Thanks for pointing out SICP/learning a non-imperative language. It's very important to learn a functional programming language and Scheme is the best choice because a) it's the language SICP (one of the best computer science books) uses b) it's "pure-enough" (you still have set! if you need it, but tail call elimination is a part of the language specification) as a functional language and doesn't introduce too many other (important, but unrelated) ideas (unlike Haskell or ML which have strict typing and unlike Scala or F# which have OO).
I would, however, disagree with you that it should be the first language: it should be taught before "blub damage" takes place (e.g., before students think only in terms of OO) but it's good to be familiar with programming and data structure to some extent (it helps to understand cons cells as pairs of pointers). C is a great language to precede Scheme: it doesn't force the OO paradigm and teaches students to differentiate between the value of the object and the address of the object (thinking in terms of multiple levels of indirection).
The reason I argue against teaching SICP to complete newbies is because otherwise students tend to think of functional programming as some sort of an academic exercise and not as an approach to solve the problem. I am basing this on actual observations of students that I know at UC Berkeley and at my own school (where a Haskell based course served the same role). Unfortunately in general I find that students don't appreciate the value they get out of a non-"Blub school" ("why am I messing with pointers and recursion instead of writing GUI apps") until much later in their career ("oh, I know how to reduce the state to take advantage of all the cores!").
(1) I use "Blub school" rather than Spolsky's "Java school", as teaching C#, Python or Ruby as sole languages would have the same effect (it's trivial to learn all of these languages once you already know C and Scheme, it's an uphill battle do the reverse).
