Today's web is a collection of applications that largely provide a frontend for browsing data. The applications and data they contain are silos: there is no easy way to separate the data from functions and compute across datasets. Every application must (re)invent its own UI for querying and displaying data.
But if the web is actually a collection of datasets, why don't we have a web browser for consuming and interacting with arbitrary structured datasets?
We can model most popular sites (HN, Instagram, Twitter, Amazon etc) as a collection of hyperlinked JSON records. Let users adjust how these records are displayed. Provide a universal way to query and navigate any dataset and invoke associated functions (eg the upvote function for an HN post).
Full separation of data and functions instead of application silos is necessary to achieve general AI compute in the future.
Example: can you email Mark a summary of the top 5 most popular HN articles 3 days before our meeting?
Not quite. The semantic web focuses on normalizing all data into a single ontology, which is great for computers but does nothing to standardize the user interface side of accessing data. Needless to say, it also hasn't gotten far.
That's a good point, and I completely agree that it hasn't gotten far.
I do think that if it were to be embraced as a more common standard on websites, building a more standard UI on top should be _relatively_ simple as each websites underlying data would _hopefully_ be structured the same.
Today's web is a collection of applications that largely provide a frontend for browsing data. The applications and data they contain are silos: there is no easy way to separate the data from functions and compute across datasets. Every application must (re)invent its own UI for querying and displaying data.
But if the web is actually a collection of datasets, why don't we have a web browser for consuming and interacting with arbitrary structured datasets?
We can model most popular sites (HN, Instagram, Twitter, Amazon etc) as a collection of hyperlinked JSON records. Let users adjust how these records are displayed. Provide a universal way to query and navigate any dataset and invoke associated functions (eg the upvote function for an HN post).
Full separation of data and functions instead of application silos is necessary to achieve general AI compute in the future.
Example: can you email Mark a summary of the top 5 most popular HN articles 3 days before our meeting?