When AI has some narrow domain, such as chess, where it can outperform the world's human masters of chess, does it make it a superintelligence or not?
There are three typical use cases for the phrase 'superintelligent':
Something that is at least as smart as a human for every task.
Something that is smart enough to improve itself on a fundamental level.
Something that is smarter than a human at a single task.
Most uses that I see are definition 2, but the other two are also somewhat common. I typically follow I. J. Good and use ultraintelligence for the first definition. (There are lots of arguments that something that fits definition 1 is likely to fit definition 2 as well, but this doesn't seem to be logically necessary.)
Obviously, Deep Blue only counts for definition 3.