Can artificial intelligence classify textual records?

Question

I am a records manager and I am being asked if I recommend Office 365. I'm having a hard time making a recommendation because I am missing an essential piece of information: can Office 365 replace the manual process of placing records into categories based on organizational function? It is important that this is done accurately, because the category determines how long the records should be kept before they are irretrievably destroyed. James Lappin seems to say that yes, Office 365 is underwritten by Project Cortex, and it is capable of doing this.

My sense is that artificial intelligence is not yet capable of determining a conceptual category for records. For this to be true, a machine would have to replicate a complex human process: reading a free-text, free-form document; identifying the relevant pieces of information in a document, while ignoring others to determine what the document is "about"; then taking the answer of what the document is "about" and matching it to a predefined set of major organizational activity.

Are there any AI experts who can comment on how realistic it is to expect Project Cortex to do this?

Answer 1

AI can categorize documents very accurately. It is not a new application but in the last year the accuracy of the underlying algorithms such as text classifiers, and language models in general, has significantly improved. There are applications of language models which now surpass human performance. Microsoft is one of the leaders in this area. For example, see their Turing Natural Language Generation (T-NLG) language model: Turing-NLG: A 17-billion-parameter language model by Microsoft. The T-NLG was developed in a research group called Project Turing which works on AI tools for Microsoft's products including Office-365. My guess is this same research group makes the AI for Project Cortex.