|
|
|
|
|
by _delirium
5437 days ago
|
|
That's been an idea since the 60s in various forms, but tractability tends to be a big problem. The classic approach is to randomly generate Lisp code trees, guided by various heuristics or randomized-optimization algorithms (e.g. genetic programming). There have been some successes, but typically smallish algorithms for well-defined problems, rather than "normal" full-sized programs. Some other attempts include: inductive logic programming (automatically infer logic programs from desired example output), partial programming (use statistical machine learning to fill in incompletely specified behavior in a program), and renewed attempts to use exhaustive or heuristic search where the search space is constrained by strong modern type systems (e.g. MagicHaskeller). |
|