Machine learning is comprises many methods not one. I am not the person to go into the method. And my purpose is not to advocate for machine learning. I want to draw out what the machine says about how we learn.

So I am going to give just a couple of machine examples from the work on the Care4/Functional Family Therapy data. I want to suggest it fills gaps in the era two approach described above. And I want to suggest it encourages we humans to learn differently.

The first example is called Natural Language Processing. It looks for pattern in speech and writing. We have seen examples in this series already, for example the n-gram analysis of how words like ‘place based’ have been used over time. Most of us are used now to ‘word maps’ beloved of newspapers and PowerPoint presenters.

The Word Cloud of a Town from an article in The Atlantic

In our analysis, we asked the machine to consider ad hoc notes made by Functional Family Therapy practitioners before, during and after their meetings with families. Here is typical entry:

E. and his mom report no problems in their relationship or in his behaviour on home visits. According to both they are ready for him to come home and see if he can apply the skills he has learned to his home in a more permanent way. E. and his mom discussed where he wants to go to school and the reasons for each of those possible options. They are working to generalise to the outside environment.

In essence, we were asking the machine to look into the black box described in Input 34 above. The machine looks at every detail, and it seeks absence as well as presence of words. It might estimate, for example, that ‘King minus Male Plus Female Equals Queen’.

We used a form of Natural Language Processing called ‘topic analysis’. The machine sifts through the words and sees how many come together to form what looks like a ‘topic’ (a pattern).

A typical output looks like this:

In this solution, the machine found 13 topics or patterns of words. The graph to the left indicates how different the topics are from each other. The bar chart to the right gives the frequency of words in topic 1 (red line) and in the overall analysis (blue line). If the analysis shifted to topic 3, different words in different frequencies would appear in the bar chart.

So far pretty meaningless? Yes. And that in a way is the point. The machine has found pattern. (It found other patterns as well as this 13 topic solution). Maybe there is no meaning in topics. The machine cannot figure that out. Only a human, what is called the ‘content expert’ can do that. The machine depends on collaboration with the human, and vice-versa.