For years, computers could utilize two of the five senses — sight and hearing. However, computers interpret these senses differently than humans do. Computers translate color into numerical values (RGB) which is how computers see. Similarly, computers hear by converting sound waves, which are usually captured by a microphones, into a digital signal which converts these waves into binary, so the computer can understand. Recently, scientists have found a way to give computers a sense of smell.
Computers do so by utilizing devices like olfactometers and GC-MSs. Olfactometers are presented with a source of the odor, which they use to simulate the human response to that odor (ex. how strong the odor is). On top of this, GC-MSs can be used to convert the smells into their chemical structure. Smells are molecules that evaporate and react with your olfactory system, so all smells have a unique chemical structure. By combining perception and structural data from the two devices, datasets can be compiled to train AI models to recognize specific scents. However, the combination of these two to make datasets is not as easy as it sounds. Extremely large data sets with a large amount of scent data must be created to train the AI. However, assuming that these models are present, thoroughly trained AI models can then recognize any scent given to the olfactometer and GC-MS. These AI models have several future applications. Some of these applications include air pollutant detection, detection of medical conditions, and personal scent detection devices.
Certain air pollutants dispel smells that can sometimes be confused with other smells. For example, hydrogen sulfide smells like rotten eggs. Once trained enough, the AI models can be presented with the smell through the olfactometer and GC-MS and it can recognize that the smell is H2S and not rotten eggs based on the chemical structure, and it can also determine the severity of the pollutant, based on the pungency.
Another application of computer scents is for the detection of certain medical conditions. For example, patients with diabetic ketoacidosis (diabetes and alcoholism) often have constant fruity breath. With the right controls in place (no foods eaten or drank for some time before the check-up), doctors can use AI models fine-tuned to recognize the scent of a DKA patient’s breath. This can be used for various conditions that cause a part of a person to smell a certain way.
Finally, a more abstract application of these models is in personal devices. For example, a toothbrush with waterproof sensors, a speaker, and an integrated AI model could be created. With this toothbrush, you could finish brushing your teeth, blow into the sensor, and the toothbrush would tell you whether your breath is fresh or if you need to keep brushing. These models have near-infinite possibilities, and as time passes and improvements are made, these scent recognition models will improve and be present in the lives of many!