Can AI guess your travel plans?

Have you ever...

  • ...waited more than 10 minutes after calling an Uber?
  • ...realized while packing your suitcase that it's too late to buy a book for the plane?
  • ...waited in bumper-to-bumper traffic while dozens of drivers tried to board the same freeway ramp?

In each case, apps could improve your experience by predicting where you want to go. Though it sounds creepy, it doesn't entail extra data collection. If I'm going to share my GPS data anyway, I want apps to use it in a way that's proactive, not reactive.

So, as a weekend side project, I set out to discover if I could use AI to predict my next move. Here's a view of my "virtual self" traveling through Seattle:

What's really cool is that you can use any place you've traveled in the past as a starting point and the model will predict your next move. So, if you vacationed in Jamaica and drop yourself there, the map will gravitate back toward your home.

How it works

  1. Takes into account 317,288 GPS coordinates collected over a 4-year period
  2. Reduces the precision of latitudes and longitudes so places can be grouped together.
  3. Uses the latitude-longitude pairs to form a Markov Chain with the condition that no place can be re-visited

I didn't have time to make a web front-end, but if you're comfortable with the command line, here's a script for analyzing your own data in about 5 minutes.

So...can AI predict your travel plans?

Yes! There's no denying that the concept sounds creepy at first, but in my opinion, an ethical, transparent use of this algorithm would generate a lot of value for customers. I would love for Uber drivers to be one step ahead of customers calling rides and to receive a neck pillow in advance of booking a 12-hour flight.