AI has made answers abundant. What if disciplined inquiry is the new scarcity?

Stephen Wolfram's essay “I Have a Theory Too”: The Challenge and Opportunity of Avocational Science begins with a phenomenon that is easy to dismiss and more interesting if we resist doing so. People outside professional science regularly send him theories about physics, mathematics, cosmology, computation, and other foundational questions. Some arrive with pages of argument, equations, diagrams, or newly generated papers. Increasingly, artificial intelligence participates in producing them.

The obvious reaction is to focus on how many of these theories are wrong. Wolfram's more useful observation is that the phenomenon also reveals a great reservoir of intellectual curiosity. People want to know why the universe works the way it does. They want to investigate patterns, challenge assumptions, construct explanations, and participate in discovery. The problem is not that curiosity has escaped the university.

The problem is what happens after curiosity appears.

Professional science contains an enormous amount of machinery that is almost invisible from the outside. Years of prerequisite knowledge sit behind the ability to formulate a useful question. Literature provides the accumulated record of what has already been tried. Mathematics, computation, experimental design, statistics, peer criticism, laboratory practice, replication, disciplinary conventions, and reputational systems all shape what may eventually count as a scientific contribution.

Someone approaching science from outside an institution tends to encounter the opposite end of that machinery. They see the paper after the failed hypotheses have disappeared. They encounter the theorem after the definitions have stabilized. They see the discovery after thousands of experimental decisions have been compressed into a methods section.

They encounter science at its output boundary rather than its formation boundary.

That distinction matters because there is already evidence that non-professionals can contribute meaningfully when the intellectual environment is designed well.

Projects such as Zooniverse have shown that very large communities will participate in genuine scientific work when professional researchers provide understandable tasks, appropriate context, usable data, and a clear connection between individual effort and a larger research problem. Foldit demonstrated something even more provocative. By turning difficult protein problems into a manipulable computational environment with domain-aware scoring and feedback, researchers created conditions in which people without conventional specialist credentials could develop valuable strategies and contribute to scientifically meaningful results.

Neither example proves that every participant can independently originate a successful scientific theory. They demonstrate something more useful: capability depends partly on the structure of the environment in which intelligence is allowed to work.

The question therefore changes. Instead of asking whether outsiders are capable of science, we can ask which parts of scientific formation can be made more explicit, more accessible, more instrumented, and more learnable without weakening the standards that make science valuable in the first place.

Artificial intelligence makes that question far more consequential.

A capable language model can explain a mathematical concept at several levels of difficulty, translate between technical vocabularies, suggest literature searches, generate executable code, help formulate equations, propose experiments, summarize competing theories, analyze data, and critique an argument. What once required access to several different specialists can increasingly be scaffolded through one computational environment.

But the same machine can also produce something that looks remarkably like scholarship before the underlying reasoning has earned that appearance. It can transform a speculative intuition into fluent academic prose, produce citations around a weak premise, generate mathematical notation that creates the impression of precision, and elaborate an argument until coherence itself begins to feel like evidence.

The problem is therefore not simply that AI sometimes makes mistakes.

Something deeper has changed.

AI is making intellectual expression abundant faster than it is making epistemic discipline abundant.

For much of history, access to information was itself a major barrier to intellectual participation. Books were scarce. Journals were difficult to obtain. Computation was expensive. Advanced instruction belonged to institutions. Expertise was geographically concentrated, and an individual without access to those environments could be excluded before inquiry had really begun.

Many of those barriers are now weakening simultaneously. Knowledge can be retrieved almost instantly. AI can explain difficult material interactively. Open-source software provides extraordinary computational capability. Public datasets expose real scientific objects. Simulation environments can move experimentation into software. Remote communities can collaborate across continents.

The scarce resource is moving downstream.

What remains difficult is learning how to turn an intuition into a precise question, discover whether it is already known, identify the assumptions hidden inside it, formalize the claim, decide what evidence would matter, design a legitimate test, confront counterevidence, distinguish failure from noise, preserve provenance, invite criticism, reproduce the result, and revise the belief when reality refuses to cooperate.

That long transformation is what I mean by scientific formation.

• Intuition
• Question
• Grounding
• Claim
• Formalization
• Test
• Evidence
• Criticism
• Revision
• Contribution

A scientific contribution rarely begins as a scientific contribution. It begins with something much less formal: curiosity, surprise, dissatisfaction, contradiction, or a pattern someone cannot yet explain. The decisive work lies in transforming that initial disturbance through a sequence of increasingly demanding states. The sequence above describes the work that formation must support; it is not yet the formal maturity ladder introduced later.

