One article I read on an AI instance had it being trained using data scraped from the internet going back to 2008. There's a boatload of junk (the old garbage in garbage out idea) on the web so how well could this instance be able to function coherently? I am guessing that the AI decision making is rule based (priority queues?) and heavy statistical modeling. It appears that if one cuts off the data it needs it to would not be able to function - speculating. If it generates database queries like Entity Framework generates database queries, then run.