/biz/ - Business & Finance

Really fucking dumb question time:
I kind of 'know' it wouldn't work, but I can't wrap my head around why it wouldn't work.

So, let's say I make some functional, but still fairly shoddy AI.
I tell it to look at some specific factors of a candlestick graph and make predictions of its future trajectory.
It follows just one single stock, to lessen complication.

Now,
over time, it will either prove to be more than 50% accurate or less than 50% accurate.
In one case I can always bet with it to win. In the other case I can always bet against it to win.
Literally cannot lose.


Feels like I'm doing one of them mathematical errors here where I "prove" 1=2, but I can't really see where it goes wrong.