Listening to Hilary Mason from bit.ly talking about machine thinking and its origins based around the presumption that ‘the essence of humanness, the ability to think and have an identity can be described so precisely as to be rendered on a bunch of tubes’.
Mason quotes Tom Mitchell saying that ‘Machine learning is the study of algorithms that continue to improve as the amount of data increases’ but also neatly demonstrates how easily data can become meaningless without context, what is MS? Microsoft, multiple sclerosis, or mass spectrometry (or one of several other meanings)? Is Paris in France or in Texas?
The same issues can arise in the sorts of recommendation programmes used, for instance, by Amazon or the movie site Netflix, which try to recognise customer behaviours to recommend future purchases and use similar purchase patterns from other customers to make suggestions for next steps. This can often backfire when the choices made are not an apparent good match. She mentions at this point the pre-emptive lawsuit that Netflix was faced with around privacy issues (as it had stored data concerning historical customer data to re-use) – this clearly an area that relates to the use of learning analytics in the context being covered in the #LAK12 mooc.
Mason looks at the application of models and explores their limitations as a result of being based on probabilities and observations – being able to predict behaviours that we have not yet seen based on behaviours that we have – she can even crack a decent joke about Bayesian theory. As she works for bit.ly, Mason is naturally keen to show the benefits of being able to predict what new content a user may be interested in having observed their previous browsing history.
This hour-long video was largely very interesting although often out of my knowledge comfort zone – fundamentally what it aimed to show was that there is much that we can learn from past behaviours that can inform our future choices.