TL;DR
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
Software engineer Jesse Waites says an AI-assisted search across historical archives surfaced a forgotten meteorite report, records of three rhinoceroses, and possible unrecorded volcanic eruptions. The findings are candidates for historical verification, not a replacement for specialist catalogues; Waites says he checked promising passages against scanned original documents.
Software engineer Jesse Waites says an AI-assisted search of historical archives surfaced a forgotten meteorite report, records of three rhinoceroses, and possible unrecorded volcanic eruptions. The work points to how machine-assisted screening can help researchers find leads in vast digitized collections, though each candidate still needs to be checked against original documents and existing specialist records.
Waites described the project in a report about searching millions of historical records, including transcriptions from the Dutch East India Company archive, digitized newspapers in the Netherlands and the United States, and selected ship logbooks. The GLOBALISE transcription collection alone contains 4.35 million pages spanning the 17th century through the 1790s, according to the source material. Waites says his system processed 5.7 million passages from that collection during an overnight run.
The workflow combined searches based on meaning with staged AI review. Waites says a smaller, inexpensive model screened large groups of passages for narrow criteria, such as whether a record referred to a real wild animal and where it was reported. A more capable model then translated and extracted details from a smaller set of candidates. Waites says he and the AI agent also checked the scans of the original pages, rather than relying on text transcriptions alone.
The report identifies a meteorite account, three rhinoceros records and possible volcanic observations among the leads. Waites says he compared candidates with catalogues used by specialists before treating them as potentially new. The supplied account does not give enough detail to independently establish the date, location or scientific classification of every find, and the discoveries should be understood as reported archival leads rather than settled revisions to historical or scientific records.
AI Can Narrow the Archive Search
The project’s broader contribution may be its method: using AI to reduce an unmanageable volume of material to a set of passages that a researcher can inspect. Historical handwriting, old spelling and imperfect optical character recognition make ordinary keyword searches unreliable. Waites says the system searched by the meaning of passages as well as their wording, then used a low-cost screening stage to cut down the candidates sent for closer review.
That approach could help scholars locate evidence dispersed across large collections, including records that were digitized but not previously connected to a particular research question. It does not make the resulting interpretation automatic. The original page, its provenance and comparisons with established catalogues remain necessary checks. The findings matter as examples of AI-assisted discovery, while their historical significance depends on confirmation by relevant specialists.
AI software for historical research
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
From Dodo Record to Wider Search
Waites says his project was inspired by historian Benjamin Breen, who published a post on October 1, 2026, describing an AI-assisted search of Dutch East India Company records. Breen reported finding a 1615 ship’s journal from Mauritius that said sailors had caught tortoises and dodos, a potential eyewitness account concerning the extinct bird. GLOBALISE has made millions of handwritten Company pages searchable through transcriptions.
Waites extended that idea by searching for several kinds of historical evidence in parallel, including animals, meteorite falls, earthquakes and a major volcanic eruption in 1808 whose location is unresolved. He emphasized that automated text recognition can make errors and that spellings varied widely, particularly in older records. His process therefore used the machine-generated transcription to find candidates, then returned to the scanned document for checking.
“AI models can be confidently wrong, so every claim had to end at a real document.”
— Jesse Waites, in his report
Which Findings Will Be Verified?
The supplied report excerpt does not provide full archival citations or detailed descriptions for each of the meteorite, rhinoceros and volcanic candidates. It is therefore not possible from this material alone to establish precisely what each document records, whether the observations are genuinely absent from existing scholarship, or whether they warrant changes to specialist catalogues.
Waites says he checked promising records against original scans and consulted specialist catalogues, but independent assessments from historians, zoologists or geologists are not included in the source material. AI screening can also miss evidence or misread text, especially where handwriting, spelling and transcription quality are poor. The reported finds remain leads whose significance depends on review of the documents and relevant records.
Archival Leads Need Specialist Review
The next step is for researchers and readers to examine the cited archival pages and compare them with the catalogues Waites consulted. Full references would allow other specialists to assess the readings, dates, locations and whether the records add genuinely new evidence. The source material does not announce a formal peer-reviewed publication or a scheduled review of the findings.
Further searches could test whether the same staged workflow works across other collections, but its results will depend on transcription quality, archive coverage and the questions researchers ask. For now, Waites’s report shows a way to surface candidates; it does not establish that every candidate is a confirmed discovery.
Key Questions
What did the AI-assisted search find?
Waites reports finding a forgotten meteorite report, records involving three rhinoceroses, and possible unrecorded volcanic eruptions. The supplied material does not include enough detail to verify each item independently.
Did AI make the discoveries on its own?
No. Waites says he designed and supervised the workflow, selected research questions and checked candidate passages against scans of original documents. AI models helped search, screen, translate and extract information.
How large was the archive search?
Waites says the GLOBALISE collection contains 4.35 million pages and that his system processed 5.7 million passages from the Company archive during an overnight run. Those figures describe the collection and the reported processing, not a verified count of new discoveries.
Are the reported findings confirmed by specialists?
The source material says Waites checked candidates against specialist catalogues and original scans, but it does not include independent specialist assessments. The findings are best treated as archival leads pending further review.
Source: hn
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
