I've never had the luxury of hiring a non-programmer to learn on the job, but I can see how it could work out. But perhaps you mean "technical ability when hired". I certainly am willing to let someone learn a lot of stuff on the job, including a new language.
However I quibble with your last point, or at least say that some jobs require an ability to grok abstraction, recursion and/or pointers and that no amount of will can overcome an inability to do that.
I'll also note that some paradigms cannot be quickly learned by most. I observed that in general for OO when it became "the thing" in the '90s and I suspect functional programming also takes "more than a month". E.g. one thing I'm saying here is that you couldn't necessarily expect a procedural programmer who already knew C to be able to really learn C++ or Java in a month. Yeah, they could learn how to write C in those languages, but learning the OO paradigm took longer.
Then again, the OO paradigms I learned (which did not include any Smalltalk ones) were both very hard (to get right) and I now think they are also very wrong. Learning functional programming might be much easier for those with flexible minds.
The way I found him was that his father was the cleaner at our offices and if I would please give his son one chance. I did and I never regretted it for one second.
Based on this and your other comments, I'd like to hear more about how you came to this conclusion. This isn't the right thread for it, though... perhaps, if you're interested in writing about it, you could make a new post?
However lately I have been wondering about the utility of this approach for GUIs.
Thankfully Rich Internet Applications are cleanly separating these concerns and you can use an OO toolkit for pure UI and a functional language the stand up RESTful services.
The days of bastardized JSP/HTML/TILES/STRUTS/Kitchen sink are numbered, for just that reason. Trying to shoehorn all of these technologies into a server side OO framework was becoming way to cumbersome.
Modularity is a win and OO only superficially encourages it.
Algorithmic complexity, like a graph database or a computer vision system or an AI? Does it matter if they just read a bunch of papers and implemented the formulas? Does it matter if they used libraries like OpenCV? Does it matter if they decided the best implementation path was to outsource it and use a SaaS like DirectedEdge?
Social complexity, like a Facebook or Meetup or Digg? Does it matter as to whether they came up with the technical architecture or social architecture to actually process all those votes and friend requests and event managements, versus just doing what the tech arch or design lead told them to do?
I've implemented device drivers in x86 assembly (a well-trod path and it wasn't very original work) and AI (for a video game, original work but with lots of abstractions provided) and an embedded systems JPEG block rewriter in C (abused the spec to create something very original, but was ultimately only a few dozen lines of code when all was said and done). Which of these is "most complicated" and why?
Any one of the answers you gave there would have sufficed to be used as the 'platform' for further discussion, if you had named those three with the 'Which of these is the most complicated and why?' I think if would have led to a very interesting discussion about your past work for a bit.
Let's take the video game AI as an example. My follow up question would have been something along the lines of 'That's an interesting one, what kind of behavior did your AI govern, and how did you go about implementing that?' or 'How did you protect your AI from getting stuck in a loop or becoming too predictable?', or any one of a bunch of questions about the sort of thing you run in to when you build a video game AI. I'd know pretty quickly if you actually did write a video game AI (I wrote a bunch of video games, and even though I wouldn't class the routines governing the bots as 'AI' I think I can relate enough to the problem domain to come up with some questions) or if you're making it up.
For myself, if someone would ask me that question I'd answer a fully integrated cad-cam system for lathes and mills. That one kept me busy for two years, ended up as about 50K lines of high level code and another 20K of assembler to make it fast. The hardest part there for me was the equidistant operation, it took me quite a while to get it right because I'm weak in the mathematical department, it's actually one of the few times in my programming career that I got so stuck on a problem that I hired a tutor to brush up on the required math.
Another thing I did was to take as much time as I wanted during an interview, I would schedule at most one interview per half day just to make sure that I didn't have to cut an interview short. This made the interview process more relaxed and on the whole gave me enough time to figure out what made someone tick. Most interviews were over very quickly anyway, but every now and then it took a long time to get a feeling for whether or not someone would fit.
One of my biggest discoveries was that when hiring someone good that they would refer their friends, and they would be pretty critical about who they'd refer, screening out a lot of potential duds for me without any effort on my part.
In the Toronto office we ended up hiring four guys that already knew each other previous to joining the company.
This can also work against you, another 'group of friends' applied and I would have been happy to hire just one of them, he stuck with his friends instead.
I once hired a guy that just wanted to learn how to program but did not yet have the ability. This was a risky move, because as a startup you don't really have the time to invest a lot of it in to a person that is not yet proven. But it worked out very well, and Jeffrey actually was one of our most reliably productive people, in part because we gave him a chance that probably nobody else would have.
Sometimes having a will and a drive trumps technical ability.