The project is the product of a few things coming together quite nicely. First, you’ve got the all important natural language processing. Now, more than ever, computers’ ability to parse natural text is increasing at a rapid pace. You’ve got chatbots that are more and more capable of maintaining actual conversations, and things like Siri, which seem to be working passably well and can only get better. This would let such a program go through any block of text and identify the things that are — or claim to be — factual statements as opposed to those that are matters of opinion.
Second of all, Schultz is working in partnership with a great website called PolitiFact. PolitiFact is a site that aggregates statements made by politicians, checks the actual numbers used in their claims, and rates the factuality of said statements on a sliding scale of “liar, liar, pants on fire” to truth. The idea is that Schultz’s program would identify the statements in a written work that are claiming to be facts, and index them with the data on PolitiFact, reporting the results back to the reader.
When the project is done, Shultz intends to make it open source, so it would probably make it into the real world as a user-side plugin of some sort that could be used to get some background on whatever the user is reading. It’s an interesting idea, but it still has a long way to go. For instance, it seems that, for now, the detector would be primarily concerned with political number checking.
It won't quite tell you why your significant other needed all that money (yet), but it's an interesting step forward in the development of the machine mind.