Perils of AI

Perils of Using AI for Local History Research

I asked Gemini to explain the "perils of using AI for local history research" since I found that inaccurate websites skew the algorithms/models used by AI. I will present an example of an inaccurate local website that AI treats as factual in my next post. Below is Gemini's response. Please consider this when using AI for research.

"Using AI for local history research can be a double-edged sword. While it offers the promise of sifting through vast amounts of data quickly, it also presents significant perils, primarily related to accuracy, bias, and context,

• Hallucinations and Fabrication: A major risk is that AI models can generate "hallucinations"—plausible-sounding but completely false information. They may invent names, dates, or events to fill in knowledge gaps, which is particularly dangerous for local history where verifiable information can be scarce.

• Data and Algorithmic Bias: AI models are trained on data from the internet, which often reflects existing societal biases. For local history, this can mean that marginalized communities or groups with fewer digital records may be underrepresented or misrepresented. The AI might prioritize mainstream narratives, ignoring or distorting lesser-known perspectives.

• Inability to Evaluate Sources: AI lacks the critical thinking skills of a human historian. It can't discern the credibility of a primary source from a secondary one or a biased blog post from a scholarly article. It may aggregate data from unreliable sources, perpetuating errors and misinformation.

• Oversimplification and Lack of Context: Local history is rich with nuance, including socio-political, economic, and cultural contexts. AI tends to oversimplify complex historical events, reducing them to generic summaries that miss the interpretive depth and emotional resonance of the past. It can't replicate the experience of a human historian who might hold a fragile, handwritten letter from the period they're studying.

• Lack of Primary Source Access: Many local history records, such as government documents, old newspapers, and personal archives, are not digitized and are therefore inaccessible to AI. This means the AI is working with an incomplete dataset, making its conclusions inherently limited and potentially misleading.

In short, while AI can be a powerful tool for analyzing large datasets and identifying patterns, it should not be considered a substitute for rigorous, human-led historical research. Its output should always be critically evaluated and cross-referenced with verified sources to ensure accuracy."