Knowledge and Symbolic Reasoning
When people gain useable insights into the operation of the universe, such as Einstein's law of Special Relativity, they have encoded those insights into "knowledge" which can be efficiently (if not completely) transmitted to other people, usually by explanation.
Generally this means that they have encoded insights into a reduced symbolic form, like E = MC^2.
When neural networks learn information, it is encoded into vast numbers of coefficients. This form of knowledge cannot be efficiently transmitted to people, and it probably does not match symbolic concepts people already have. So LLM's implicitly learn how different words are used, without necessarily learning or using our general categories such as noun and verb (which may be the most intuitive to us even if they present troubling exceptions to LLM's).
Proponents of LLM believe this is fine. They want AI to deliver solutions. They don't need AI to explain itself. They feel that it is unnecessary for AI to help us learn how to solve problems for ourselves. We stand on the shoulders of AI to do more.
But if AI is letting us solve problems without fully understanding them, then it is making us dumber. The "solutions" that we are thereby creating might best be understood as slop (the word popularized by Cory Doctorow, who is an excellent critic of AI).
When you are a homeless person passing through a soup kitchen, they dish out slop. That's fine because they have limited resources and limited staff and that is the only way they can feed so many people cheaply. But when you are an aristocrat dining in a fine restaurant, or just a person who has enough time to do so, you want a carefully prepared meal, not slop.
Slop is perhaps unavoidable, but generally it is something we should preferably avoid. Ideally everyone should have carefully prepared meals. That's part of a quality life.
Resisting AI means we will not achieve the (alleged) productivity benefits.
But preparing slop makes us dumber. Preparing meals carefully makes us smarter. In the long run, this is more important than being "more productive." It is much better to do less, and to understand what we are doing and learn how to do it better, than just to "do more."
Consuming more slop makes us poorer, not richer. (Don't trust GDP and similar metrics here. What is really most important is not how much we consume, or how much money circulates, but deepening our quality of life.)
We should seek to invent the technologies which make us smarter, not dumber. Only by being smarter can we know and appreciate quality and how to get there.
Therefore, we should seek to build the society that makes us learn more, think, and create, not just dish out more and more slop.
Making people dumber and dumber is the quickest road to collapse of everything.
That also happens to be what you get by mindlessly raising "productivity."
"Higher Level" thinking
Proponents of AI think the sloppiness is fine and it enables us to think at a "higher" (more abstract) level while the AI does the lower level thinking for us.
But this higher level often becomes little more than BS and hand waving.
It is my feeling and my belief that the strongest learning comes from working things all the way through. This is not a new idea. Euclid famously told King Ptolemy I: There is no royal road to Geometry.
So when I build my programs, I do it this way. I think problems through with paper summaries or diagrams first. I think about the different kinds of ways they could be solved, and choose what appears to be the best one. If it proves to have been a wrong choice, I flip to another one before I have written much code, if possible. I build everything from the raw ingredients of my operating system and programming language as much as possible. Only if things appear to be particularly tricky do I look for previous solutions (aka libraries) that I can use. If fairly easy, I even reimplement the parts of those libraries that I need. I rely heavily on built-in language features or libraries including things like associative arrays (aka hashtables) which are capable of dealing with many if not most hard problems.
I know this goes against the grain. From the very beginning of my 39 year career in computer programming I was taught the mantra "Reuse." But I reject that as a general rule for many reasons:
1) Learn (everything) by doing (everything).
2) Programs built upon combinations of even fairly simple libraries can become ever more impossible to fully understand. Often different libraries do not intuitively connect with one another. Then all your code becomes translating information from one library to another--very dull.
3) Copyright, patent, and similar issues.
During the whole process, even before starting to code, I start writing the user documentation as well. This is invaluable in determining the fine details of the interfaces. If something is hard to describe, it's probably not designed well either.
I don't create a 'detailed design' such as including all variables and data structures before coding. That's basically humanly impossible. When I was required to use a formal design process, most people could not actually perform a useful Design Review to being well into the coding process if not nearly complete. As one of my colorful (and PhD) colleagues remarked, "We're supposed to do Design after Coding. I prefer design while coding."
