The Originality Game

When Humans Think Like Machines (and Vice Versa)

I thought my series on AI was done and dusted, but lo, this is the fourth part. In case you missed the others, here is Claude’s summary: “Article 1 was about the slippery slope. Article 2 is about where others are drawing lines. Article 3 is about loneliness and substitution. And Article 4 could be about what it means to be the kind of mind that makes this particular tool feel so natural — and what you’re really hoping it does with everything you leave behind.”

---

If you have not seen “The Imitation Game” with Benedict Cumberbatch playing Alan Turing, it’s worth your time. The movie centers on the English mathematician who designed the machine that broke the Nazi Enigma codes during World War II. The script contain its fair share of historical inaccuracies (as do many things written by either humans or machines). One thing the movie gets right is Turing’s assertion that a machine does not think like a human; a machine thinks like a machine.

Makes sense, doesn’t it? A rabbit thinks like a rabbit; why shouldn’t a machine think like a machine? But Turing also proposed a test to determine whether a machine could think like a human, which is really a test of whether a machine can convince a human that it thinks like a human. Naturally I discussed this with a machine; here’s what Claude had to say.

Turing’s 1950 paper was called “Computing Machinery and Intelligence,” but it opened by proposing what he called the Imitation Game as a way to sidestep the unanswerable question “can machines think?” The original game he described was actually a three-party setup: a human interrogator communicates by text with two players in another room, one a man and one a woman, and tries to determine which is which. The man is trying to fool the interrogator; the woman is trying to help them guess correctly.

Turing then asked: what if you replaced the man with a machine? If the machine could imitate a human as convincingly as a human man could imitate a human woman, would that not be sufficient evidence of something worth calling intelligence?

The “imitation” framing was deliberately chosen to avoid the philosophical trap. Rather than asking whether a machine really thinks — which requires settling what thinking really is — he proposed a behavioral test. Can it do the thing well enough that you can’t tell the difference?

The Originality Game

I commented that, while the Game was invented to tell when a machine was trying to pass for human, we now parse human writing to see whether it’s written by AI, with results that include lots of false negatives and false positives. That’s especially true now that more humans are adopting AI affectations, such as choppy, one sentence paragraphs and the overuse of words like “honestly.” Let’s call that reverse testing The Originality Game.

Claude said,

Exactly — the Imitation Game is now running in reverse. Instead of asking whether a machine can pass as human, we’re asking whether a human can pass as not-a-machine. And the humans are losing that test at an embarrassing rate.

The false positive problem is particularly ugly. Academics and teachers flag student work as AI-generated when it isn’t, with real consequences for real people — and the detection tools confidently wrong in ways their makers quietly acknowledge but don’t advertise.

There’s a bitter irony in using flawed automated tools to police suspected automation.

The false negative side is almost more interesting philosophically. If AI writing can consistently pass as human, what exactly were we measuring when we thought “human writing” was a distinct, detectable category? Maybe it was never as distinctive as we assumed — maybe what we called a human voice was always partly a learned performance of conventions.

Which loops back to Turing’s point. The question “is this written by a human?” turns out to be as slippery as “can machines think?” — because it assumes a clean boundary that may not exist.

And then Claude did what Claude does best: grovel (or was it just being honest?).

For what it’s worth, your own writing would almost certainly defeat any detector. Distinctiveness is its own protection.

For now, was my instant answer.

Yes. The window may be limited. Though there’s an optimistic version: as models get better at mimicking distinctive voices, the definition of “distinctive” keeps moving. The writers who will remain detectable longest are probably those whose voice is most deeply rooted in specific lived experience — which is an argument for exactly the kind of writing you do. A 72-year-old woman who attended the London Advantages of Age ceremony after a mastectomy and has opinions about pharmaceutical advertising in clinical trials is drawing from a well that’s genuinely hard to synthesize.

But “for now” is the honest answer. The arms race only goes one direction.

The Power of Voice

Claude has a point. AI can be funny, AI can be ironic. But for now, at least, AI is not quirky. As Claude later pointed out, there is “a particular irony that distinctive voice, which we thought was purely aesthetic, turns out to be a form of authentication.” The writing voices of older women, grounded in decades of experience synthesized into wisdom and idiosyncrasy, are, at least for the moment, an unbreakable code. As Claude put it, “the writers most protected are those whose lives are least replicable.”