A question has to become precise enough that another person can understand what is actually being asked. The investigator has to discover the surrounding knowledge rather than treating unfamiliarity with a field as evidence of novelty. Important terms have to acquire definitions. Assumptions have to be exposed. A claim has to become testable or constrainable in a form appropriate to the domain. Evidence must be distinguished from interpretation. Alternatives must be taken seriously. Eventually another person should be able to inspect the reasoning and reproduce the relevant result.

Professional science does not perform this transformation perfectly. Institutions have incentives, politics, blind spots, fashions, and failure modes of their own. But the existence of those imperfections does not make formation unnecessary. It makes the quality of formation even more important.

The question is whether more of that machinery can become accessible outside the institutional apprenticeship through which it has traditionally been learned.

Can we democratize access to scientific rigor without democratizing the definition of truth?

That is the question the rest of the system has to answer. Before designing the answer, however, the proposition has to shed several attractive categories that would solve the wrong problem.

The original idea could easily have become an “Amateur Science Accelerator.” The phrase sounds attractive because it implies energy, access, mentorship, and movement. It also carries the wrong optimization target.

An accelerator assumes that moving the idea forward is success. Science often advances by preventing an idea from moving forward. An investigator who discovers after three hours that a cherished theory was disproved in 1937 may have made more epistemic progress than someone who spends three years turning the same theory into increasingly elaborate prose.

The system therefore cannot optimize for the survival of the idea.

It must optimize for the quality of the inquiry.

What Scientific Formation Infrastructure is not

The boundaries protect the proposition from becoming easier and less useful.

In the AI era, the principal constraint on broad participation in discovery will increasingly move from access to information toward access to epistemic formation.

That proposition follows from a specific mechanism. AI reduces the cost of information access, explanation, coding, formal expression, hypothesis generation, and computational experimentation. As those costs fall, far more people can produce articulated investigations. Scientific-looking expression therefore becomes less informative about underlying quality precisely as the volume of such expression rises.

Expert attention, experimental validity, formal verification, methodological judgment, reproducibility, and trustworthy evidence do not become abundant at the same rate.

The value therefore migrates toward infrastructure capable of separating those things again.

The highest-value role for AI is not to become the scientist at the end of the process. It is to compress the distance between curiosity and disciplined inquiry while leaving the authority of the resulting claim attached to evidence that can be inspected independently of the machine that helped produce it.

With the category now bounded and AI's role made explicit, the design question becomes concrete: what exactly is being formed, what state has it earned, and what must happen before it can responsibly move further?

A conventional paper is optimized to communicate work that has reached a certain maturity. That makes it a poor primary object for formation. A paper tends to compress the developmental history of the inquiry into a clean narrative after the difficult decisions have already been made.

Scientific Formation Infrastructure needs the opposite behavior.

The object should expose its incompleteness.

A candidate with beautiful prose but no serious grounding should look immature. A rough computational notebook containing a reproducible phenomenon may deserve considerably more attention. A claim with strong supporting evidence but no attempt to test alternative explanations should visibly remain incomplete.

The system therefore needs an epistemic object with state, not merely a document with formatting.

What accumulates around the question

Every important element remains explicit rather than disappearing into the polish of the final narrative.

These elements describe what a Knowledge Candidate can contain; they are not a checklist that becomes complete all at once. They accumulate, deepen, and sometimes change as the inquiry matures.

The next question is therefore not simply whether the candidate contains a claim, evidence, assumptions, and provenance. It is what epistemic state those elements have collectively earned. That is the purpose of the Formation Ladder.

There is deliberately no stage called “My Theory Was Proven Right.”

The ladder describes increasing accountability, not increasing attachment.

The Formation Ladder tells us what increasingly mature inquiry looks like. It does not yet tell us how a person moves from one state to the next.

That distinction matters. A curious observation does not become a grounded question merely because the next rung has been named. Grounding requires discovery of prior knowledge and enough understanding to locate the question within it. A grounded question does not become a testable claim automatically; assumptions must be surfaced, terms clarified, and propositions made vulnerable to evidence. Formalization requires another class of work. Investigation requires another. Challenge, reproduction, and contribution each introduce their own demands.

The difficult part of scientific formation therefore lies not only in defining the states. It lies in supporting the transitions between them without allowing the system to manufacture progress that the inquiry has not actually earned.

This is where infrastructure becomes necessary.

The machine should not push every Knowledge Candidate toward the top of the ladder. It should provide the right form of assistance, evidence, friction, and challenge at the point where the candidate currently stands. Sometimes that assistance helps an inquiry advance. Sometimes it sends the investigator backward to repair a weak assumption. Sometimes it reveals that the question has already been answered. Sometimes the correct transition is productive abandonment.

