Would You Let AI Pick Your Christmas Dinner?

We gave our model 3,000 ingredients from 350 festive recipes to generate the most sustainable Christmas menu.

Christmas lunch always has the potential to raise more than its share of tension. The burned turkey is a staple of festive slapstick comedies and once you work in different traditions and personal tastes, putting together a meal can come to seem more like a political negotiation than a domestic chore. With Christmas 2021 falling in the immediate aftermath of COP26 and the climate crisis feeling more urgent than ever, there’s now another factor to think about: the carbon footprint of our celebrations. The last thing we want is for our festive get-togethers to be soured by climate guilt.

We wondered if the kind of machine learning and advance analysis that we carry out on the financial markets could help optimise a festive menu and address the key issue of sustainability as people go about planning their Christmas dinners. From the more than 1,600 recipes on the BBC Good Food website, we isolated the 357 tagged with the term “Christmas” and began organising them into categories.

Problems loading this infographic? - Please click here

Source: BBCGoodFood.com, Man Group calculations. Sustainability data kindly provided by Dr Michael Clark, Oxford Martin Programme on the Future of Food.

We then used Natural Language Processing (NLP) – a system by which artificial intelligence parses large amounts of text – to make sense of over 3,000 ingredients found in the Christmas recipes. For example, from the instruction “2 medium Desirée potatoes, peeled and thinly sliced”, our artificial intelligence model can simplify this to just “Desirée potatoes”. Next, we used a technique called linear programming to find the ‘optimal’ combination of ingredients. In other words, the model had to find the ingredients that together unlocked the largest number of recipes.

Problems loading this infographic? - Please click here

Source: BBCGoodFood.com, Man Group calculations.

With the ingredients and recipes optimised for efficiency, we now needed to optimise for sustainability. To do this, we turned to Dr Michael Clark, Fellow at the Oxford Martin Programme on the Future of Food, and the Livestock Environment and People Programme (LEAP). Dr Clark provided us with a deeply researched and impressive databank of the carbon cost of the ingredients in our shopping baskets. We were then able to score each recipe for its sustainability, highlighting those specific ingredients that carried the greatest carbon footprint. Finally, we were able to construct a menu that delivered the lowest possible carbon footprint while drawing on a predetermined number of ingredients.

Without any restrictions on the number of items in your shopping basket, the most sustainable meal our model recommends is:

It’s no surprise that meat is largely – but not entirely – missing from our shopping basket. The emissions generated and the land and water required dwarfs that needed by vegetarian options. If these don’t appeal, though, you can plot the sustainability of your meal of choice on the scatter chart below.

Problems loading this infographic? - Please click here

Source: BBCGoodFood.com, Man Group calculations. Sustainability data kindly provided by Dr Michael Clark, Oxford Martin Programme on the Future of Food.

We will be managing the climate crisis for many years to come, but it feels like the festive period, which is, yes, about consuming and celebration, but also about giving and kindness, should be a time for us to focus particularly on what we can do to leave a better world for those who come after. With the help of machine learning techniques and advanced analytics, there’s a way to help you ensure that your festive meal doesn’t cost the earth.

 

Our thanks go out to Gary Broughton for his original work on the BBC Good Food Recipe data and to Erica Greene and Adam Mckaig from the New York Times, and Michael Lynch for open sourcing their work on conditional random fields.

User Country: United States (237)
User Language: en-gb
User Role: Public (Guest) (1)
User Access Groups:
Node Access Groups: 1