Wanted: Algorithms that create useful outputs
I love smart algorithmic approaches to problems. Created and used well they make many things in our lives and environment better. I trust the algorithms in my thermostat to work out the most efficient way to heat or cool my home and shut it down when I leave. As I do with the plethora of algorithm driven safety systems in my modern cars constantly analyzing the objects around me. I find some recommendation engines really useful because when buying a box of bolts, I want the nuts and washers that go with them and I may in fact be in the market for a new wrench to tighten them up with.
What I do not like are algorithms that create bad outputs due to either inappropriate design or techniques to build them. Or poor quality or understanding of the data that powers them. A case in point is an email that I just received from an on-line travel agent I used to book some hotel accommodation earlier this year.
West Drayton for those of you who do not know it is not a small exclusive Caribbean or Pacific Island resort. It is a residential neighborhood just to the north of London’s Heathrow Airport. Do I as the subject of the email asks “remember West Drayton?” Nope, firstly because the Airport hotel I stayed in is not actually in West Drayton and secondly it was just an overnight when coming into LHR very late on my way back from the arctic circle. Apart from the very enjoyable breakfast I had with an old friend the next morning, the only thing I remember was jumping in a rental car and getting out of there. I am hoping that there was some data science behind the 5-month filter, but I am starting to doubt it.
If vendors are not going to use good algorithms on my data I am going to take some guidance from Andreas Weigend’s excellent book Data for the People: How to Make Our Post-Privacy Economy Work for You and choose not to provide this particular on-line travel agent with data about my future travels for free by using them again. I intend to do the same with others, and encourage you to do the same when presented with a bad algorithms output.
So to answer some questions I got elsewhere on this. Q: This is not really the result of a bad algorithm just a bad SQL query and some bad copy? A: Technically yes that is probably what this is. I do however suspect that the people who came up with it think of it as an algorithm and so do I. That is because they think that it solves a problem. Also I do not see any publicly accessible data that connects my hotel with West Drayton someone wrote something to find what they thought would be a more eye catching (well at least recognizable) location name associated with the hotel. Q: How could you make it better? A: Simple stuff. Firstly to fit the marketing copy Screen out all non vacation locations, airport hotels, hotels attached to major medical facilities, hotels in business districts etc. Then develop a smarter algorithm that to identify a location. This will not be simple as what you are searching for is the name that will be most recognizable. That may change based on the recipient, the type of hotel or the location.