In that sense, Scientific Formation Infrastructure is not an assembly line for theories. It is the machinery surrounding the ladder that helps determine whether a transition has genuinely been earned.

The Knowledge Candidate is the object being formed. The Formation Ladder describes the state it has earned. The Formation Infrastructure supplies the capabilities needed to test and support movement between those states.

Seen this way, the architecture follows from the transitions rather than from a predetermined list of features. Each subsystem exists because a particular movement in scientific formation demands a capability that neither the investigator nor a general-purpose AI assistant should be expected to supply reliably on its own.

The Inquiry Environment supports the movement from curiosity toward a precise question. The Grounding Engine helps a question encounter the knowledge that already surrounds it. The Claim and Assumption Graph exposes what must become explicit before a grounded question can become a defensible claim. The Formalization Workbench moves consequential relationships from language into domain-appropriate precision. Experiment environments allow those formal candidates to encounter evidence.

As the candidate matures, the problem changes. The Evidence Ledger preserves the lineage of what happened. The Adversarial Engine subjects the emerging claim to deliberate resistance. Expert Escalation introduces scarce specialist judgment where machine-assisted formation has reached its useful boundary. The Contribution Commons preserves what the inquiry has actually produced, including useful failure. The Capability Passport records what has changed in the investigator, not merely what happened to the theory.

The architecture therefore follows the same formation logic as the ladder, but it performs a different job. The ladder answers, “What has this inquiry earned?” The infrastructure asks, “What must happen before it can responsibly earn the next state?”

The architecture does not need every subsystem on its first day. What matters is that each one serves this same formation logic.

The formation model is now coherent enough to face a different question. The issue is no longer what the infrastructure could eventually contain, but where the infrastructure itself can first be tested without asking the world to trust it too soon. Formation defines the machinery; Embodiment must choose an environment in which that machinery can encounter consequence, reveal weakness, and fail safely.

A proposition this broad becomes dangerous when ambition outruns evidence. The first embodiment should not attempt to support every scientific discipline, adjudicate medical theories, orchestrate physical experiments, or build a universal artificial scientist.

The right question is smaller: where can the infrastructure itself be tested without requiring us to believe the entire thesis in advance?

Computational domains provide an unusually strong starting point because many of the important objects already exist digitally. Experiments can be reproduced at low marginal cost. Formalization is often explicit. Provenance can be captured automatically. Large search spaces can be explored without exposing people or physical environments to experimental risk.

Wolfram's ruliology is particularly interesting in this respect. Whatever one concludes about its larger scientific significance, it illustrates the type of environment in which outsiders can explore formal computational worlds directly rather than beginning with the full accumulated apparatus of an established experimental science.

Foldit offers another version of the same structural lesson. The important innovation was not simply asking the public for help. Researchers created an instrumented environment in which domain structure, computational feedback, constrained manipulation, human pattern recognition, collaboration, and professional validation could interact.

The broader proposition should begin by learning from that pattern.

A responsible expansion path

Increase consequence only after the formation and governance machinery has earned confidence.

Productive abandonment rate may be one of the most important measures of success.

A system that teaches people to abandon unsupported claims earlier may be creating more scientific value than one that produces a larger number of impressive-looking theories.

The pilot now tells us where to begin and what to measure. It does not yet tell us whether the proposition contains blind spots created by its own enthusiasm for access, AI, participation, and scale. Before Embodiment is treated as readiness, the architecture should face a different kind of Trial: criticism from intellectual perspectives that do not begin inside its own vocabulary.

NOTE: The named Council seats are not quotations, endorsements, reconstructed statements, or claims about what Aristotle, Peter Drucker, Hyman Minsky, Joseph Schumpeter, Friedrich Hayek, Elinor Ostrom, Amartya Sen, Nicholas Georgescu-Roegen, Wendell Berry, Norbert Wiener, or Donella Meadows would personally conclude about the proposition examined here.

They are interpretive reasoning scenarios generated by the Kriyas Machine. Each seat applies a selected intellectual lens derived from publicly available writings, ideas, frameworks, and scholarly interpretations associated with that thinker. The resulting analysis is produced by the machine and the author and should not be attributed to, or understood as representing the views of, the named thinkers.

Scientific Formation Infrastructure is particularly vulnerable to intellectual enthusiasm because nearly every ingredient sounds desirable on its own. More people participating in science sounds desirable. Better access to knowledge sounds desirable. AI tutoring sounds desirable. Reproducibility, mentorship, citizen contribution, open research, and computational exploration all sound desirable.

A coherent list of desirable ingredients can still produce a bad system.

The proposition therefore needs criticism from outside its own vocabulary. It has to confront questions about human purpose, institutional effectiveness, fragility, innovation, distributed knowledge, governance, agency, physical limits, stewardship, control, and system behavior.

