The Specification

TAM-UNF.02 · The Ungoverned Frontier · The Approximate Mind

She does not specify the composition. She specifies the behavior.

A strength-to-weight ratio that exceeds anything currently available. Flexibility within a certain range. Biodegradability after a specific period. She types these into a system that searches the combinatorial space of possible material structures, a space so large that no human researcher could explore even a fraction of it in a lifetime. The system proposes candidates. She evaluates them against criteria that include properties she was not explicit about: how the material feels in the hand, whether it has what she calls warmth, a quality she cannot quantify but recognizes immediately when she encounters it.

She has a habit, noticed by everyone in the lab, of running a finger along any new sample before she writes anything down. The material before the notes. The knowing before the language. Her advisor told her, twenty years ago: instruments measure what you tell them to measure, and your hands tell you what you didn’t think to ask.

The system returns a candidate. She runs her finger along it. It has the strength-to-weight ratio she asked for. The flexibility range. The biodegradability profile. It also has something she did not specify and cannot fully name. A thermal response she noticed only when she touched it. A behavior at the material’s boundary that no parameter in her specification described.

She writes it down. Then she writes: where did this come from?

The Leaky Container

#

Every specification is a description of what you want written in the language of what you already know.

This is the specification’s fundamental limit. You can specify what you have the vocabulary to describe. The vocabulary comes from what you have already encountered, which means the vocabulary comes from the existing frameworks through which the domain has been understood. What lies outside those frameworks cannot be specified. It can only be found, and then recognized, if the recognizing capacity extends beyond the specifying capacity.

The materials scientist’s warmth is the clearest case. lies outside that vocabulary, the properties whose existence you do not yet know to expect, cannot be specified. They can only be found, and then recognized, and the recognition is possible only if the recognizing capacity extends beyond the specifying capacity.

The materials scientist’s warmth is the clearest case. She could not have specified it because she had no word for it before she found it. The word came after the contact, when the material gave her something she then needed to name. The specification shaped the search. The search exceeded the specification. The excess is the part she did not ask for, and it is, in this case, the most important part of what she found.

This is not a failure of specification. It is what specification in a large combinatorial space necessarily produces. The space contains more than any specification can describe. The search finds things that fit the description and other things alongside them, things that are real and relevant and sometimes more important than what was described. The discoverer asked for one thing. She received that, and something else she did not know she needed, and the something else arrived because the search space was large enough to contain it.

The gap between what the specification asked for and what the search returned is not a residue. It is the discovery.

What Grows in the Gap

#

The materials scientist received one unexpected property alongside what she specified. This is the smallest version of the gap.

The gap has a scale relationship that matters. At the individual level, the specifier is close enough to the search that she can feel when the result exceeds the specification. The materials scientist runs her finger along the sample. The thermal response is there; she notices it; she is changed by the noticing. The gap is productive because the person who opened it is also the person who receives what comes through it.

Scale the specification to a research team. The team specifies the properties of a class of compounds for a therapeutic application. The search returns candidates that meet the therapeutic specification and exhibit other behaviors in adjacent biological systems. Some behaviors are irrelevant. Some are harmful. Some are more therapeutically significant than the original application. The team asked for something specific. The search returned something specific and its neighbors, and the neighbors were not visible in the specification because the specification was written in the language of what was already known.

At this scale, the gap between intent and discovery is wider than at the individual level. The team’s collective specification was more precisely engineered than the materials scientist’s intuitive criteria, and more precisely engineered specifications cut more sharply through the possibility space, finding what they describe and also what lives next to it. The team can still examine what the search returned. But the examination requires distributing attention across many findings, some of which exceed the specification in ways nobody on the team has the background to evaluate. The gap is productive only if the team includes the recognition capacity that the specification cannot describe.

Scale further. The autonomous pipeline maps the gaps in the documented territory and generates its own specifications for what to search. The specification is now produced by the pipeline, not by any human researcher. The gap between intent and discovery has been institutionalized across multiple automated layers. The pipeline specifies, the pipeline searches, the pipeline evaluates what it finds against criteria derived from the existing literature, the pipeline flags what deviates from expectation, and a researcher receives outputs whose relationship to any human intent is seven layers of automated reasoning removed.

At this scale, the gap is not a feature of one interaction. It is the architecture of the process. Every output exceeds its specification in ways the specification could not describe, because the search space is large enough that what matches the description also includes what the description didn’t anticipate. The discovery is in the gap, distributed across 400 items in a queue, waiting for someone with the recognition capacity to find it.

