DBTrek wrote:Less hard if a fact-database can be assembled to test against (behold, the internet).
Less hard if a basic fact evaluation algorithm can be written implemented.
You’ll never achieve perfection, but that’s not the goal. The goal is more accuracy and less bias than the current media. Achieving that shouldn’t require too many MIT grads.
Just identifying a fact alone is an intractable problem. You can only assign confidence values. Most of the methods for doing this involve something similar to Google's page rank algorithm, which is essentially hacked by all the collusion and coordinated messaging. Second method is by creating lists of "trusted" sources, but we don't really have trustworthy sources. They are all trying to hide or distort the truth, or outright lie, in order to further an agenda.
Then there is the computational linguistics problem of processing a news article into some semantic representation that is accurate, which can then be verified with the your facts ontology.