It means that every algorithm, given enough (potentialy infinite) memory, can store precomputed output values based on input values in an (indexed) array, and thus convert the algorithm to O(1) array lookup. It's a trick as old as computing itself, predating computers by centuries. Scientists always used tables of precomputed values to get results quickly without doing actual computation.
The Wikipedia article covering this topic is https://en.wikipedia.org/wiki/Lookup_table .
BTW, many real-life algorithms (CRC comes to mind here) do it too, especially when speed is crucial (real-time applications). Note that the amount of memory needed to store the precomputed values is directly related to expected precision and amount of input variables.