You're right, I'm still not as much of an expert as I would like to be in my field, and I've been doing it since dirt was rocks :0

Pattern recognition was exactly my first objective, but I only had kilobytes so language was out of the question.

The machine would ease the burden on me as far as the pattern recognition was concerned, but I expected to continue to do all of the judgment myself for the foreseeable future until someday when I had more powerful computers.

Very helpful to separate the recognition from the judgment, but I still found it best to perform both simultaneously. That was what I would have wanted AI for, to do both if I could get it good enough for reliable judgment one day.

>The code was a transcription of that understanding. Acquiring the understanding was the job.

Well in the mid-1970's almost nobody had the title of Software Engineer or software product manager compared to today.

But "Coders" as a job title were as common as the professional "Programmers", they worked hand-in-hand. The Coders were the ones operating the keypunch machines which took the manuscripts from the programmers plus data from the users and turned it into code on the punch cards so it would quickly run on somebody else's out-of-reach machine without tying it up for very long. Those colossal remote mainframes were expensive. But there was nothing else so what were you going to do? It sucked to be tied to some huge data center though before you could do any programming at all :(

If AI makes some coders of the 21st century feel like they are being bumped down closer to keypunch operators than ever imagined, serving a massive machine they will never be able to own, that would not be too surprising.

>You can now produce the software without ever building the model, and that breaks an assumption the whole profession was organized around.

This is exactly what I said back then, but with reverse angst. I was observing all the professions up to that point in time, almost all of which had nothing to do with software or computers at all. Since actual stand-alone "software companies" were still rarer than hen's teeth. With desktop computers beginning to take hold, Bill Gates and backers like that put maximum effort into getting software recognized under copyright and not just patent coverage.

Next thing you know there were two handfuls of software companies which is still pretty insignificant, but that is exponential growth and it can be quite tempting.

That's when I realized if that keeps up, people leading the first wave of computerization are going to start producing "the software without ever building the model", especially with a lot of professions that require decades of domain expertise which can sometimes be more infinitely rewarding to leverage with each accumulated decade.

If it was going to take decades anyway, might as well do it. The idea that computerization was going to take place using purchased software, without so many companies having their own home-grown programming expertise from the beginning, is what breaks the assumption that all other professions were organized around! That in itself was going to leave a lot of money on the table.

>The domain expert had no equivalent path, because learning to build reliable software is years of work they were never going to do.

I had already spent years learning to build more reliable code than you could generally get from popular software, because reliability is what I needed more than anything. I wasn't going to spend the years of additional work making my frameworks into things that even resembled commercially appealing products though. If I ever decided to go that route, there were going to one day be high-performance teams having well-honed experience in that area if nothing else. Gave me more time to concentrate on other things.

Even though I had a teenage head start in programming itself similar to Gates during the same 1970's, by the mid-'80's it was not only programming but AI too was plainly going to only get more popular faster than I could keep up. I had already started to do a little ML a few years earlier which really worked, but it was expensive and "nobody" around here could afford it once the oil crash kicked in. It was plain to see that "all I had to do" was wait and any domain expertise I could develop in natural science could be leveraged later on if AI gets good someday. I even explored the neural networks of the 1990's but that wasn't going to cut it either.

And here we are.

If I got a wild hair and decided to launch a (non-mission-critical) "product" at this late date I guess I could consider the use of an AI agent not much differently than I would have engaged with a software team once they became an entity themselves. Same business model from my point of view over the long term.

The most artificial thing in the whole timeline is copyright which lots of massive sand castles have been built from, and that is where this language-model-approach to AI strikes the weakest foundation so far, as the tide finally rises too much to be denied.

One big artificial thing gets ugly when confronting another big artificial thing as they're vying for king of the artificial mountain :\