Financial Times column says smell is still hard for AI to copy

In short: A Financial Times opinion column argues that human smell remains one of the few personal traits that today’s AI still struggles to identify and copy.

What's going on

The column, written by Elaine Moore, says many people feel privacy is shrinking because AI can already copy faces and voices and because some companies have faced reports about mishandling data. Against that backdrop, it points to smell as a more stubborn holdout.

Barry Smith, director of the Institute of Philosophy at the University of London’s School of Advanced Study, is quoted saying he is glad some human attributes are “not digitally coded.” The column argues that smell has been neglected in research for years, so there is less information for AI systems to learn from.

The piece also says smell helps keep a clear line between real life and digital life. Smith argues that even very realistic virtual worlds will still feel incomplete without smell. It is like watching food videos when you are hungry, you can see it, but you cannot fully experience it.

What to watch

The column notes that researchers and startups are trying to change this. Alex Wiltschko, who helped spin the startup Osmo out of Google Brain, says the biggest barrier is data, meaning enough labeled examples that connect molecules to what people report smelling. MIT researchers have also built a smell database called SmellNet.

If machines ever get better at smelling, the column says it could lead to practical tools, like detecting disease or spotting rotten produce. For now, it argues, smell remains difficult for AI because it is deeply physical, when you smell something you are literally taking tiny particles of it into your body.

Source: Financial Times