In fact, I’ve been involved with ‘artificial intelligence’ since about 1990, when I developed Wave 3 AI – expert systems. Wave 4 is the current incarnation. Still no ‘intelligence’ to speak of, but marketers and hypsters love the term. Perhaps in Wave 5, the name will finally be correct.
Aside from my historical connexion, I want to share how I am using AI in my writing – in this case, ChatGPT 5.1. I’m not going to give much backstory on the setup, but I’ll point out some internal process logic.
I have completed the manuscript for a Language Insufficiency Hypothesis, so I have been sharing screenshots of each page – usually a spread – and using the GPT as a second set of eyes. I’ll feed it an image and a request, in this case, to find key terms so I can capitalise and italicise them appropriately. In this example, this is the ending paragraph of Chapter 6.
This first screenshot is an example of output. As is evident, it was looking, among other things, for the capitalisation of the concepts of Presumption Gap and Effectiveness Horizon.
Notice the iconographic language is a bit off. The red X is a bit out of sync with the rest of the message, which says the entry is already correct. So, two instances; no problems. Next.
In this message, I warned that it was OCRing the screenshots but not retaining the formatting, and which is a reason I was sharing images over text.
What’s interesting is that it informed me that it would now treat the image as canonical. In Image 3 (above), it’s engaging in introspection – or at least self-dialogue. This is evidence that it (1) reviewed the results of the OCR, reviewed the image (as an image), and (3) compared 1 and 2 to arrive at the conclusion that the OCR had indeed dropped the formatting.
It wasn’t enough to inform me that everything was ok or, better still, not to bother me with noise since it was already in good order. Instead, it’s like an autist talking to itself. It reminds me of Raymond in Rain Man.
Image 34 (next) is the last example. Here, the OCR confounds rendering Horizon as Hπrizon, and then points out that I should avoid the same mistake of viewing o as π.
Thanks for the advice. I was losing sleep worrying about this possibility.
Conclusion
This is obviously a late-stage use case. I use GPT for ideation and research. Perhaps I’ll share an example of this later. I might be able to review my earlier notes for this project, but it was started years before the latest Wave arrived.
Philosophics 