AI Knowledge Discovery & Preservation Initiative
Better Than Synthetic Data: Preserving Human Expertise, Connecting Verified Knowledge, and Accelerating Discovery
Executive Summary
The AI Knowledge Discovery & Preservation Initiative proposes investigating whether AI can become more accurate and capable of discovery by connecting real-world evidence, preserving human expertise, and uncovering relationships hidden across disconnected sources.
Rather than relying on increasingly large models or synthetic data alone, the initiative would test whether better access to real, verified, interconnected knowledge can improve AI performance and scientific discovery.
The initiative begins with two priorities:
Better Than Synthetic Data: Compare real, verified, interconnected information with synthetic data and hybrid approaches using controlled evaluations.
Preserve Human Knowledge Before It Disappears: Use AI to identify and preserve valuable professional knowledge, research, and personal archives before they are discarded or lost.
If these approaches demonstrate measurable benefits, the initiative could expand into scientific research, environmental history, archaeology, genetics, disease research, and other fields.
The goal is to test the concept with a small, measurable pilot before committing to a large-scale program.
1. Research Priority: Better Than Synthetic Data
Central question: Can AI improve its accuracy, reasoning, and ability to discover new connections by learning from real, verified, interconnected knowledge rather than relying primarily on synthetic examples?
Synthetic data can be useful, but generated information can also reproduce errors, amplify biases, or reinforce conclusions that lack independent evidence.
The initiative would investigate where real-world evidence offers advantages, where synthetic data helps, and whether combining the two produces better results.
Connect reliable scientific publications, experimental results, historical documents, environmental observations, and authorized human archives.
Preserve source information so AI can trace conclusions back to supporting evidence.
Compare real-data, synthetic-data, and hybrid approaches on equivalent tasks.
Evaluate factual accuracy, source verification, scientific reasoning, cross-disciplinary connections, and generalization to unfamiliar questions.
Investigate whether previously disconnected information reveals useful patterns or research opportunities.
Clearly distinguish documented facts, reasonable inferences, hypotheses, and generated material.
Success criterion: Demonstrate measurable improvements on defined tasks rather than assuming real data or synthetic data is universally superior.
2. First Practical Project: Preserve Human Knowledge Before It Disappears
The problem
People can spend decades developing specialized knowledge, yet their records may be destroyed after they die.
My father worked as a claims adjuster for 63 years and accumulated approximately 500 pounds of professional papers. Those papers were eventually discarded.
This illustrates the potential loss of knowledge that may never have been published, digitized, or incorporated into formal research.
Proposed solution: AI-assisted knowledge recovery
Step 1 — Identify experts
Examine public obituaries, professional biographies, retirement announcements, institutional histories, and other lawful sources.
Identify people whose careers suggest valuable, specialized expertise.
Prioritize potentially irreplaceable knowledge that is poorly documented elsewhere.
Step 2 — Locate potential archives
Look for notebooks, professional papers, correspondence, unpublished research, technical drawings, photographs, and digital records.
Identify leads through families, former colleagues, employers, professional associations, universities, and historical societies.
Use obituaries as starting points, not as proof that private records exist or are available.
Step 3 — Obtain permission
Contact families, estates, executors, or other authorized representatives.
Explain the preservation project's purpose and potential benefits.
Respect privacy, copyright, ownership, confidentiality, and access restrictions.
Step 4 — Digitize and organize
Convert authorized documents into searchable digital collections.
Use AI to classify, index, and summarize materials.
Preserve original records and their historical context.
Step 5 — Connect the knowledge
Link documented expertise to related scientific research, historical evidence, and technical literature.
Identify potentially valuable insights that conventional searches may overlook.
Ensure conclusions remain traceable to their sources.
Step 6 — Preserve for future generations
Establish appropriate storage, access controls, attribution, and long-term preservation.
Allow authorized rights holders to determine appropriate access and usage conditions.
Why begin here? Knowledge preservation is time-sensitive. Once unique records are destroyed, they may be impossible to recover. This project also provides a concrete way to test AI-assisted discovery using real information.
3. Expand Into High-Value Research Areas
If the initial work demonstrates value, the initiative could investigate additional opportunities.
A. Scientific research and failed experiments
Connect scientific publications, experimental results, and research datasets.
Preserve documented failures and negative findings.
Help researchers avoid repeating unsuccessful experiments and identify unanswered questions.
B. Earth's history and environmental change
Connect NASA, NOAA, Copernicus satellite observations, environmental sensors, climate records, and historical documents.
Reconstruct changes in forests, ecosystems, coastlines, and landscapes.
Investigate historical deforestation in Iceland and Greenland, including timber use for housing, ships, heating, charcoal production, and iron smelting.
