Machines really are able to act more intelligently due recent discoveries in mathematics!
Paul Phillips’ presentation was fascinating to me as I could instantly see so many connections in how the application of decision automation can significantly improve multiple areas of e-business and online learning.
Much of the history of communication has been finding ways to do the same things we have done in the past – but more efficiently. For example, the focus is usually on how to get information to humans more quickly and efficiently. But the problem is humans only have a certain “bandwidth” – and we now face information overload (e.g. Paul estimated that 70% of his work day is spent responding to emails).
Paul pointed out how certain domains allow for the computer to receive information/data and make decisions more effectively and quickly than if a human needed to receive, interpret, and execute action on the data. Obviously certain decisions won’t and should never be made by a machine, but with certain things it just makes sense to let the computer do it faster and more effectively than any human could.
The domain he picked for this lecture is online advertising, although it could apply in a number of areas. It is true that if you go into a store or restaurant multiple times, most likely the employees there will remember you and customize their service to your needs. Most web sites do not…yet.
The mathematical algorithm (much of which has been invented in the last 7 years) which makes this possible allows the computer to take into account huge amounts of data (6 “buckets” of variables) which are used to predict and display the most likely messages or creatives to lead to conversion (in the case of online advertising) while at the same time constantly testing (and making alterations based on the) risk that something else might be better. In other words, the computer is constantly using the what it predicts will be the best thing in the immediate situation (specific person at a specific time in a specific place with a specific known history) while simultaneously measuring and monitoring the risk that it might be wrong and something else might be better.
The results are measurable too. There have been astounding uplifts in conversion rates of certain sites that have already employed this method (sometimes over 100% uplift) when compared to random selection of messages/creatives.
I’m currently in the process of studying the mathematics and methods behind decision automation and will continue to post what I am learning…