I submitted a proposal to DARPA yesterday for extending Fortran with an AI extension.
Please be reassured that the chances of my being funded are zero, but I liked the idea.
Absolutely any comments are welcome.
The key technical challenges are implementation of a LISP AI like system in Intel Fortran. It must be considered unusual that this has not already been completed. The PI implemented the first AI system in AutoCAD using Auto LISP. Auto LISP has a very limited function set required for AI code. It took about 6 months to develop the code on an IBM XT computer. This program was first published at an AutoCAD Conference in Sydney in 1988. Modern Fortran can be used for a faster AI system, but this is a challenging but not impossible step in Fortran.
The significant technical challenge in Fortran, once solved, will mitigate the need for interfacing an interpreted language such as Python or LISP to Fortran, which the PI considers will be significantly more klutzy than developing FORTRAN with a few LISP-like extensions. The issue is not the technology, Intel Fortran is a stable language with excellent integration into Visual Studio 2019. LISP is an interesting language that is interpreted, but there is no reason to not take the functional characteristics of the language and use them to control a major Fortran modelling program. Fortran will execute on most computers from Raspberry Pi’s and up at a reasonable speed and multithreaded. This provides a method that will run from PI’s to supercomputers and then able to be implemented into systems, such as AWS.