In a comment thread at n-category cafe, John Baez has linked to the electronic version of a big bowl of ice cream. Investigating the link in his comment, I came across the web page of Francis Sergeraert, who has linked to his papers and talks.
Sergeraert and his collaborators have pioneered a program of computational algebraic topology, and it is amazing what they have already acheived. For example, they have developed effective versions of the Serre and Eilenberg-Moore spectral sequences.
These kinds of algorithms exert a powerful hold on my imagination. When I first tried to learn commutative algebra, I found much of the subject impenetrable. Then later when I learned about GrÃ¶bner bases, I suddenly found everything I found hard to understand became easy to understand. Now the GrÃ¶bner basis algorithm is too slow to implement by hand other than toy examples, but having effectively computable toy examples was enough for me. Commutative algebra textbooks are full of toy examples, but my suspicious unconscious mind was sure that they were tricking me, and that the toy examples were not to be trusted. Learning how to compute new examples allowed me to shut my unconscious up.