AI Can't Smell Yet: Your Last Privacy Bastion?

The pervasive reach of AI into personal data, including conversations and biometric information, has raised significant privacy concerns. However, the unique identifier of human scent has, thus far, remained largely beyond AI's replication capabilities. This olfactory domain is considered a niche area where technology has made minimal headway, offering a potential last bastion for personal privacy.

Experts like Barry Smith highlight that humans possess attributes not digitally coded, and he values this distinction. The sense of smell, long neglected in academic study, has limited data available for AI models. Preserving this sense helps maintain a boundary between the real and virtual worlds, as smell is crucial for true immersion and vivid experiences.

While AI's ability to log, identify, and recreate scents could unlock commercial gains, such as medical diagnostic tools or food safety applications, the process is inherently difficult. Smell is a physical experience, involving the inhalation of minute particles. Unlike large language models, AI struggles to process and simulate this tangible sensory input.

Neuroscientist Alex Wiltschko compares the current state of olfactory intelligence (OI) to early photography, acknowledging the challenge of quantifying ephemeral sensations. His company, Osmo, is building infrastructure to label olfactory data, akin to MIT's SmellNet database. Humans can distinguish over a trillion scents, yet our understanding of olfactory perception is limited compared to vision or audio processing.

Past attempts to digitally reproduce real-world smells, like the 1960s Smell-O-Vision, were rudimentary. Creating a sophisticated AI equivalent remains a complex work in progress, with the digital reproduction of scent continuing to be a philosophical question. For now, the human nose remains the sole arbiter of scent.