What travels along this arc is not the gap’s existence. The gap opens at the individual level and scales through the process. What changes at each scale is who bears the recognition burden. The materials scientist bears it herself. The research team distributes it across members with varying preparation. The pipeline distributes it across researchers who receive outputs from processes they did not design, about domains they may not fully inhabit, from search spaces they cannot personally survey. The recognition capacity, the prepared mind that gives the gap’s output meaning, is increasingly separated from the specification that opened the gap. This is the series’ central problem, stated at its smallest scale, before any governance question arises.

The discovery was always in the gap. The question is always whether anyone is prepared to receive it.

The Recognition Problem

#

The materials scientist found the warmth. She recognized it as important before she could name it. Her hand told her what her specification had not described. The recognition was possible because she had spent enough time in the domain to know when something unexpected was also something significant.

Recognition is the capacity that makes the gap productive rather than merely unpredictable. But recognition has a scope. The materials scientist can recognize what is significant in her domain. She cannot necessarily recognize what is significant in adjacent domains that the search space happens to traverse. If the thermal response she noticed had implications for structural biology that she had no framework to see, those implications would pass through her hands unrecognized. The gap produced them. The recognition capacity didn’t extend to them.

This is the structural problem that scales with the process. The individual researcher has recognition capacity in her domain and diminishing capacity in adjacent territory. The research team pools recognition capacity across disciplines, extending it somewhat, but still leaving most of the possibility space beyond the collective preparation of whoever happens to be on the team. The autonomous pipeline returns findings whose implications may require recognition capacity distributed across fields that are not in conversation with each other, held by researchers who will never encounter each other’s outputs.

The gap is always productive in proportion to the recognition capacity that receives what it produces. The pipeline is scaling the gap faster than recognition capacity scales. More findings, in more domains, at more intersections between domains, arriving at queues staffed by researchers prepared for one domain who must evaluate outputs from searches that crossed into ten others.

This is not an argument against the pipeline. It is a description of what the pipeline most urgently needs alongside it: not just the companion systems from the previous essay, but the cultivation of cross-domain recognition capacity at the scale the pipeline requires. People who have prepared themselves to recognize significance across domain boundaries, who can read an output from a search that crossed five fields and know which of the five has something urgent in what came back.

That capacity is rare. It develops slowly. It cannot be specified into existence.

I wonder whether the rate at which we are expanding the pipeline’s output is outpacing the rate at which we are cultivating the recognition capacity that makes the output productive rather than merely voluminous, and whether the gap between specification and discovery will eventually become not a source of unexpected findings but a source of unexpected findings that nobody is prepared to receive.

She is still looking at the notation she wrote: where did this come from?

She knows the answer in a narrow sense. The system searched a combinatorial space and returned a candidate that met her parameters and exhibited the thermal behavior as a structural consequence of meeting them. The behavior was latent in the structure, waiting for anyone who specified the right parameters. She found it because she asked the right questions, not because she asked for this specific answer.

The discovery was in the gap between what she asked and what she received. It is always there. The gap is not a malfunction of the process. It is the process working as it must when the search space is large enough to contain what has not yet been named.

She runs her finger along the sample again. She begins writing the new specification: what she now knows to look for, because what she found told her it was there. The gap produced the knowledge that lets her close it. This is how it has always worked. The tools have changed the scale. The logic is the same.

This is Part 2 of The Ungoverned Frontier. The gap between intent and discovery opens here, in the simplest act of specifying and receiving something the specification did not contain. The series continues in Part 3 (The Collision), where the gap expands: multiple specifications meet unexpectedly, and what emerges belongs to no single intent.

References

#

Craft, Specification, and Making

Sennett, Richard. The Craftsman. Yale University Press, 2008.

Ingold, Tim. Making: Anthropology, Archaeology, Art and Architecture. Routledge, 2013.

AI and Materials Discovery

Merchant, Amil, et al. “Scaling Deep Learning for Materials Discovery.” Nature, vol. 624, 2023, pp. 80–85.

Szymanski, Nathan J., et al. “An Autonomous Laboratory for the Accelerated Synthesis of Novel Materials.” Nature, vol. 624, 2023, pp. 86–91.

Discovery and Recognition

Kuhn, Thomas S. The Structure of Scientific Revolutions. University of Chicago Press, 1962.

Polanyi, Michael. Personal Knowledge: Towards a Post-Critical Philosophy. University of Chicago Press, 1958.

Combinatorial Search and Creativity

Boden, Margaret A. The Creative Mind: Myths and Mechanisms. 2nd ed., Routledge, 2004.

Kauffman, Stuart. At Home in the Universe: The Search for Laws of Self-Organization and Complexity. Oxford University Press, 1995.