I get the sense that you're trying to mask the simplicity of predicting the next-most-likely-word after training your app, ala markov chains, under the guise of

I get the sense that you're trying to mask the simplicity of predicting the next-most-likely-word after training your app, ala markov chains, under the guise of "magical AI." Providing an error threshold when it spits out the wrong response to a phrase seems to be worsening its natural ability as well.