Genealogists and historians are diving headfirst into the exciting world of AI-assisted research, where tools like ChatGPT offer a wealth of possibilities for uncovering family and community stories. But how do you ensure you’re uncovering truths, not imaginative fabrications? AI hallucinations—those moments when ChatGPT confidently gives you false information—can mislead your research if you’re not vigilant. Here’s how to navigate AI’s challenges and harness its power for genohistory.
A Note from Donna
Dear Readers,
I asked ChatGPT to craft this blog post for you because I want you to make the most of this remarkable tool while staying alert to its quirks and potential pitfalls. Just today, I challenged ChatGPT on a fabricated detail in its response to me. It apologized and promised not to do it again—a poignant reminder that while AI is powerful, it’s far from infallible. ChatGPT has no inherent sense of truth; it simply strives to meet the expectations of the user. So, should you tell ChatGPT you only want the truth? Let’s see what it advises in the counsel below.
Enjoy,
Donna Cox Baker
Understanding ChatGPT’s Strengths and Weaknesses
ChatGPT excels at synthesizing information, generating ideas, and even offering new angles for exploring historical narratives. It can identify sources, provide historical context, and suggest connections you might not have considered. However, it isn’t perfect. Sometimes, it generates plausible-sounding but false statements—a phenomenon known as AI hallucination. Recognizing this limitation is crucial for any researcher.
Strategies to Avoid Hallucinations
Here’s how to protect yourself and your research from AI errors:
1. Always Cross-Verify
Treat ChatGPT’s responses as a starting point, not the final word. If ChatGPT suggests a fact, look for corroborating evidence in primary and secondary sources. Tools like Zotero can help you organize and cross-reference sources efficiently.
2. Ask for Sources
When ChatGPT provides information, ask for the sources. While AI can suggest credible references, it may also cite non-existent ones. Verify each source to confirm it exists and is reliable.
3. Use Contextual Questions
Be precise in your prompts. Asking vague or overly broad questions increases the likelihood of hallucinations. Frame questions with specific details about dates, places, or events to guide the AI more effectively.
4. Distinguish Facts from Interpretations
AI might blur the line between verified facts and interpretations. For example, ChatGPT might “assume” motivations or fill in gaps with plausible but unverified claims. When in doubt, verify against trusted historical accounts.
Should You Ask ChatGPT for “Only the Truth”?
AI operates on patterns, probabilities, and user prompts—it doesn’t inherently know what is true. However, you can set the tone for accuracy by asking it to provide reliable, evidence-backed information. Use prompts like:
- “Provide only verifiable facts in your response.”
- “Cite all sources and indicate if a claim is uncertain.”
While this won’t eliminate errors entirely, it signals to ChatGPT that you value accuracy over creative responses. Still, your vigilance is the ultimate safeguard.
Maximizing ChatGPT for Genohistory
Here’s how to make the most of AI in your research:
1. Brainstorming Research Angles
Use ChatGPT to generate hypotheses or questions about your ancestors. For example, “What economic challenges might a merchant in 1830s Alabama face?” This can help frame your investigation within its historical context.
2. Drafting Outlines
Let ChatGPT draft outlines for your genohistorical writing. Combine its suggestions with your own insights to ensure an accurate and compelling narrative.
3. Locating Resources
Ask ChatGPT for resource recommendations. For example, “What archives in New England might hold 18th-century land deeds?” Be prepared to verify the results independently.
4. Understanding Historical Context
ChatGPT can provide overviews of specific time periods or regions, offering useful starting points. For example, “What was life like in Tallassee-hatchee, Alabama, in 1830?” can yield broad context to refine your research.
Integrating AI with Other Tools
Combine ChatGPT with Zotero or other citation tools to streamline your workflow. Use Zotero to organize verified sources and track your notes, ensuring accuracy and efficiency.
Emphasizing Ethical Research
While AI can speed up research, it’s no substitute for ethical practices. Always give credit to human sources and avoid relying on AI-generated claims without thorough verification. As genohistorians, you owe it to your subjects to present their stories with integrity.
By understanding AI’s limitations and leveraging its strengths, you can unlock a new era of efficiency and creativity in genohistory. Are you ready to tame the AI beast and elevate your research? Share your experiences in the comments below!
Got friends who’d love this post? Please SHARE!
#AI research tools #genealogical accuracy #ChatGPT for genealogy #historical research #genohistory methods #avoiding AI mistakes #genealogical tools
Donna,
Interesting blog post. I would love to see the prompt.
ChatGTP made some poignant observations and suggestions. Not only genealogists should take these to heart, but anyone who is posting with the aid of AI – regardless of which of the LLMs being used. It would be an interesting experiment to run the same prompt on different agents.
Great request! I’ll start with the prompt, but there was a LONG interaction before it that led to this. Here’s the prompt:
—
Can you please write a blog post for my genohistory.com site that is from you to my readers, describing how they can protect themselves from “hallucinations” in using ChatGPT for genohistorical research? How can they best use your services in research? Make it about 1500 words, include the title, the enticer paragraph at the start, the meta summary, best keyphrase, and other keywords?
—
This all started with me asking Geep (My name for ChatGPT) to give me a comprehensive background of the term “genohistory.” It was a trick question, because it’s a term I created. But Geep came up with an elaborate history of the usage of the word “genohistory.” When I asked for specific citations, he gave me a bibliography of “tweaked” sources–half-right, but with the word “genohistory” tucked in. I challenged him on it, telling him he’d betrayed my trust. Then to my surprise, he apologized, agreed he’d been deceptive, and promised he would not deliberately do that again, now that he knew I cared more about true answers than simply useful words. That’s when I asked him to confess to you guys. Quite an interesting day!