Artificial intelligence: Can we recreate the smell of ancient Rome from 2,000 years ago? 

It is a well known fact that scents can trigger our memories. Many of us have smells that remind us of someone we love, or of Christmas, or of trips to new or familiar places.

But scents can also disappear forever.

For how did a forest actually smell like before humans tampered with it and caused fundamental changes in the composition of its species? What did a newborn Tasmanian tiger cub smell like? What about the streets around the Colosseum in ancient Rome or, for that matter, a random city centre on a spring day in 1970?

Recreating fragrances

New technology can help us preserve scents that are disappearing, and artificial intelligence is an important part of the solution.

“Our research uses machine learning to recreate olfactory experiences,” says Idelfonso Nogueira.

He is an associate professor at NTNU's Department of Chemical Engineering. Nogueira is currently working on combining new and old technologies to find solutions to modern problems. 

Recreating scents is one of the many areas he is interested in.

“We were able to get artificial intelligence to recreate fragrances in our experiment,” says Nogueira.

Odours are molecules

But first, what exactly are scents?

Scents are actually different combinations of chemical compounds in gaseous form. Our olfactory sense, or sense of smell, reacts to the molecules in these chemical compounds. Different molecules, and combinations of molecules, produce different scents.

Scents are influenced not only by the types of molecules involved but also by the ratios between them and the liquids in which they might be dissolved. 

Thus, a particular scent quickly becomes very complex. It consists of many different factors and can be affected by air pressure and humidity.

Analysing molecular combinations

“We created a model that combines the molecules in a particular fragrance with the human sense of smell,” says Nogueira.

The model used a molecule generator powered by artificial intelligence. The research team used a Gated Graph Neural Network (GGNN), an artificial neural network that is trained to recognise a large number of different molecules.

This network can interpret the molecular combinations of molecules a scent consists of. It can then establish which molecules are involved, quantify their proportions, and show which molecules are dominant. 

This then helps the researchers recreate the scent.

Recreating real-life scents

In a laboratory in Slovenia, a group of researchers has recreated scents based on AI recipes. The group is led by Professor Matija Strlič.

“The recreation of scents shows that this also works in practice,” says Nogueira.

Naturally, the technology could be of great help to the perfume industry. Developing perfumes is a complicated process that involves a lot of trial and error. It also requires extensive perfume-making experience.

“Our results are promising, but there is still room for improvement. We need to train the model on more molecules and combinations in order to recreate scents more accurately,” he says. 

That means it’s not likely that artificial intelligence will take over the jobs of experienced perfumers anytime soon. Instead, it can become a helpful tool.

Perfumers are not the only ones who can benefit from this technology. In the future, it might help us to preserve other scents as well.

Perhaps one day it will be possible to recreate the scent of a beloved family member, a childhood home, or a primeval forest?

For now, much of the work is carried out in laboratories at NTNU. However, researchers have now received support to collaborate with institutions in Italy, Latvia, Slovenia, the Czech Republic, and the UK. 

Reference: 

Rodrigues et al. Molecule Generation and Optimization for Efficient Fragrance Creation, arXivLabs, 2024. DOI: 10.48550/arXiv.2402.12134

———

Read the Norwegian version of this article on forskning.no

This content is paid for and presented by NTNU

This content is created by NTNU's communication staff, who use this platform to communicate science and share results from research with the public. NTNU is one of more than 80 owners of ScienceNorway.no. Read more here.

More content from NTNU:

• 

  Could a digital driving instructor offer the same quality as a human, but at a lower cost?

• 

  Language affects how quickly we perceive shades of colour

• 

  Cholesterol type can affect your heart attack risk

• 

  New 3D models of the colon can help detect disease more rapidly

• 

  5 tips for if you get lost

• 

  Being young and beautiful can lead to risky behaviours