Does a Machine “See” Our Data?

In my seemingly never-ending quest to grasp the underlying technical components that feed into the various linked data efforts out there, I happened on an interesting talk given at last year’s SymfonyLive* conference. This presentation, by Markus Lanthaler, was called “Building Next-Generation Web APIs with JSON-LD and Hydra.”

If you’re interested in JSON coding at all I would recommend the first half of this presentation for Lanthaler’s nice overview of the differences between JSON and JSON-LD (JSON for Linking Data).

However, what really struck me in this presentation were a couple of slides he used to compare how a human sees and interprets information in an application and how a “machine” might do the same thing. Semantic web developers often talk about “machine learning” or what Karen Coyle has referred to in the context of bibliographic control as “machine actionable” data. The example Lanthaler provides illustrates these differences well.

He uses a simple JSON code snippet. This is what a human might see:

When we look at this we can pretty easily understand what we have in front of us: there’s an ID number; a title; description; a status about whether this is open; a date of creation; and a path to associated comments.

When a machine encounters the same example it might “see” the following:

Every character is a nice hex code. But a computer doesn’t know that “69 64” is the label for the unique identifier or that “74 69 74 6c 65” holds the title. Nor does it “know” what a unique identifier or a title is. I thought it was a nice visualization.

Lanthaler goes on to describe how JSON-LD helps solve this problem by using “context” to map terms to identifiers. I recommend having a look if you are at all interested in this problem.

* Symfony is a PHP framework for web projects, see <http://symfony.com/>