A bit random, but...
I wonder whether machine learning could help decide which types and primitives (eg whether booleans should exist instead of mere single bits; whether a 'where' function should exist) would be a promising approach for designing "restricted instruction" languages like APL, J and K.
If you have say n slots for concepts, where n is equal to the number of ASCII characters to which to assign an abstract idea, you could aim to minimise "# characters required" times "# secs required" (weighted as appropriate) given a benchmark suite of problems.
I wonder whether machine learning could help decide which types and primitives (eg whether booleans should exist instead of mere single bits; whether a 'where' function should exist) would be a promising approach for designing "restricted instruction" languages like APL, J and K.
If you have say n slots for concepts, where n is equal to the number of ASCII characters to which to assign an abstract idea, you could aim to minimise "# characters required" times "# secs required" (weighted as appropriate) given a benchmark suite of problems.