Alignment
Every participant responds to the same question. The responses therefore share a known point of origin.
Pollagraph
Language models have the text. Pollagraph adds human reasoning.
Writing in Parallel gives multiple people the same carefully designed prompt and captures their independent reasoning in structured paragraphs.
One Prompt. Multiple Minds. Structured Human Reasoning.
Why Pollagraph
Pollagraph adds something different: human reasoning deliberately produced around known questions under known conditions.
Existing text gives language models extraordinary breadth. Pollagraph creates complementary reasoning data by asking people to respond independently to carefully designed questions.
The Principle
Classic expert systems established an important idea: human knowledge could be deliberately acquired for computers.
Knowledge engineers interviewed experts, captured what they knew, and organized that knowledge for machine use.
Pollagraph carries that principle forward.
The Method
Writing in Parallel is Pollagraph's method for deliberately acquiring human reasoning.
Multiple people receive the same carefully designed prompt. Each person responds independently in a structured paragraph.
Because everyone begins with the same question, the responses share a common point of origin.
Because everyone reasons independently, different perspectives, assumptions, judgments, and reasoning paths emerge.
One prompt. Multiple minds. Multiple reasoning paths.
Writing in Parallel
The Relationship
Pollagraph calls the relationship among these independent responses Aligned Variation.
Every participant responds to the same question. The responses therefore share a known point of origin.
Every participant reasons independently, bringing different judgment, experience, assumptions, and interpretation to the question.
Different reasoning paths remain connected through the common prompt that produced them.
The Product
A standard Pollagraph reasoning set makes the idea concrete.
One carefully designed question establishes the common point of origin.
Ten people consider the same question independently.
Each writer contributes one structured response.
Different approaches to the same question remain visible.
The responses remain connected through their common prompt.
The Data
Pollagraph calls the resulting data Structured Human Reasoning Data, or SHRD.
The value comes not only from what each person writes, but from the known relationship among the responses.
From Expert Systems to Language Models
Pollagraph does not recreate the architecture of the classic expert system. It carries forward the principle of deliberately acquiring human knowledge for machines.
The Pollagraph Thesis
Language models already have enormous quantities of human language.
Pollagraph proposes something complementary: human reasoning deliberately produced around known questions under known conditions.
Writing in Parallel creates the responses. Aligned Variation describes their relationship. SHRD preserves that relationship as structured data.
Pollagraph exists to pursue that possibility.
Where the Data Begins
A Writing in Parallel session puts Pollagraph's knowledge-acquisition method into practice.
Participants receive common prompts and respond independently in structured paragraphs.
The common prompt creates alignment. The participants create variation. Writing in Parallel captures both.
The resulting responses can be organized into SHRD datasets around particular questions, subjects, problems, or knowledge domains.
The Pollagraph Process
The Pollagraph System
Each part of Pollagraph has a distinct role.
Deliberately acquire human knowledge for machines.
Ask multiple people the same carefully designed question.
Preserve different reasoning paths around a common point of origin.
Structure the responses and preserve their relationships.
Pollagraph
Pollagraph connects institutions that can produce structured human reasoning with AI organizations interested in exploring that reasoning as a complementary source of data.
For Institutions
Bring participants together around common prompts and participate in producing structured human reasoning data.
Host a SessionFor AI Organizations
Explore deliberately produced human reasoning as a complementary source of data for language-model development.
Explore SHRDRead the Book
How Polling for Paragraphs Can Solve the AI Trust Problem
Go deeper into the ideas behind Writing in Parallel, Pollagraph, and structured human reasoning.