How do graphing calculators do numeical integration?

Here's a question I don't know the answer to, which Sam Shah asked: how does the TI-83 do ∫_-z^z exp(-x²) dx, or more generally numerical integration? Of course as z goes to ∞ this approaches √π, with very small tails. (The link goes to an old post of mine that unfortunately has broken LaTeX; you can read the alt text for the images. The idea is that in∫_z^∞ exp(-x²) dx, the integrand can be bounded above by the exponential exp(-z²-2xz); integrating this, the original integral is less than exp(-z²)/2z and this is pretty tight. And yes, I know, I should switch to a platform with LaTeX support.)

So you expect to get something near √π for any reasonably large value of z. But if z is large enough -- say 1000 -- then you get a very small value, in this case on the order of 10^-7. Presumably if the range of integration is wide enough, then the integration method used by the calculator doesn't actually pick up the central region where the action is actually happening.