The algorithm that drew me wrong

Recently, one of my students found a photo of me on the internet. I am posing with a smile, with the Bosphorus Bridge in the background. They uploaded this photo to an artificial intelligence tool and asked it to generate an image of me. They must have also mentioned that I am an academic, as I was depicted holding a pen and writing something on paper. The resulting image is surprisingly good. My face looks exactly like me, the expression is like mine, and the lighting and details are convincing. They had it framed and gave it to me as a gift. I thanked them for this lovely present. Then, looking at the picture, I realized there was a mistake: I am holding the pen with my right hand. However, I am left-handed and hold a pen with my left hand.

The artificial intelligence drew me incorrectly because it does not know me in this way.

This small error opens up a big question: Why did the artificial intelligence make such a mistake, and how can we fix it?

The first possibility is algorithmic bias. Artificial intelligence systems are trained on large datasets. In these datasets, the majority of people use their right hands. Therefore, for the system, a person holding a pen is right-handed by default. Left-handedness, being in the minority, becomes invisible. As Safiya Noble has shown, algorithms do not represent the world as it is, but rather as the majority of the data suggests.

The second possibility is a lack of information. Because the artificial intelligence did not have enough information about me, it chose the most likely option. It made an “educated guess.” Because these systems substitute the highest probability for reality. At this point, an attractive solution emerges: Let algorithms know us better. Let them have more information about us. Then they might represent us more accurately. At first glance, this doesn't seem like a bad idea.

But it is right here that we encounter the real question. Because knowledge is not innocent. As Foucault said, knowledge is always a power relationship. As Zuboff points out, this information is the raw material of today's behavioral prediction and guidance economy.

A system that wants to draw me more accurately eventually turns into a system that predicts me better, guides me better, and manages me better. As Langdon Winner said, technological systems are not just tools; they set norms. A person writing with their right hand becomes the norm. Others are the exception. This is a small but instructive example of the epistemic violence described by Spivak. Some subjects are systematically misrepresented or not represented at all.

The resulting image is a simulation. As Baudrillard said, a simulation does not reflect reality; it replaces it. A version of me that is not me, but looks like me, enters circulation.

And all of this creates a risk. As Beck said, this risk is not external; it is inherent to the system.

Finally, this process has an emotional dimension. Wanting artificial intelligence to represent us correctly is also a desire to be understood correctly. As Papacharissi and Illouz have shown, emotions are also a managed and exploited field today.

Therefore, the issue is not just about which hand I hold the pen with.

The issue is this:

How accurately should algorithms know us?

An algorithm that draws me wrong is annoying.

But an algorithm that draws me perfectly is more dangerous.

Because perfection is where distance disappears.

And sometimes, distance is safer than error.