The future is fewer people writing code?

>But with new programs like Microsoft Word coming into existence, typography (e.g. formatting a document, setting the margins, making sure the lettering is appealing, etc.) became something everyone could do easily without much thinking.

Correction: Most people do it without thinking. This does not mean the majority of the typography on the internet (and eventually printed) is good or well done.

My LaTeX setup produces fairly good typography for math work, without too much thinking on my part, after setting up the packages, fonts, etc and learning LaTeX, which did took some time. (And ironically enough, writing LaTeX feels quite a bit like coding...) But first of all, deciding to use and learn LaTeX (or some other workable solution to produce good typography, it probably isn't the only one but the one I'm familiar with) requires you think about typography and realize that good typography is needed in the first place.

> In particular, any series of computations that can be represented as a directed graph, e.g. a streaming data workflow, or state machine, is much more easily understood pictorially than textually.

As long as it is very simple. Electronics already have a highly developed visual language for describing their functions - but if what was going on inside every chip was illustrated just as what was going on between chips, it would be entirely unintelligible. Instead, any visual representation is at a particular scale, and well known portions are represented as blocks with cryptic textual notes next to each interface (ACK, EN, V0+, CLK, PT2, HVSD, WTFBBQ, etc.), labels to identify company or type, and an expectation that you know what they do or can find out on your own (and not an expectation that you understand how they do it.)

Anything simple enough to be completely expressed in human-comprehensible pictures should be exposed to the user and modifiable (even if not by using pictures, but forms.) I totally agree, if that's what you and this article are trying to say. My experiences in trying to encode actual human workflows in BPMN have taught me that when using pictures it's harder to express things of any sophistication than in words - because of words like "with" and "each" and "all", "if" and "when," and because of ways things change over time, and because of separate but overlapping/interacting flows that languages can express easily but pictures not so much.

In pictures, that involves looking all over your picture for different things, trying to figure out how to draw lines to them; if the condition is once or twice removed from the object of the search, it involves trying to untangle massive knots with your eyes and memory. Theoretically, that is. What it involves in practice is scrawling words all over your picture (just like in a circuit diagram.) Words that express the same types of relationships over time and type as the picture is trying to express projected onto a plane, words that could be easily expanded to include those relationships and eliminate the 18 types of lines, the 25 types of shapes, the 12 types of shape borders, the 16 color schemes and the long list of rules for connecting them that had to be invented to avoid coming up with a textual syntax.

Isn't doing programming projects much less cumbersome when you hire a professional programmer and express a high-level project spec to him/her? Why cannot a sufficiently advanced machine-learning model behave just like such a professional?

Modern models learn to achieve goals in increasingly sophisticated 3d environments and even learn execute commands in some very limited form of natural language. You could say that the question of achieving more humanlike performance may be "just" a question of engineering and scale.

Programming won't be any easier. It's like claiming that in the future thinking will get easier. I'd claim otherwise: keeping in mind recent trends of "outsourcing" the knowledge many brains just don't have enough building block in working memory to do complicated thinking.

I don't think we can distinguish text and pictures so easily. Look at Chinese, look at Egytian hieroglyphs. Even when "hierglyphics" is used as as a term of abuse of for programming languages synatax -- it ends up pretty popular.

I thoroughly hated LabView when I had to program in it, but it did convince me that a graphical programming language could work -- if only it refrained from doing the cking stupid things that LabView did (such as the strongly typed editor* that would automatically progate the any type error it found, but not your fixes).

In my current C++ work, I would dealy love a graphical tool that showed me where any given value came from, much like LabView does by its very nature.