The Guardian recently reported on the a PWC report into the risk of UK job losses from breakthroughs in robotics and artificial intelligence. Whilst technological unemployment isn’t a new concept and the report and article certainly do comment on job creation too, the percentages at risk in some sectors are quite striking.

The graphic in the article appears to be clean and easy to read at a glance, however I found it a little hard to interpret the “job automation at potential high risk” percentage, particularly as it is not immediately comparable to the other percentage shown, and that the definition of high risk is itself based on a percentage. The article does go on to clarify the value to be fair but it did initially confuse me.

 

Can we automate our own job to do a makeover?

Before tackling a makeover in Tableau I thought it would be interesting to see if the act of doing a data visualisation could actually be automated, e.g. by IBM Watson Analytics. Signing up for a 30-day IBM Watson trial was quick and easy after which I could upload the source data spreadsheet and have it prepared for analysis. Once prepared I could either ask a structured question of the data set or choose from some recommended forms of analysis for the measures and dimensions identified. I didn’t spend long on this, and whilst a visualisation could indeed be produced with little input from me, it certainly had some flaws. For example the total/all dimension value was included which skews the chart.

Here is the viz that was produced:

.

Doing the makeover manually – much better?

Moving on to the Tableau makeover, I really liked a submission by Tamara Gross and have borrowed some of the styling from that. My makeover follows below. I’ve converted the percentage of jobs at risk into a percentage of total UK jobs so that it is more comparable with the overall share of employment – we don’t lose the actual proportion at risk as that is apparent from the stacked bars. However we gain a set of percentages that sum up to the overall headline of 30%. The viz is also available on Tableau Public.

.

But surely we can automate some of the data visualisation process?

Continuing the theme of whether the data visualisation process can be automated I took a look at the Narrative Science Narratives for Tableau Google Chrome extension. This extension is intended to give you an automated explanation of a Tableau visualisation, in “natural, conversational language everyone can understand”. How much time do you spend crafting a headline and narrative for your charts or dashboards? Could this extension really automate that hard work? I took it for a test drive with my makeover.

First up you need to open the link above in Chrome, click ”Try it Now” and install the extension. You then get a leaf icon in the top right of your Chrome tool bar. Click the icon and you should be prompted to register. Once registered if you visit one of your visualisations in Tableau Public and click the icon again you’ll be prompted to select the dimensions and measures that you want to write about:

.

This bit took some fiddling! I couldn’t get the narrative to work without including “measure names” in the dimensions list despite not wanting to write about measure names. Perhaps this hiccup is a reflection of undue complexity in my visualisation though. For simplicity I removed all measures except the percentage share of UK jobs at risk measure.

Once you click next you get a draft narrative and can then tweak various settings, like specifying the word “industry” rather than “entity”, whether the measure is a number or percentage and whether higher values are good, bad or neutral. After changing these settings you can go back to the narrative which will have refreshed.

I selected the least verbose narrative, unlike if I’d written it myself!

.

I’m reasonably happy with the end result. I wouldn’t use it as is, but parts of it would certainly be a great starting point for a narrative. Here’s the start of the text produced:

This analysis measures Percentage share of jobs at high risk by Industry and by Measure Names.

Average Percentage share of jobs at high risk is 4.53% across all 20 industries. Wholesale and retail trade stood out with a very high Percentage share of jobs at high risk value. The individual industries discussed in detail are selected based on the total Percentage share of jobs at high risk across all subcategories.

For Domestic personnel and self-subsistence:

Average Percentage share of jobs at high risk is 0.02% across all three subcategories (representing 0.08% of the total Percentage share of jobs at high risk across all industries).”

I’ve emphasised the aspects I’d keep which nicely highlight a couple of the industries that many in the Makeover Monday community identified to talk about.

Want to find out more about natural language and Tableau? Check out this Tableau blog post if you haven’t already.

 

Wrapping up

In a week where the topic was automation of jobs due to breakthroughs in robotics and AI it was really interesting to try to automate aspects of my own job using AI / analytics tools. I’ve clearly just scratched the surface in terms of what I’ve tried, but my impression is that tools like those discussed can certainly augment the job of data analysis and visualisation. However, there will still be a reliance on human intuition and design skills in this particular field for some time.