A large part of the answer is historical(80 column text is derived from the IBM punchcard standards of yesteryear). The remainder is in information theory: we tend to use textual language because the data is a nice compromise between "compressed" and "editable". While it's true that "a picture is a thousand words", those thousand words are generally not abstract concepts like "homomorphism" or "ideology", and yet you can type those words with a few characters.
What you suggest is possible to some extent, it's just hard to standardize, much harder than gluing more stuff on top of the existing systems. You're looking at something that you could build a career on, if you find the right approach. There a lot of companies and projects that have tried to research moonshots and come up with good demos but bad production environments.