[Haskell-cafe] transcoding - Haskell?!

[Joachim Durchholz]

Am 12.06.20 um 19:44 schrieb Branimir Maksimovic:
> It would choke on assembler as well. Digital computer hard AI is 
> impossible, because there is no
> 
> algorithm for making algorithms. Mathematicaly impossible...

You know you *can* work with undecidable algorithms. You put a resource 
constraint on them and let them declare "don't know", which then can be 
handled appropriately. There's even software that works like that in 
practice, nowadays - register allocators in compilers, query plan 
optimizers in databases, travel route optimization software, sheet 
cutting software for irregular shapes, various types systems, proof 
exploration software, etc.
These work by trying out various options and returning "don't know 
whether this is the optimum, but it's the best I could find with the 
available resource limits", and with good heuristics, it's still good 
enough.

So the undecidability is not important here.
You need a good heuristics.
And AI can be good about heuristics.
So this aspect is not the one that's messed up about the idea.

Regards,
Jo