AI, Accuracy, and the Problem With Inventing Farrah

---

One problem with AI-generated images is that they can look completely convincing while being completely wrong. The mistake doesn't have to involve some obscure historical detail, either. Sometimes it can be something as basic as eye color. I have seen several synthetic images of Farrah Fawcett that give her blue eyes. Her eyes were green.

On the surface, that may seem insignificant. Changing someone’s eye color does not rewrite her career, alter a major event in her life, or change how we understand her as a person. But that is precisely why the example is useful. We are not talking about reconstructing a disputed event from decades ago. We are talking about a basic physical characteristic that can be checked against thousands of photographs and hours of film and television footage.

If AI can get something that simple wrong while still producing an image that looks photographic, it raises a larger problem. Viewers are being asked to trust what they see because the image looks real, even though parts of it may have been invented. That is where synthetic imagery becomes more than a technical curiosity. It begins to interfere with visual accuracy.

The problem is not that artificial intelligence is incapable of being useful. I use AI myself, and I have no objection to using it to repair damage, improve poor reproductions, reduce noise, or help restore old photographs. In those cases, the original image remains the source. The goal is to recover or present what was already there more clearly.

Creating a new photographic image of a real person is different. Once that happens, the software is no longer simply cleaning up the historical record. It fills in details, makes choices, and sometimes gets them wrong. Hair color can shift. Facial structure can change. Clothing can be invented. Expressions can appear that never existed. Even something as basic as eye color can be altered without the image looking obviously fake.

Most people won't stop and analyze every image they see online. If a synthetic picture looks convincing enough, many viewers will accept it. Then the image gets downloaded, reposted, cropped, shared again, and separated from whatever explanation originally identified it as AI-generated. Eventually, the fabricated version can circulate alongside authentic photographs with little distinction between the two.

I have already seen this happen many times with Farrah. Artificially altered and completely synthetic images can travel far because they look familiar enough to seem legitimate. Once that happens, correcting the record becomes difficult, if not impossible. A real photograph has to compete with an invention that may be brighter, smoother, more dramatic, or simply more eye-catching.

This is why small mistakes matter. The blue eyes themselves are not some major historical scandal. The significance is in how easily the mistake can pass unnoticed. If viewers accept an incorrect eye color because the image looks believable, what else might they accept for the same reason?

The use of AI feels especially unnecessary with someone as extensively photographed as Farrah Fawcett. She left behind an enormous visual record. We know what she looked like. We have studio portraits, magazine photography, television appearances, films, interviews, candid images, and decades of footage. No software needs to guess at her appearance.

That is really the point. AI-generated images of real people do not simply create new pictures. They create new information, whether the person making them intends that or not. Sometimes that information is harmless. Sometimes it is inaccurate. The danger is that both can look equally convincing.
​
If AI cannot reliably preserve something as basic as eye color, we should be much more cautious about treating synthetic images of real people as anything more than what they are: inventions.