Again, buddy, no. That's not how math works. Math does not fit things to curves, math generates the curves. The object of math is the function, the ones that take data sets and fit them to curves are data analysts, for the purpose of predicting future behavior.

Zooming in on a particular section of a curve and observing that it looks roughly linear at that scale does not make the underlying function, which generates that curve, linear. Exponential growth is exponential growth, and it starts before the "knee". It's there the whole time, even when it looks linear.

Every continuous function looks linear when you zoom in enough, that's how derivatives work in calculus. The exponential function looks linear right up until it starts to not look linear anymore. The point of mapping real world systems to functions is to predict their future behavior, not just describe their present status.

The prediction that AI will go exponential is based on the premise of AI generating future AIs. Obviously, as AI gets better, the AIs that it generates will get better. As AI increases, the AIs thus generated increase by a factor of AI^2 . Once AI generated AIs are equivalent to those developed by a human, i.e. AI = 1, the rate of increase will accelerate, since every new model can make an even better model, which can make even better ones, ad infinitum.

No one knows for sure exactly what is going to enable AI to generate powerful AIs, but once it happens that's the knee. That's why it's hypothesized to be exponential. And that has big consequences, which is why people are eager not to miss the signs that it's ramping up.