Running with Scissors

On occasion I come across something that just makes me shake my head. I’m not sure if I’m impressed, or put off by this one by Amazon.

The Mechanical Turk (http://www.mturk.com/mturk/welcome)at Amazon.com is a way to get humans to do work that computers have a hard time doing – like identifying what objects are in a photograph. (Wonder where they got this name from?) In a nutshell, humans perform HITs (Human Intelligence Tasks) and get paid a certain amount for each one, say 2 cents each for identifying the name of the artist on an album cover. These funds are credited to your Amazon account.

Amazon describes this as “artificial artificial intelligence.” This takes advantage of what the human can do easily and uses the computer to organize and run the show – really a great idea in taking advantage of strengths.

David L. Margulius from InfoWorld coined the “Webshoring” phrase to describe this system. This is from his brief article about this:

I gave it a try and performed six HITs (human intelligence tasks). Each, if approved, will pay three cents into my Amazon account. By my crude reckoning, if I get paid for all six, I’d be making about $3.60 an hour (pre-tax). What fascinates me — aside from the creativity on display here and the fact that this is technically less than minimum wage — is that Amazon has figured out how to take the outsourcing trend to its ultimate extreme, which I’ll call Webshoring. It doesn’t have to care which country the worker is in, as long as the work gets done and there is no management overhead. Wow.

Like I said, I’m not sure if I’m impressed or put off.

Maybe after I earn enough credit to purchase the new Tears for Fears DVD I’ll have a better answer for you. :)

Recently I attended Edward Tufte’s course, “Presenting Data and Information� in Arlington, VA. It was a full day course with Tufte doing all the presenting. One nice part of this course is that included in your registration fee, you get 3 of Tufte’s books – The Visual Display of Information, Envisioning Information, and Visual Explanations.

The course started at 10:00 AM, but I got there about an hour early since I rode the VRE commuter train from Fredericksburg to Arlington. The course was held in a large, theatre-type room – probably 500 seats (and by the time the course started, the room was pretty full). For about an hour before the class, Tufte works the room meeting the students – holding what he calls “office hours� where he signs his books and answers questions – well, at least he gives his opinion on what you need to read in his books – not much actual conversation. He also recommends some reading to do in the books before the course starts – I did some of that, but mostly I just browsed the books. I was really impressed with the quality of the books and the tons of illustrations and discussions about them.

Tufte is a rare book lover – in fact he had several 400 year old books that he showed to the audience that illustrated points he was trying to make at various points in the class (one of these rare gems was a first edition by Galileo).

Actually, Tufte is not just a rare book lover, but a lover of all printed pages. He argues that the resolution of the printed page is much greater than any other media available. So he advocates the printed page over and over again during the course (or, he alternately suggests the use of huge computer monitors on everyone’s desk in an effort to match the resolution of paper).

At the heart of his presentation was what he calls the “Grand Principles of Analytical Design� of which I will try to summarize below:

A well designed analytical graphic will:

1. Show Comparisons – show different data types side by side
2. Show Causality – articulate the causes of the displayed data
3. Show Multivariate Data – a pie chart is one dimensional – only shows one kind of data which is not very analytical. Multiple types of data presented together in one graphic can illustrate the relationships between the data more clearly, and thus make the graphic more useful.
4. Completely integrate word, image and data into a single graphic – don’t break the graph, data, images, figures or text onto separate pages.
5. Show the raw data to the audience – this adds credibility by letting the audience see the data that is the basis for your analysis.
6. Contain quality data – bad data cannot be saved by good design.
7. Make comparisons side by side, adjacent in space – let the audience see all relevant materials for a particular graphic on one page.
8. Use “small multiples� of data, not just “cherry picked� data that supports your points.

I agree with these – they are common sense. The real trick is to come up with a way to do all these things when you have a stack of data and a conclusion to present, but these guidelines are helpful in knowing what to shoot for.

I must say that I was a bit disappointed in that there was no real exchange of views during this course – we just heard the Tufte side. I guess it would be difficult to take questions from a room of 500 students, but I was surprised that there was not one question asked or that Tufte never invited questions at any point during the presentation. Of course, you could ask a question while he autographs your book, but I must say that doesn’t quite do it for me. To be fair, he does have a section of his web site called “Ask ET” (http://www.edwardtufte.com/bboard) where questions can be asked and answered dispensed – there is a lot of good stuff there.