Shortest path in conformal maps of a surface

My intuition tells me that the shortest distance between two points on the surface corresponds to a line segment joining the two points on the map of said surface, because, the path on the surface is same as the shortest path in the map. However, this turns out to be wrong.

Take for instance, the Beltrami-Poincare half-plane model of $\mathbb{H}^2$, the shortest path between two points seems to be an arc of a semi circle centered at somewhere on the horizon. Picture:

Why is the shortest distance not a straight line in the map here?

Probably I am missing something quite basic, but I just can't seem to figure it out.

Here's one way to think of the issue: Suppose $$ ds^{2} = \lambda(x, y)^{2} (dx^{2} + dy^{2}) $$ for some smooth, positive function $\lambda$. Given two points $p$ and $q$, a path from $p$ to $q$ may "shorten itself" by veering off the Euclidean segment in order to travel in a region where $\lambda$ is smaller.

In the hyperbolic plane, $\lambda(x, y) = 1/y$, which grows rapidly as $y \to 0^{+}$ and which drops off as $y$ grows. As you know, the shortest paths are arcs of circles meeting the boundary $y = 0$ at right angles. Conceptually, these arcs of circles "climb just enough" into a region of larger $y$ to minimize their length. (That's not a proof, of course, just a conceptual interpretation. You're invited to sharpen the interpretation, e.g., explaining why circles meet the boundary at right angles, or why if $y_{0}$ is large the geodesic connecting two points $p = (x_{0}, y_{0})$ and $q = (x_{0} + 1, y_{0})$ is nearly a Euclidean segment.)

More vividly, we could construct a function $\lambda$ that is close to zero in some disk and very larger elsewhere. In the resulting conformal metric, a path might deviate considerably from a Euclidean segment in order to pass through the disk where $\lambda$ is small.

What you might be missing out is that the natural state of Hyperbolic plane is as an abstract space, independent of any embedding/immersions in Euclidean space. The half plane model attempts to bring it in to the Euclidean plane, but there are distortions involved.

As an analogy, we can try mapping the Earth's surface, to a map, a 2D plane. Inevitably, there will be distortions involved. You might be able to preserve some of your preferred properties depending on the map you choose, but it might be that the the straight lines on the globe is no longer the straight lines on a map.

Here, in the Beltrami Poincare Half-plane model, the model preserves the angles, but the straight lines are no longer straight in the usual, Euclidean sense, due to the distortions involved.