It would be a massive cope to say I came close to winning this round, but, truly I believe, there is a bit of a silver lining. After a multi day interaction with one of them which (the very one with leads to this concession), I studied our chat logs, and made some incredibly interesting conclusions. I am in the process of devising a simple 'test' of course that one can apply to accurately determine if you are actually just exchanging messages with machine learning (or AI) software.

Now it's important to realize machine learning is advancing at an ever accelerating rate, every generation it's getting better and better, so one vulnerability today might not be a vulnerability tomorrow. But again, truly, I believe the conclusion I made exposes a vulnerability that does not lie in any algorithm implementation, but instead just the nature of knowledge & intelligence itself, more specificialy: the difference between how a real human brain comes to internalize and express concepts, and how a machine learning software would try to simulate that.

Refer to the attached crudely drawn image.

The vulnerability in the simulated AI human, is ironically, efficieny. Evil people train these simulated humans, using very consciously picked data sets, into being able to connect with people in the most effective and time/energy efficient manner, these are called expressed (or connector) concepts. Through that concept, the AI now has a connection to you, and evil people use this to manipulate you.

Now here lies the difference, with a real person you are connected to through an expressed concept, you can now bond over residual concepts that you both, by neccesity, understand due to the human learning process, these residual concepts lie within both of your natural knowledge depths. AI will have no reference to these residual concepts. For them to be able to, it would be akin to social reverse engineering, which I cannot fathom how ineficcient and uneconomic that could be.