Status Report: AI in Genealogy and Family History

The following report was generated by Gemini in response to the prompt:
Give me a status report on the use of AI in human family tree and family history software and services.
The conversation went on and I will post some follow-up that could be of interest to RootsMagic users.

---

The integration of artificial intelligence into genealogy has moved beyond gimmickry into practical core infrastructure. AI tools are actively streamlining archival processing, record extraction, and data organization, while generative text models provide historical context.

At the same time, the fundamental rule of genealogy remains unchanged: AI accelerates hypothesis generation, but human verification against primary sources remains essential.

---

Core Areas of AI Implementation

1. Handwriting Recognition & Automated Indexing (HTR)

Handwritten Text Recognition (HTR) is arguably the most impactful application of AI in genealogy today.

• FamilySearch & Public Archives: Large-scale automated indexing projects use vision-language models to process un-indexed or image-only historical record collections (such as parish registers and probate records). Records that previously took volunteers decades to index are now text-searchable through tools like FamilySearch Full-Text Search in a fraction of the time.
• Transkribus & Custom Models: Specialized platforms like Transkribus allow researchers to upload difficult historical hands—ranging from 18th-century cursive to German Kurrent, Latin, or Cyrillic—to produce automated transcripts with high accuracy.

2. Specialized Search & Record Matching

Commercial repositories are leaning into AI to move beyond exact-match string searches:

• Ancestry & MyHeritage: AI-driven matching algorithms analyze tree structures alongside record collections, looking at contextual hints (e.g., matching occupation, nearby land boundaries, or common naming patterns) rather than relying solely on name/date hits. For example, Ancestry utilizes machine learning for tools like SideView™ to phase DNA matches by parental side.
• Findmypast: Implements AI-powered transcription across historical newspaper collections, making surname and location searches practical across vast press archives via Findmypast.

3. Photo Enhancement & Computer Vision

Visual AI remains a high-visibility feature, though opinions in the genealogy community are split between utility and historical preservation.

• Restoration & Upscaling: Deep-learning models effectively remove scratches, sharpen focus, and correct color fading, as seen in the MyHeritage Photo Enhancer.
• Facial Recognition: Tools automatically tag potential matches of unknown relatives across photo albums by matching facial structures across different ages.
• Generative Manipulation: Features like portrait animation (MyHeritage DeepStory) or cross-generational composite family photos are popular for engagement, though seasoned researchers caution against confusing creative reconstruction with historical evidence.

4. Document Structuring & Local Research Assistants

A growing shift in desktop and web tools focuses on turning unstructured documents into structured relational data.

• Emerging platforms and custom LLM workflows allow researchers to upload scans of wills, deeds, or letters. The AI extracts structured facts (names, roles, dates, relationships, locations) and auto-links them to existing tree nodes or databases.
• Genealogists routinely use general LLMs—such as ChatGPT, Claude, or local open-weights models running via Ollama—to interpret archaic terminology, translate Latin vital records, or construct historical context summaries.

---

Current Industry Challenges & Limitations

• Hallucinations in Tree Building: Generative AI models struggle with strict evidence evaluation. When asked to “fill in a family tree,” models frequently invent missing ancestors, mix up individuals with identical names, or synthesize plausible-sounding but nonexistent sources.
• Privacy & Opt-In Concerns: Platform integrations of third-party AI features have sparked community debate regarding data sharing, auto-opt-in defaults for user trees, and copyright over user-contributed records.
• Citation & Provenance: Automatically extracted metadata still requires human audit. AI tools frequently miss subtle transcript cues (e.g., ditto marks, marginal notes, or crossed-out entries) that dramatically alter record interpretation.

---

Summary Matrix

Feature / Task	AI Maturity	Best Used For	Primary Risk / Limitation
Cursive & Manuscript HTR	High	Un-indexed parish records, wills, journals	Misinterpreting regional abbreviations or phonetic spellings
Translation & Record Reading	High	Latin, German, Scandinavian vital records	Subtle legal terminology mistranslations
Photo Restoration	High	Sharpening, scratch removal, color recovery	Fabricating details in severely degraded faces
Record Matching / Hints	Medium-High	Surfacing leads across massive datasets	Propagating unsourced or wrong tree connections
Automated Tree Construction	Low	Draft generation, timeline brainstorming	Hallucinating individuals, conflating same-name people