Um, what part of “undercuts the concept of AI at a primitive level” do you not understand? I am claiming to work at a deeper level than AI!
“Artificial intelligence” is merely the public-facing term for the vastly narrower concept of “neural network”. These are glorified pattern matching engines – that is, irregular – as opposed to the humble regular expressions that have been in the arsenal of computer programmers for several decades.
I am able to speak about “undercutting” it, precisely because I have provably created something (the link was give above) in which future-forward computer programs – that make use of function calls to pattern matchine engines, based on the logic of various computationally feasible expressions, no matter how “regular” they might be – may be embedded.
This is all just to say that I’ve created a new kind of system, which admittedly being much like the same old system (Unix/Linux/*nix), the devil is in the details when it comes to issues like distributability, portability, security, and so on.
As a thought experiment, imagine walking into a random VC’s office on Sand Hill Road in Palo Alto, trying to convince him/her that you have a startup that makes use of neural network technology, and has even gone so far as to replace all usages of regular expressions with it. Say you’ve created an application that validates whether a string of text input by a user into an HTML form is a valid email address. Your pitch is that you’ve been able to do your big data magic with a billion real email addresses, using all the latest technologies, so that no one could ever fool your company with a fake email address like “haha@whatever.ur.dumb”. I find it quite obvious that if this VC thought you were attempting to actually be serious, they would start looking at you cross-eyed and would begin nervously fumbling for whatever they had at hand in order to summon office security.
Neural networks are in the form of data structures that may be accessed by invocations of specific functions within systems of various other functions called programs (or applications), and these exist within a larger framework – taking care of lower-level details like memory management and access to peripheral devices – called an operating system.
All function calls have a CS-theoretical “expense” based on factors like algorithmic complexity and memory allocation. So the idea that you would find it a good idea to pay such an enormouse expense, which making function calls into neural networks – as opposed to regular expression engines – would clearly entail, in order to do something as trivial as the validation of the email addresses that users input in your web forms: that idea would strike the VC as perfectly insane. You would clearly find yourself in restraints if you were compelled to cross a certain threshhold of aggression during your meeting.
So my claim is that I’ve done something of a more fundamental nature than what the designers of specific data structures or algorithms can possibly lay claim to.