Examine evidence of ancient ecosystems, including Antarctica's past forest environments.
Investigate how historical environmental changes can inform forest restoration and modern ecological research.
C. Archaeology and ancient societies
Connect archaeological sites, human remains, artifacts, ancient DNA, geological evidence, and historical records.
Investigate migration, settlement, adaptation, and relationships between ancient populations and their environments.
Identify gaps where multiple independent forms of evidence could improve historical understanding.
D. Scientific infrastructure and major research datasets
Investigate access to suitable resources from CERN and other scientific research institutions.
Connect relevant datasets with publications and documented experimental results.
Respect access requirements, licensing, and data-use restrictions.
E. Historical epidemics and chronic disease
Connect historical medical records, documented outbreaks, population studies, and modern research.
Investigate changes in disease patterns over time.
Account for incomplete records and differences in historical diagnosis.
F. Genetics, evolution, and viruses
Connect ancient and modern genetic research with evolutionary biology.
Investigate documented relationships between organisms and their environments.
Examine viruses that infect other viruses, including virophages, to identify promising scientific questions.
G. Carefully validated synthetic data
Identify situations where synthetic data can supplement scarce real-world information.
Compare synthetic and real-data approaches under controlled conditions.
Validate conclusions independently and prevent generated material from being mistaken for observed evidence.
These are research opportunities to evaluate, not guaranteed discoveries.
4. Create a Continuous Knowledge Discovery Process
The long-term vision is a repeatable research cycle:
Discover: Identify gaps, contradictions, and unanswered questions.
Search: Find relevant evidence across connected sources.
Connect: Identify relationships between previously separated information.
Verify: Check provenance, reliability, and competing explanations.
Test: Evaluate hypotheses against independent evidence.
Preserve: Record findings, failed approaches, and remaining uncertainties.
Repeat: Use the results to guide further research.
This approach would support human researchers rather than replace them.
5. Run a Three-to-Six-Month Pilot
Begin with a limited project using existing AI systems and legally accessible information.
Pilot activities
Select a manageable set of research questions.
Build a curated, source-traceable knowledge collection.
Include a small, authorized archive-preservation component where feasible.
Compare conventional retrieval, connected real-world evidence, synthetic data, and hybrid approaches where appropriate.
Evaluate performance using the same questions and standards.
Document costs, limitations, and results.
Measurements
Factual accuracy and source verification.
Quality of reasoning across multiple sources.
Ability to identify useful connections.
Performance on unfamiliar questions.
Research time and cost.
Reliability of document extraction and indexing.
Compliance with privacy, ownership, and provenance requirements.
A two-to-five-percentage-point absolute improvement in accuracy could be an experimental target, but it should not be treated as guaranteed or assumed to apply to every task.
6. Safeguards
The initiative should:
Respect privacy, consent, copyright, licensing, and ownership.
Maintain source provenance and attribution.
Distinguish facts from hypotheses and AI-generated interpretations.
Preserve conflicting evidence and negative findings.
Prevent synthetic data from serving as false independent confirmation.
Use independent evaluations and human oversight where appropriate.
7. Preliminary Budget
An initial planning estimate for a three-to-six-month pilot is $115,000–$355,000.
This is an illustrative estimate, not a verified quote. Actual costs would depend on staffing, archive access, digitization needs, computing, and evaluation requirements.
Existing infrastructure and available datasets should be used wherever practical.
8. Request for Consideration
I invite AI research organizations to:
Evaluate whether connecting verified knowledge can improve AI performance.
Assess the feasibility of an obituary-led human knowledge preservation project.
Support or evaluate a limited pilot with measurable outcomes.
Determine which scientific and environmental research areas offer the most promising next steps.
Expand the initiative only if independent results justify further investment.
Conclusion
Humanity has accumulated enormous amounts of knowledge, but much remains scattered across scientific publications, research datasets, historical records, environmental observations, and personal archives.
Some knowledge disappears because nobody preserves it. Other knowledge remains inaccessible because the relationships between sources have never been explored.
AI may help address both problems.
The central proposition is that AI research should investigate whether connecting verified real-world knowledge, preserving human expertise, and systematically evaluating evidence can deliver improvements that synthetic data or larger models alone may not provide.
A small, carefully evaluated pilot could determine whether this approach is practical, beneficial, and worth expanding.
More on Bible Tech: The Physics of How the World Ends · Why did God create Marijuana
More from Jason Bulsa
Books, playbooks and kits from Rainbow Innovations — packaging fixes, psychology, longevity, and the systems behind this blog — are all in one store, every checkout test-verified:
Visit the Rainbow Innovations Store →