For over a quarter century, I've either written the documentation into the program itself, or straight into fairly simple HTML. I like being that close to the metal. I hate word processing programs. I do all my editing in Gnu Emacs.
I've had some experience doing things other ways. Java programming, for example, is traditionally done with the importation of dozens or even hundreds of libraries, with interactions so complex that fancy tools are needed to work out the ramifications and keep each library installed at a compatible version and all the interfaces correct for that version. General code does little more than call one library after another. This is the pinnacle of the "Reuse" concept. I hated it. It wasn't programming in my opinion, it was dishing out slop.
AI is a vastly greater extension of this.
Now I am very happy to be able to search the web to find code to solve each unfamiliar issue as it comes up. I don't just cut and paste the bits of found (or generated!) code. I read them and figure out how they work. Then I write them into my program. (My post-retirement program MakePlaylist was created exactly as described above, except I haven't written HTML documentation for it, only in-line documentation that gets spit out into help messages and full documents by built in program options. But now I am writing HTML for a far more challenging project: a multivolume book about my life. I can view the result immediate, and also apply simple pre-processing editors I have in mind, along with CSS which I haven't much messed with before.)
DO WE REALLY NEED MORE PRODUCTIVITY ?
Capitalists, oligarchs, and their high priests known as Economists insist that all good things come from increasing productivity. But they do not. Increasing productivity may mean more income for them, but lower quality of life for all as everything becomes slop, prepared and consumed mindlessly.
Now old fashioned machines and even automation may be just fine, when they do the heavy lifting and boring routine tasks for us.
But the creative and thinking parts are not only the parts we most like doing, they are also the parts that make us better when we do them.
Now suppose you are a departmental manager responsible for several projects. You could either have project manager staff for each project, or do them all yourself with AI.
Having a staff working on each project means you can have informed feedback about the practicality of each project. Doing it all yourself means you don't get that essential feedback. It is an error of pride to believe that you don't need that feedback from another person. The end result is slop which lacks humanity and depth, the hallmarks of great art.
It reminds me of the music created by electronic and automated music generation pioneer Raymond Scott. Scott invented machines to do things like sequencing, pioneering devices that became very useful to many musicians. For that he should rightly be honored. But he invented these machines so he wouldn't have to work with other musicians. The result in his own subsequent life's work is very lively music which is also very shallow.
The world we want to construct is one in which each person contributes what they are best doing, which is quite often what they like doing best or something adjacent to it. Turning all jobs into dishing out slop is exactly the opposite.
What we want to do is the thinking and creative parts, and have machines do the heavy, awful, boring, and repetitive parts. That's what previous automation has done.
And in many cases still and forever, the best machines are machines custom built for their purposes.
In both shirts and intellectual products, hand made is best and always has been.
And it makes us better to make such things, at least so much as we find our calling in doing so.
Limits to Growth
Creating the supercharged high value worlds where most everything is done by AI that people just command, and yet everyone has a job doing something more to their liking commanding that AI, can only be possible by large amounts of growth, the kind of growth nobody is planning for anymore anyway. It seems more that people are simply being laid off rather than retrained for even more creative positions.
We need to scale back our assault on the environment, including especially our consumption of fossil fuels. But even generating electricity the very best and most environmentally friendly ways, with wind and solar, still has considerable environmental impact. We need to use as little electricity as possible. As little of 'everything' as possible in fact, except our creative minds.
Instead, as everyone knows, data centers of obscene size are being built with obscene levels of consumption of water and electricity--which were going to be if not already scarce anyway, and scarcer still going forwards.
And that's not even counting the environmental cost of the 'value' AI may be adding to society, if it were keeping everyone employed at an ever higher level. Im not counting that because it's unlikely.
Even just the Data Centers being built are only going to bring on the collapse of everything faster, let alone the vast future of data centers planned and/or approved.
The environmental cost is another problem that AI can't solve. Though if it were intelligent and free thinking it would tell people not to build any more data centers for a while as part of the solution.
