material and developmental conditions. A person, organization, or state can be a meaningful local unit without being a self-sufficient causal unit: each depends on flows and systems outside its boundary. “Internal” names Brand’s missing half, but it is not a clean causal boundary.

6. Systems select behavior as well as constraining it. If an arrangement rewards extraction, manipulation, or premature deployment, the people and organizations using those strategies may gain resources and authority. Their success can help reproduce the rules that selected them, while competitors imitate them simply to remain viable. A collectively damaging outcome does not require every participant to want it.

7. Coordination is necessary, but more coordination is not always better. A group can coordinate effectively while imposing costs on everyone outside it. The relevant questions are which people can act together, toward what goal, under which checks, and whether the wider system can detect and answer harmful coordination.

8. The mismatch matters more as leverage grows. A failure of attention, judgment, or coordination has different consequences when it affects a conversation, a city, a nuclear arsenal, a biological system, or infrastructure used by billions. Nuclear weapons do not prove that humans are uniquely irrational, or that deterrence has one simple explanation. They show that technical capability can make a small number of decisions carry catastrophic stakes.

9. The general problem becomes actionable only through particular causal paths. For any attempted intervention, I need to distinguish the desired outcome, the constraint currently limiting it, a leverage point capable of changing the system’s behavior, the people or mechanisms able to act there, what blocks them, and the feedback that would show whether the intervention worked. These are not interchangeable descriptions of the same thing.

10. Better representations are one enabling capacity. No person can hold every relevant scale, dependency, uncertainty, actor, and piece of evidence in unaided attention. Prose, diagrams, simulations, maps, and institutions each preserve some relations and hide others. Better ways of seeing can expose constraints and missing feedback, but they do not by themselves supply goals, authority, coordination, capital, engineering, or action.

The dependencies are important:

• 1 and 2 establish that external capability is necessary but insufficient.
• 3 and 4 establish that steering capacity can be changed, but not by treating either the isolated brain or “culture” as the sole cause.
• 5 turns the two halves into a feedback system rather than a contest between materialism and spirituality.
• 6 and 7 explain how individually intelligible behavior—and successful coordination within the wrong boundary—can reproduce a system that its participants do not collectively want.
• 8 supplies the urgency.
• 9 turns the general problem into a sequence that can be investigated in a real system.
• 10 locates representation inside that sequence without mistaking it for the whole intervention.

The conclusion does not follow from “technology magnifies character.” Technology and character are not independent variables: technologies, bodies, habits, norms, and institutions continually help produce one another.

For a particular problem, the working sequence is: desired outcome → binding constraint → leverage point → actor able to intervene → blocker → action → observed result. The result should update the model of the constraint. My interest is especially in the parts of that sequence that are missing or disconnected.

Footnotes

1. UNICEF, “Progress in reducing child deaths slows as 4.9 million children under five die in 2024”. ↩

2. Food and Agriculture Organization, An Introduction to the Basic Concepts of Food Security. ↩

3. World Health Organization, Guideline: Fortification of Food-Grade Salt with Iodine; World Health Organization, “Lead poisoning and health”. ↩

4. Stuart Ritchie and Elliot Tucker-Drob, “How Much Does Education Improve Intelligence? A Meta-Analysis”; “Systematic review and meta-analysis of educational approaches to reduce cognitive biases among students”. ↩

5. Daniel Schmachtenberger, “A Synthesis: Why Humanity Won’t Survive Unless We Change”. ↩

6. Congressional Research Service, “Authority to Launch Nuclear Forces” and “Nuclear Command, Control, and Communications”. ↩