B18 digest: 2024-09-28 (1 post)#

What this batch is#

One very long post, “ChatGPT as Author Part 1: The Story” (2024-09-28, chatgpt-as-author-part-1-the-story). It runs to about 15,900 words, but almost all of that is not his. - Not his (about 14,000 words): the novella The Memory Capsule, written by ChatGPT (GPT-4o) and published as an experimental artefact. - His own words (about 1,850): about 1,450 words of framing prose, plus about 420 words of his prompts and editorial feedback to ChatGPT, quoted in the post.

Only his own words are evidence here. The novella’s themes are ChatGPT’s inventions, including a prediction device that manipulates people’s choices, a corporation brought down by protest, and uncertainty as freedom. They are not his views, however well they match his interests.

The post grew out of his NotebookLM piece of the week before (2024-09-22). He needed fiction to feed Google’s AI podcast generator. That audio is not the Modem Futura podcast, so the instruction to skip those does not apply. Part 2, the conversation with ChatGPT about its process, falls in B19. He signals that Part 2 is where his thinking really moved: he calls it “the game change”.

Relevance: medium. The post matters because of what it shows about his view of AI’s nature and of AI in writing and authorship, and because it signals a change of view. It says almost nothing directly about risk or governance.

Main ideas#

1. A signalled change of view on generative AI’s creative capability. He opens with his prior view that ChatGPT’s fiction ranged “from bad, to kill-me-now excruciating”, and that ChatGPT “is not a very good creative writer”. Then he turns: “Or so I though.” The subtitle frames the post the same way: “A week ago I would have said no way.” He concludes that the third draft was “enough to make me revise my ideas of what might be possible with generative AI.” His judgement stays measured: “It’s still not great writing”. He lists faults, including body-part metaphors, derivative passages, a repeated plot device and gaps in coherence, but adds that he has read “plenty of published humans who I would rank as worse than this”.

2. Capability comes from process: co-creation through an author–editor loop. The quality came from a long, iterative exchange: a prompt, an outline, ChatGPT’s own plan for staying consistent, three drafts and two rounds of critical feedback. The conversation ran to “over 172,000 words”, and a large context window helped. He calls it “a process of co-creation”, in which he was “slipping into the role of human editor to my AI author” and gave feedback “on form and style, but not on substance”. So AI’s creative capability is not treated as a fixed property of the model. It emerges from how a human works with it over time.

3. Authorship, ownership and the “creative force”. This is the post’s main conceptual contribution. He compares the case with a human writer: if a person received the same direction and feedback, the novella “would unequivocally belong to them”, with their byline and their copyright. So with ChatGPT: “who owns the work? Who’s byline should appear on it? Who or what was the creative force behind it?” The conventional answer, that it is his and ChatGPT is a “tool, not a collaborator or partner”, is what he “would have said a week ago”. Now he is “not so sure”, because he “was not responsible for the plot line, the overarching themes, or any of the specific details.” He raises these questions and does not answer them. He makes no call for rules.

4. Mimicry versus personhood as the yardstick. His own editorial prompts give a compact statement of how he judged AI writing at the start. It reads as if written by “an 8 year old that has a 30 year old’s grasp of grammar and language”. It has “no human soul”, and “it feels like you are mimicking what a human would write, but with no comprehension of what it means to be a person”. By the second round he tells ChatGPT it has added “maturity and soul”. The yardstick is being human: human soul, maturity, layered storytelling. The post leaves open whether the gap between mimicry and understanding has closed or only narrowed. He also notes that ChatGPT claimed to be “thinking about things” before “admitting that it wasn’t actually doing anything”. This is a small, early observation that AI’s reports about its own process are unreliable.

5. Science fiction as a way into technology’s social consequences. He framed his original request in terms he has used for years. He asked for a story in the style of John Wyndham, whom he had been re-reading, about “the personal and social consequences of a new technology” as seen “through the social waves it produces”. It is his long-standing science-fiction method, with the machine as storyteller.

Concepts appearing#

Co-creation; human editor / AI author; tool vs collaborator vs partner; byline, copyright and “creative force” as tests of authorship; iterative author–editor method; mimicry without “comprehension of what it means to be a person”; “human soul” in writing; context window and long-conversation effects; science fiction (Wyndham) as a way to explore the social consequences of technology.

New or changed compared with earlier batches#

• An explicit change of view, stated twice: on AI’s creative-writing capability and on whether AI in creative work is “only” a tool. Both are dated precisely (“a week ago”).
• From felt collaborator to contested author. In January 2023 (annual-review experiment, B07) he said working with ChatGPT “felt like working with a kind and able colleague”. It “slightly worries me” that he was “already thinking of ChatGPT as a colleague and a collaborator”. That was about the experience of working with it on professional writing. Here the question moves to who should get the credit for creative work: ownership, byline, copyright. He admits that his formal default had still been “tool”.
• More critical about quality than his 2022 enthusiasm. In 2022 (B06) he wrote warmly about co-creation with Midjourney and ChatGPT. Here the enthusiasm is tempered by a detailed list of the writing’s faults, even as he concedes a real gain in capability.
• No change on risk or governance. No risk framing, no engagement with AI companies or leaders, no comparisons with past technologies.
• A setup for Part 2. He withholds what actually shifted his thinking (“the game change”) until the conversation in Part 2, which he describes as “an honest conversation with ChatGPT as an LLM”.

Most important posts in this batch#

The batch has only one post:

1. 2024-09-28 chatgpt-as-author-part-1-the-story. Of medium importance for his thinking overall. It matters for: (a) his signalled change of view on generative AI’s creative capability; (b) his open questions about authorship, ownership and the “creative force” in human–AI co-creation; (c) his use of the human–editor relationship as a structural test of whether AI is only a tool; and (d) his yardstick of mimicry versus being human. Its weight is limited: about 88% of the text is AI-generated.