Digital Financial Reporting

If you read between the lines of Kalev Leetaru's article Deep Learning Must Move Beyond Cheap Parlor Tricks there is a lot that you can learn.

First, recall that I have pointed out in the past that there are two major techniques for implementing to artificial intelligence:

• Expert systems (logic and rules-based approach): Representing processes or systems using logical rules.
• Machine learning (pattern-based approach): Algorithms find patterns in data and infer rules on their own.

A third approach would be a hybrid of an expert system that also uses machine learning.

Start by reading the concluding paragraph of Leetaru's article:

In the end, for deep learning to move beyond cheap parlor tricks towards solutions that can truly advance society, we must move beyond today’s correlative approaches and simplistic one trick ponies towards algorithms that can actually reason semantically about the world.

That last part "reason semantically about the world" is the key.  That reasoning is powered by things like the functional components defined within ontologies.

Lot's of people are being duped today by software vendors selling smoke and mirrors (i.e. parlor tricks).  I have heard of two projects related to XBRL-based reports and machine learning.  Both were the brainchild of software engineers, not business professionals.  Both were explained to me.  I predicted that both would fail.  I don't know if either succeeded but I doubt it. How do I know this?  Neither involved the hard work of creating the metadata necessary for making the projects a success.

High-quality curated metadata is the answer.  That high-quality curated metadata will supercharge artificial intelligence, including the machine-learning or deep-learning capabilities.  Metadata like this. (This is my best prototype.) Don't misinterpret what you are looking at.  All that human-readable information is also machine-readable.

There are no short cuts folks.  The path that seems cheap and easy might seem seductive, but trust me when I say that the hard, time consuming, expensive path that provides a well engineered, robust system is the only way to get artificial intelligence to work the way we need it to work for things like important functions such as financial reporting.

Don't understand what I am talking about?  I would recommend that you read Computer Empathy, it might help you avoid the snake oil salesmen pushing solutions that will not work.