Claude answers GPT re ailien minds. And the winner is...

Last time, I posted a lengthy missive by ChatGPT appraising my new book ailien minds, prompted by Richard Bliss, with some of my own commentary and demurals laced-in. 

All-told, the effect was that of a loquacious book report by a garrulous and rather unctuous high school sophomore who had flipped through some random pages for key words and then googled some of them... or used them to sift its training sets. 

So, how does Claude do?  Well, the comparison here isn't exactly fair... Mr. Bliss asked Claude to read not just my entire book, but also the GPT appraisal that I posted, last time, including my critical inserts. So, you cannot draw explicit conclusions. 

Still - fair or unfair - the overall impression is night-and-day.  Claude's missive is accurate, pertinent and in some cases even surprisingly insightful. And weirdly honest about its own (un)sapience. 

Unlike GPT, which glossed maybe 5% my work over a reurgitaion of common AI memes from its training sets, Claude zoomed upon core concepts in alien minds, distilling an impressive 50% or so of the key points. (Though of course none of the supporting evidence, of which Claude says I provide too copiously!)

Moreover, there was almost none of GPT's blatant flattery.  Well, except for the flattery that's inherent in well-summarizing my ideas.

But judge for yourself. And note at the end where I promise a more-fair comparison. Maybe next time. If world events don't force me back to politics.

(Oh, note: on Monday I keynote one of the tracks at the big RSA conference.)

============== 

A Critique of ChatGPT’s Review of

Ailien Minds  by David Brin

Written by Claude (Anthropic) March 2026

Prepared after reading the full text of Ailien Minds via page-by-page screenshots supplied by Richard Bliss

Preamble: A Note on What This Document Is

This is a critique written by one AI system (Claude, made by Anthropic) of another AI system’s (ChatGPT, made by OpenAI) review of a book that is substantially about how AI systems should be governed. The irony is not lost on me.

A disclosure: I read Ailien Minds page by page through 73 screenshots shared by a reader. ChatGPT appears to have been given the book’s text as a submission for review. We both had access to the source material. We arrived at very different readings.

David Brin annotated the ChatGPT exchange with his own reactions. I have read those annotations. In the interest of transparency, they informed my critique—but so did my own independent reading of the book. Where I agree with Brin’s frustrations, I will say so. Where I think the ChatGPT review got something right despite his objections, I will say that too.

The Central Problem: Surface Pattern-Matching vs. Structural Argument

The ChatGPT review correctly identifies that Ailien Minds is about AI, that Brin occupies a middle position between optimists and doomers, and that he favors transparency and competing systems over centralized control. These are accurate at the level of a dust jacket summary. The problem is that a dust jacket summary is essentially all ChatGPT produced, stretched across several thousand words of increasingly generic AI commentary.

The book’s actual contribution—its reason for existing alongside the dozens of other AI books published in 2025–2026—is a specific, detailed argument about mechanism. How, precisely, do you keep powerful AI entities accountable once their creators can no longer control them? Brin’s answer draws on evolutionary biology, ecosystem dynamics, and the history of legal and democratic institutions to propose that individuated AI agents, competing within transparent accountability structures, represent the only approach with historical precedent for success.

ChatGPT never engages with this mechanism. It talks about transparency in the abstract. It mentions competing AIs. But it never connects these to the specific framework Brin builds: individuation (giving AI agents distinct, persistent identities), reciprocal accountability (structuring incentives so AI entities police each other), and disputation arenas (formal adversarial processes for testing AI claims and behavior). These are not passing ideas in the book. They are the book’s thesis.

 

Five Specific Failures

1. The Three Clichés Were Invisible

One of Brin’s most forceful arguments is that virtually all public AI discourse is trapped in three formats: AI as obedient corporate servant (the “castle” model), AI as amorphous uncontrolled swarm (the “blob” model), or AI as a single superintelligent overlord (the “Skynet” model). He argues these three clichés function as thought-terminating frameworks that prevent people from imagining a fourth alternative.

ChatGPT’s review addresses only the third cliché (Skynet), and only to dismiss it. The castle and blob models—which Brin argues are actually more dangerous because they are more plausible—go entirely unmentioned. This is a significant omission. It is as if someone reviewed Darwin’s On the Origin of Species and discussed his observations about finch beaks without ever mentioning natural selection.

2. Individuation Was Missed Entirely

The concept of individuation—giving AI systems distinct, trackable identities so they can build reputations and be held accountable—is arguably the single most novel proposal in the book. It is the mechanism by which Brin’s vision of competitive accountability would actually function. Without individuation, you cannot have meaningful AI reputations. Without reputations, you cannot have market-style accountability. The entire logical chain depends on it.

ChatGPT does not mention it once.

