January 31, 2005

When Do Assertions Become Facts?

At the core of all of the apps that I have developed is a relational database. I am used to it being there, available to answer any question, and ready to receive updates about the current reality of any given object instance in the system.

Now I find myself developing a web application and semantic data service which are fed from a very large number of SEC filings and news stories. All of the data will be available for others to interpret in their own apps, but I would also like to provide a "current interpretation" for any given object instance.

Poor me, I must have my oracle.

Take the case of an executive's various roles in a corporation: "Chairman and CEO, Former President, Semantic Widgets Division". This is complex data that changes. As time passes, the latest assertions about role will inevitably contradict previous assertions. Such as shareholders firing their CEO.

When should an assertion lead to a re-assessment of the current understanding?

It goes without saying that interpretations of reality are formed in the mind through an ongoing process of re-assessment. We qualitatively compare present impressions with recent impressions. Enough contradiction, and we form a new working state of understanding.

Larger thinking systems such as scientific communities also follow this basic process.

I will use this approach in generating the current interpretation (CI) for changeable objects such as company role. Each CI of an object will have a strength, based on the number and reliability of the assertion sources. Any contradictory assertion rolling into the system puts the CI into a "state of question".

After the strength of a CI has been sufficiently undermined, it is replaced with a new CI reflecting the congruent assertions that caused the shift. This instance is reflects a "state of re-assertion". Example: a WSJ article about the CEO resigning amidst scandal and shareholder revolt would be sufficient to only momentarily put the CI into question, soon to be followed by a re-assert to: Former CEO.

And what if a series of contradictory assertions from reliable sources are received by the interpreter? Well, I'd have to work out how to properly reflect a "state of confusion" for that object instance. :)