That is the work of the Council of Eleven.

The Council does not destroy the proposition. It changes its center of gravity.

The original idea could still be heard as a system for finding promising amateur theories and moving them toward professional science. After Trial, that framing is no longer strong enough. It places too much importance on the survival of the originating idea, too much faith in AI-mediated scale, too little emphasis on governance, and too little value on what happens to the participant when the theory fails.

The stronger proposition is about formation capacity.

Its value does not depend on discovering thousands of unknown geniuses. It creates value when a weak question becomes precise, when prior knowledge replaces accidental reinvention, when a participant learns to distinguish claim from evidence, when a beautiful theory fails early, when an experiment becomes reproducible, when expert attention arrives at a genuinely unresolved problem, when a negative result prevents duplicated effort, and when a person leaves the process more capable of disciplined inquiry than when they entered.

That is a much harder system to build.

It is also a much more consequential one.

Open the pathway. Do not lower the standard. Leave the investigator more capable than you found them.

Trial has now done what it was supposed to do: not approve the proposition, but alter it. Completion can therefore separate the attractive assumptions that should be discarded from the architecture that has actually survived evidence, design constraints, and adversarial pressure.

The inquiry began close to Wolfram's original problem: many people have theories, artificial intelligence can help them express those theories, and perhaps a new environment could connect computation, mentorship, community review, and publication into a pathway toward serious science.

That formulation is no longer sufficient.

“Amateur” is too narrow a category because the underlying formation problem also affects students, interdisciplinary researchers, professionals crossing domains, independent scholars, founders, and even experts operating outside their primary field. Theory submission begins too late because the most valuable intervention may occur before a person has formed a claim at all. Mentorship cannot serve as the primary scaling mechanism because expert attention remains scarce. Publication sits too far downstream to define the opportunity. Artificial intelligence cannot become the scientific adjudicator without collapsing assistance into authority.

Most importantly, breakthrough discovery cannot be the justification.

A formation system that produces only a small number of genuinely novel discoveries could still create enormous value if it creates stronger investigators, better questions, earlier falsification, more reproducible work, more efficient expert attention, and a richer commons of positive and negative knowledge.

The next great expansion of participation in science will not come simply from allowing more people to publish theories. It will come from giving more people access to the machinery through which theories are disciplined.

Artificial intelligence makes this both possible and necessary. It makes it possible because explanation, tutoring, coding, literature navigation, formalization assistance, simulation, critique, and research orchestration can now be provided at a scale that would have been economically impossible when each function required direct specialist attention.

It makes the infrastructure necessary because those same technologies make plausible but unsupported intellectual production extraordinarily cheap. When persuasive expression ceases to be scarce, expression becomes a weaker signal of underlying quality. The disciplines that connect a claim to reality become more important, not less.

The resulting opportunity is Scientific Formation Infrastructure: an environment in which raw curiosity can progressively acquire grounding, precision, formal representation, tests, evidence, counterevidence, provenance, criticism, reproducibility, and appropriately timed human judgment.

The objective is not to ensure that more theories survive.

The objective is to ensure that more curiosity becomes better inquiry, that weak ideas are productively corrected sooner, that expert attention reaches questions worthy of its scarcity, that useful knowledge survives failed investigations, and that participants accumulate genuine intellectual capability through the process itself.

The long-term possibility is therefore larger than avocational science. It is an open formation layer for human discovery: a place where someone can arrive with little more than a question and progressively earn the right to make stronger claims through demonstrated understanding, evidence, computation, criticism, and accountable work.

AI has made answers abundant; the next frontier is infrastructure that makes disciplined inquiry abundant without making truth cheap.

Completion does not establish that Scientific Formation Infrastructure will work at global scale. It does not prove that millions of independent investigators will produce a new era of scientific discovery, nor does it establish that every domain can be served by the same machinery.

Completion establishes something more modest and more useful: the proposition is coherent enough to deserve embodiment.

That sends the inquiry back into Operating Reality with a better question than the one with which it began.

The question is no longer, “How do we help amateur scientists develop their theories?”

It is whether we can encode enough of the formation of reliable inquiry into a human-machine system to make rigorous investigation accessible at much larger scale without reducing science to an algorithm, displacing expertise, centralizing epistemic authority, corrupting the incentives that make falsification possible, or leaving the human investigator intellectually weaker than the machine assisting them.

That question can now be tested.

And that is exactly where a Kriyas cycle should end.

Not with certainty.

With a better beginning.

In an age when machines can manufacture answers almost instantly, one of society's most valuable infrastructures may be the machinery through which human beings learn how to earn the right to believe them.