Recent breakthroughs in artificial intelligence have produced a steady stream of hyperbolic headlines. We are told that chatbots can pass medical board exams, that AI-designed drugs are entering clinical trials, and, increasingly, that machines have developed a sense of smell. From reports of computers \"learning to smell\" to claims that AI \"tastebuds\" outperform humans, the narrative suggests a machine that is rapidly colonizing the five senses.
However, a closer look at these milestones reveals a significant gap between chemical perception and statistical correlation. As Philip Maughan notes in *Noem*, many of these \"sensory\" breakthroughs are merely Large Language Models (LLMs) echoing human associations found in their training data. When an AI links the color pink or a round shape to sweetness, it isn't \"tasting\" the abstract; it is simply repeating a linguistic pattern common in human description. It is a mimicry of sentiment, not a mastery of molecules.
The lack of genuine progress in digital olfaction stems from a fundamental lack of interest within the industry. While vision and language have been digitized with immense success, the volatile and complex world of scent remains largely ignored. For all its prowess in logic and generative art, AI remains trapped in a sterile environment, disconnected from the chemical reality that defines much of the biological experience.
With reporting from 3 Quarks Daily.
Source · 3 Quarks Daily
