Some Thoughts on Decision Automation

Sat, Feb 16, 2019 3-minute read 498 words

Decision fatigue is a real-life phenomenon, not just a theoretical concept. Scientific studies show that humans have a limited amount of willpower, and that decision fatigue inevitably leads to poor decisions or even decision paralysis. One way to preserve willpower is to automate the most mundane decisions. In this post I share some thoughts on designing such system for my personal use.


In my current vision, automated decision making system has the following components:

  • problem that requires solution
  • set of solutions to choose from
  • procedure for selecting the optimal solution
  • feedback mechanism

Everyday Problems

Choosing daily meals is a great example of a problem that I encounter regularly and would love to automate. Nothing induces decision fatigue more than trying to decide what to cook for next week! But the same system can be applied to other problems, including choosing what to wear, how to spend spare time, etc.

Start with the Choices

The simplest approach to decision automation is to have a set of predefined choices so no creative input is required. This initial investment of time and willpower needs to be made only once. The same options then can be used again and again.

For instance, instead of asking “what should I eat for breakfast”, I ask “should I eat a) cereal b) scrambled eggs or c) peanut butter toast?". The “cereal”, “scrambled eggs” and “peanut butter toast” choices were pre-made based on my nutritional requirements and personal preferences.

I am on a six meals a day plan, so I am currently developing a fixed set of options for each meal type:

  • 6 options for breakfast
  • 12 options for lunch
  • 12 options for dinner
  • 6 options for snacks

It’s a work in progress, as I continue refining the options. The ultimate goal is a set of 36 meals that require minimal time and ingredients to make, and satisfy my palate and nutritional needs.

Making Decisions

Now that the system has a set of choices, I need to set up a procedure for building the weekly meal plan. The simplest approach would be to randomly choose each meal separately. A more efficient system would have more advanced features:

  • random selection with no replacement (so each option is chosen only once a month, except for snacks and breakfasts)
  • bundled meals (breakfast, lunch, dinner and snack for the same day chosen as a bundle)
  • accounting for social rhythm (e.g., higher calorie meals on workout days)
  • seasonal variations (meals optimized for seasonal ingredients)

An ideal system woud need to be used only once, generating a meal plan and grocery list for the entire month with one click.

Feedback Mechanism

Finally, a simple mechanism for collecting feedback could help to continue improving the system. This includes evaluating meals based on gratification (some meals become tiresome over time) or fullness factor, recording changes in cost (money or time), etc.

I am still in the process of building components of the system, and will be occasionally sharing my progress on this blog.