How technology is changing human perception

Getting a tattoo has traditionally consisted in making the prediction about your future self: “Would I still love this band, symbol or the name of this person currently occupying a stretch of my forearm years from now?” Researchers at the University of Colorado Boulder have developed PhotoTat, an experimental system of photochromic inks that introduces a little more flexibility into that decision.

Photochromic materials change the way they absorb light when exposed to particular wavelengths. In PhotoTat, blue, magenta and yellow dyes are encapsulated in tiny particles made from polymethyl methacrylate, or PMMA. Ultraviolet light activates the pigments and makes the colors visible. Bright light reverses the effect and returns them to a nearly colorless state. Mixing the three inks can create a broader palette, allowing multicolor images to be switched, faded or rewritten with light.

This, however, doesn’t mean the ink fully disappears from the skin. The change only affects its visibility, so this means the pigment remains in the skin while its visibility changes. This work is still experimental and broader human-testing would be needed before it could be offered as an everyday tattoo option. Even so, it points toward body art that could be more adaptable, interactive and potentially useful beyond mere body modifications. 

Describing colors could be a straightforward task. However, describing a smell often requires a comparison: roses, toasted bread, or, if the moment calls for it, teen spirit. A Yale-led study is exploring how machine learning can help organize those impressions and predict how similar complex odors mixtures will smell to people.

The research grew out of an international competition where teams developed AI models to predict how similar different smell mixtures would smell to people. After the challenge, researchers combined predictions from six high-performing models into an ensemble. The model outperformed existing approaches on a hidden test set and performed strongly on an independent set of 50 newly designed scent-mixture pairs.

One of the most interesting findings was that the model relied heavily on semantic descriptions of odors, not only on molecular structure. This is key, as smell has very limited vocabulary for everyday use. We usually name the object with what a scent reminds us of rather than the scent itself. Instead of being a weakness, language provided highly useful in predicting relationships between mixtures. “Something ‘smells like flowers’ or like watermelon or like a rotten egg, whereas colors have specific names like ‘green’ or ‘blue,’” lead author Vahid Satarifard of Yale’s Human Nature Lab told the Yale News. “That was thought to make semantic odor descriptors a poor basis for predicting how smells relate, but we found the opposite.”

For now, the study demonstrates something very important: AI could help give structure to a sense that has remained difficult to measure, explain or digitize.