What about everything that isn't AI?

There is a historical argument I've heard a few times. I unfortunately couldn't find any evidence for or against it beyond some Reddit thread, and a Wikipedia article that cites another encyclopedia. Nevertheless, it's been rolling around my head for the last few days.

Since the 18th century, critics have argued that slavery hinders technological advancement because the focus is on increasing the number of slaves doing simple tasks rather than upgrading their efficiency. For example, it is sometimes argued that, because of this narrow focus, technology in Greece – and later in Rome – was not applied to ease physical labour or improve manufacturing.

-- Wikipedia, Slavery, § "Economics", https://en.wikipedia.org/wiki/Slavery#Economics

The reason I've been thinking about this is that I think the recent wave of AI advancements might have a similar effect on the field of software.

We have been inundated with a wave of tools that are glorified ways to ingest some data and feed it to an OpenAI API. Hell, I work on one. More ambitious advocates speculate about how we could "solve" entire fields of research by throwing LLMs at it.

On several occasions, for instance, I've heard that we should abandon all efforts to make applications work well with screen readers, and instead have a tarted-up version of GPT Vision read information from a screen capture. Essentially implying that instead of making computers that the blind can use, we should make robot assistants to work the computer for them. An idea so bad, I feel cruel for entertaining it. (Have you ever had to instruct someone over the phone or over text to do something on the computer? Would you like to do your day job like that?)

In the process of applying computers to an aspect of human life where they were previously not used, we often discover new things about what we did before. A "paperless office" gives insight on what paper was to the office before, and what it can be now. In developing a system to automate some process, we look at that system from new angles.

In contrast, today's AI implementations largely give us a cheaper version of what we would do without a computer. Where conventional software development imposes structure on a business domain through requirements engineering, systems being improved through deliberate analysis, neural networks can be improved through pure scale -- more training data, more neurons, more GPUs, more sludge.

Write a blog post about the effect AI will have on progress within the software industry, focusing on its potential to hinder or even regress new research, drawing parallels to the supposed effect of slavery on technological progress in history. Detailed. Contemplative. Well-researched. Flowing prose. In the style of...