Machine learning and wine: what can we learn?

As a software engineer, I’ve spent all of my career thinking about how to make processes more efficient, and these days, the hype in automation centers around one topic: artificial intelligence. Machine learning, in particular, is a subset of artificial intelligence that has been driven to the forefront in the wake of declining costs and increased availability of computing power and data storage. Now that it’s cheap to store and analyze lots of data, the possibilities for different industries to change and grow are multiplying, and fast.

There are three primary ways in which artificial intelligence is already impacting the world of wine.

1. Agricultural advances: Vineyard managers can collect data from their vineyards and use it to inform decisions about how to manage the winemaking process. Aggregating data across producers and from different sources, such as weather, could remove more of the uncertainty inherent in the winemaking process.
2. Improving the customer experience: As we’ve discussed in class, consumers can be overwhelmed by the number of options they have and don’t have great signals to help them choose the wines they drink. Machine learning can provide more personalized recommendations to consumers based on their preferences.
3. More efficient pricing and predictions: A standard use of machine learning is to take sets of structured data and make predictions about a particular variable. One application is to predict demand and pricing for wine based on historical data. For example, it might be possible to predict the best price for wines in your newest vintage by looking at historical pricing, weather, and chemical composition data to develop a better guess on how it will be received in the market.

Organizations like the University of California, Irvine, are collecting and publishing datasets. The public availability of these datasets has enabled independent researchers, including off-duty software engineers and data scientists, to experiment with new models for using this data. UCI has published datasets on chemical analysis of wines (1999) and wine quality (2009), and some researchers have started to collect their own data. As the amount of data increases, the possibilities for applying statistical techniques to wine will only continue to expand.

The big question is: what will you build to take advantage of this new opportunity?