AI Engineering Blog | Read the AI Articles | Monolith

The Great baike off | Monolith AI

Written by Admin | Dec 17, 2020 12:00:00 AM

For almost all of 2020, everyone at Monolith has been working remotely. Many of us have sorely missed the social aspect of working in an office together. 

So, every day at 3 pm, everyone who works at Monolith can join a remote 'tea time' call, our way of adding the social aspect of office space. Those who join talk about everything other than our jobs. It's 20 minutes of random conversations and epic tangents. 

Just like everybody else who has religiously watched The Great British Bake-off and Masterchef, many of us love cooking or baking. Food comes up often in tea time calls, and recently, cookies became a talking point. 

What's the best cookie? Some like it gooey, and some like an extra crunch. So, an idea started forming - why not use machine learning to optimize chocolate chip cookies? It's not that different from what our platform is used for by engineers in manufacturing. 

 

  • Ingredients - The manufacturing materials
  • Pre-bake preparation - Desire production output and initial design conception
  • Baking - The manufacturing process

Our CEO, Richard, decided that the process will be a mixed-initiative system where both human chefs, human cookie-raters, and a machine optimizer will participate in 120 experiments.

 

Process allocation was conducted in Monolith AI

 

Each participant was randomly allocated different tunable parameters to use in their cookie bake.

  • Either you have to melt or cream the butter
  • Brown sugar, white sugar or a 50 per cent mix of both
  • Rest the dough at room temperature or an hour in the fridge
  • A baking temperature between 170 and 190

Everyone will bake six cookies. 

  • One pair will be baked for 10 mins
  • Another at 12 mins
  • The last set will be baked for 14 mins

Finally, the baked cookies will be given a rating on three criteria:

  • Texture - from brittle to soft
  • Colour - from light to dark
  • Taste - from unedible to incredible

120 items in a dataset are enough for our platform to train a machine learning model but as everyone at Monolith shared the idea with our loved ones and others, we realised more people wanted to participate. Either for science or just to have an excuse to think that Paul Hollywood would shake their hand. So, we are opening up the experiment to everyone. You can fill in the form below and you will get your baking recipe, instructions, and a form to add your rating for each bake! Of course, we will notify you when the final model is ready!