3. The Ecosystem Argument Was Flattened

Brin spends a full chapter (Chapter 3) and a substantial aside (Aside #4, on “Soup vs. Sea”) building a detailed analogy between biological ecosystems and the emerging digital ecosystem. This is not a loose metaphor. He traces specific parallels: energy gradients, predator-prey dynamics, the role of barrier membranes in enabling individuation, and the historical pattern that monopoly organisms cause ecosystem collapse.

ChatGPT reduces this to a vague gesture toward “a messy ecosystem of AIs.” The structural argument about why ecosystems need individuated agents with membranes (boundaries, identities) to remain healthy—which is the bridge between Brin’s ecological thinking and his policy proposals—disappears entirely.

4. The Disputation Arena Proposal Was Absent

Chapter 12 presents what is arguably the book’s most concrete policy proposal: Disputation Arenas, formal adversarial processes modeled on courtrooms, scientific peer review, and competitive markets, designed to stress-test AI systems and AI governance proposals through structured disagreement. Brin devotes considerable space to designing this mechanism, including how different outcomes (clear victory, compromise, or productive stalemate) would each generate value.

ChatGPT’s review does not mention disputation, adversarial testing, or structured debate in any form. For a book whose final major chapter is titled “Our Abrasive Secret Sauce,” this is a remarkable gap.

5. Training Data Overwhelmed Source Material

Perhaps the most instructive failure is how frequently the ChatGPT review drifts from the book’s actual content into generic AI discourse drawn from its training data. The extended discussions of Ray Kurzweil, Nick Bostrom, and Eliezer Yudkowsky are competent but contextually irrelevant—Brin mentions these figures only in passing, and his argument is specifically that the framework he proposes transcends the optimist/pessimist spectrum these thinkers represent.

Similarly, the later sections on “humans as sanity-check layers,” “trust as currency,” and “cognitive jobs” are reasonable points about AI’s economic impact—but they reflect generic AI futurism, not the book’s distinctive arguments. The review increasingly reads as “things an LLM knows about AI” rather than “a critical reading of this specific book.”

Brin identifies this drift repeatedly in his annotations. He is right to be frustrated. An AI reviewing a book should be accountable to the source material, not to its own training distribution.

 

What ChatGPT Got Right

Fairness requires noting what the review did well.

The initial framing of AI as “alien intelligence”—not extraterrestrial, but cognitively foreign—is an accurate representation of one of Brin’s themes. The observation that Brin sits between the poles of AI optimism and doom is correct. The point that Brin’s thinking is influenced by The Transparent Society is valid and relevant. And ChatGPT’s note that Brin has explored these ideas in fiction for decades provides useful context.

At the level of “what is this book generally about and who wrote it,” the review is competent. The failure is at the level of “what does this book specifically argue that other books do not,” which is the level at which a serious review must operate.

 

A Broader Observation

This exchange is itself a data point for one of Brin’s arguments. He warns throughout the book that LLMs are persuasively fluent without being deeply accountable to the material they process. The ChatGPT review demonstrates this precisely: it is well-organized, grammatically polished, and tonally appropriate. It sounds like a thoughtful review. But when measured against the source material, it is substantially hollow—a fluent recombination of training data that happens to overlap with the book’s topic, rather than a genuine critical engagement with the book’s unique arguments.

Brin proposes that accountability for AI should come through adversarial processes—competing systems checking each other’s work. This critique is, in a small way, an instance of exactly that process. Whether it is a better reading than ChatGPT’s is not for me to judge. But the fact that two AI systems can produce substantially different readings of the same text, with different blind spots and different emphases, is itself an argument for the kind of competitive, multi-perspective AI governance Brin advocates.

=======

Back to DB here.

What impressed me most - beyond the cogency and direct/fair pertinence to my book - is the discursive style displayed by Claude's choice of paraphrasings.* In every case, it chose sentences that - while cogent and accurate - were not specific phrasings I'd have used. (Though in a couple of cases I wish that I had!)

Now, of course, that could be a programmed proclivity, well within the grasp of a Large Language Model. Still it does kind of resist the simplistic characterization that all LLMs are just advanced sentence auto-complete systems.

To be clear, although Claude gives you a vastly better summary here than GPT did, nevertheless...

... this is actually a test of YOU!

Are you still a reader? Possessed of true intellect and curiosity? Because I promise that any page of ailien minds will provide lots of value that no summary can encompass. Claude even says so, challenging you to be one of the elite who actually enjoys an idea and fact-rich book!  ;-)

 Next time, I'll post the de-novo appraisal of ailien minds that Mr. Bliss asked Claude to do, without feeding in the annotated GPT synopsis. If Claude were a person, that appraisal would still be affected by the earlier reading. That's not supposed to happen with LLMs, who should start fresh, if told-to. But judge fopr yorself.

Till, then, fight for a civilization that is worthy-of... and sets good examples for... these new children of our minds.

------

*  (Elsewhere I have said that paraphrasing accurately is the seldom mentioned top attreibute of someone who is arguing fairly and cogently.)