I read recently that a noted software developer fired many hundreds of Junior Software Engineers and replaced them with AIs. Not just AIs, but AIs now to be monitored, supervised by the remaining Senior Software Engineers. So, gosh, what could go wrong? Well, one Senior Software Engineer pointed out as he was heading out the door (He quit, obviously.) with the question: “Where do Senior Software Engineers come from?” Obviously they come from the ranks of Junior Software Engineers, who are tasked with less important tasks, to cut their teeth, so to speak. They need to work, make mistakes, correct them, etc. There is much to learn and without a large pool of Junior Software Engineers, there won’t be any Senior Software Engineers to do their work, so this guy saw the writing on the wall and hightailed it out of town.
The “leaders” in so many tech firms are now business types, no longer the tech types who used to run these places, you know, like Steve Jobs and Steve Wozniak, Sam Altman, Larry Ellison and even Bill Gates (to some extent). The business guys look at the salaries “saved” when the Junior Software Engineers were “made redundant,” “let go,” “terminated,” etc. and think that is a net savings with an increase in productivity thrown in. Except going out the door with them is much institutional knowledge, even a bit of wisdom, and losing that Senior Software Engineer, is an even bigger hit to those categories. They might not even see those “Juniors” as being “Seniors in Training” or as resources of institutional knowledge, etc.
And when it comes to writing software code, AIs shine, but there is another problem come up. You may be aware of what is called “AI slop” sometimes referred to as “AI Hallucinations” or AI Bullshit,” but output from an AI which is severely flawed, let us say. And AIs are being “taught” to train themselves, so fairly soon, they will be raking through the slop from other AIs, even their own, and the problem magnifies. We were alerted to this problem when photocopiers were invented. We learned that if you copied a copy of a copy, etc. the copies soon became unrecognizable. Then when computers came along we were taught the same lesson again, with “lossy” file formats such as JPEG. When JPEG images were saved over and over, they too became “muddy” if not downright unrecognizable.
Now, in an experiment AIs have been fed their own output, then re-fed and re-fed it and, guess what, what you get is bizarre and of no use whatsoever.
So, if we give over training (the expensive part of AI enterprises) to the AIs themselves, what can we expect as results? You know, from AI medical advisers, and that sort? Hello, SkyNet!
Uncommon Sense 