History in the Gaps

As I continue exploring history with ChatGPT, I keep finding gaps in its answers. At first, I thought they were random, but I was always suspicious—and now the pattern is clearer. It’s not always about lies; it’s about what gets left out.

Here are some examples:

• Tulsa Massacre (1921): For nearly a century, schools didn’t teach it. Only recently did it begin to re-enter mainstream memory.
• Night of Terror (1917): Suffragists jailed and beaten at Occoquan Workhouse for demanding the vote. A brutal moment, barely covered in many history books.
• Trump’s first term (2016–2017): For a period, there was real LGBT+ progress, including openly trans people serving in the military. It worked. Units trained, deployed, and carried on. Yet today, that reality is already being erased from some tellings.
• What’s happening in Los Angeles right now: Guard deployments are being legally contested, but official narratives downplay or sanitize the turmoil.
• The floor of the House of Representatives: On the day pandemonium broke out, the official record said nothing happened.

In each case, something real happened—but depending on how you ask or where you look, it can vanish.

Why These Gaps Happen

I know there are soft and hard guardrails. I know there is bias in the data. But I’ve also learned how these erasures happen:

• Frequency bias: Events that aren’t taught in schools or widely referenced appear less in the data, so AI and search surface them less often.
• Metadata over media: Unless specifically asked (and at great cost), AI doesn’t deeply analyze every image or video. A video of peaceful protestors described as a riot gets remembered as a riot.
• Authority bias: Official sources, especially government records, short-circuit further inquiry. If the record says nothing happened, then nothing happened.

In none of these cases is the AI lying. But omissions and confident framing leave us with falsehoods all the same.

Why It Matters

History can be rewritten not just by burning books, but by skipping chapters—or quietly tearing out a page. A truth unreferenced is a truth forgotten. And if we let yesterday’s facts be smoothed away, then today’s facts are next.

It’s up to us to figure this out. If we don’t want to lose our past, then we must make sure our AIs—and our own archives—carry all of our stories, not just the ones convenient to those in power.

Experimental Call to Action: Catch Erasure in Real Time

Here’s your experiment:

1. Pick something you just witnessed. A headline, a policy change, a floor fight in Congress, a deployment order. Doesn’t matter if it feels “big” or not—what matters is that you saw it unfold.
2. Capture the raw record. Save a screenshot, transcript, or firsthand note. Don’t rely on summaries—grab the thing as it happened.
3. Revisit in a week. Look at the official record, the news write-ups, and even AI summaries. What details have gone missing? What’s been softened, reframed, or contradicted?
4. Name the gaps. Don’t argue interpretation yet—just point to what disappeared. That act alone keeps memory alive.

---

Erasure doesn’t only happen to last century’s tragedies or last decade’s reforms. It happens to yesterday, and it can happen to today before the dust even settles. The AI isn’t lying, but omissions and framing leave us with falsehoods all the same.

If we want to keep our past—and our present—from being rewritten, then it’s on us to make sure all of our stories survive, not just the ones convenient to those in power.

Try this experiment once, and you may never unsee it again. And once you do, you’ll recognize how quickly we can lose not just yesterday’s history, but today’s.