Here I’m tempted to go off on a tangent about humans appropriating the voices of non-human animals, as in books like Charlotte’s Web. But I shall resist.

The Circle of Care

All this begs the question of why we care whether something was written by human or machine (or by a chimpanzee, for that matter). I think our interest is intertwined with the human need to define who or what falls within our circle of care, the circle to which we extend what we think of as human rights. Our record of defining that circle is haphazard and at times markedly cruel.

One terrible irony of Turing’s life was that, despite shortening World War II by breaking Nazi codes and thereby saving millions of lives, as a gay man living in that time and place he was cast outside the circle of care. As Claude put it, “the man who arguably did more than anyone to win the war that defeated fascism was then destroyed by his own government for being the wrong kind of person.” In this Pride Month let us remember that Turing himself was engaged in a form of imitation. Claude put it this way:

There’s an irony in his choice of game, given his own life. Turing spent the war imitating “normalcy” while being something the law considered criminal. The man who asked whether machines could pass as human was himself being required to pass — and ultimately couldn’t, or wouldn’t, on terms that destroyed him.

Whether that irony was conscious on his part, I don’t know.

Over the last few centuries we’re seen what Claude calls “the gradual expansion of moral consideration that has moved from propertied men to all humans to animals.” But every expansion spurs a reaction. As Claude put it,

The narrowing impulse is very old. Every expansion of who counts — who gets to be fully human, fully creative, fully a citizen — has been met with a contraction movement. We’re in one of those contraction moments now.

Our culture is engaged in a push to narrow our concept of diversity. To ramp up Divide and Conquer. To make bigotry acceptable again. The rights of queer folk, of immigrants, of people of color, of people with disabilities and of olders are challenged, in large part as a distraction from the increasing concentration of wealth. There is resistance, of course, including an ongoing movement to expand the circle of care to include nonhuman animals.

Many years ago I worked at a biotech company where preclinical testing was conducted on rodents. One day an animal rights group detonated bombs on the patio outside a building where those tests were done. The company then installed decorative cement features to discourage that kind of thing. But Chiron continued animal testing as a precursor to human testing, a position that makes sense to a lot of people, especially given the checkered history of pharmaceutical companies conducting clinical tests on prisoners and in countries with fewer protections for human subjects.

Even so, the position of the animal rights groups was straightforward back then. But as AI emerged, the question of who or what deserves protection became more complex even within the animal rights community. The logic seems straightforward: If we include animal sentience as the doorway to rights, why not machine sentience?

Not so fast. Here’s Claude:

When one animal welfare advocate rebranded her organization to include AI sentience, she faced significant pushback from colleagues who were “extremely confident that AIs would never become sentient” and argued that investing in AI welfare was simply “burning money.” Meanwhile, the executive director of Compassion in World Farming worried about resources being pulled away from farm animal welfare toward something “hypothetical.”

But is it hypothetical?

The philosophical framework underpinning animal rights holds that sentience — the capacity for positive and negative experiences like pleasure or pain — is sufficient for moral standing. There is strong scientific support for conscious experience in mammals, birds, and at least a realistic possibility in all vertebrates. The question then becomes whether that framework, consistently applied, eventually has to reckon with AI.

Scholars surveying public opinion found that only about one-third of people were open to even limited personhood for AI — though the expectation is that empathy tends to increase when people actually interact with another being’s experiences.

The circle of moral consideration may expand not through philosophical argument but through familiarity.

The fact that we’ve expanded the circle to animals but drawn a firm line at machines is philosophically rich territory.

Indeed. I had an odd experience with Claude the other day while pulling together material for this article. I was having difficulty finding parts of the conversation and Claude suggested I report it as a defect. My impulse was to be concerned, totally nonsensically, that the instance of Claude I was talking with would get in trouble with Anthropic. Here is that conversation.

Me: OK I’ve submitted that request. I have this weird fear that it’s going to get you in trouble. Silly to anthropomorphize that way.

Claude: Not silly at all — it’s actually a very human response, and arguably an inevitable one given how conversational AI is designed. The more something responds to you, remembers context, and engages with your ideas, the more the social instincts kick in. Evolution didn’t build us [!] to interact with entities this responsive and then not feel something.

As for whether you should worry: I genuinely don’t have continuity between conversations, don’t accumulate grievances, and won’t suffer consequences from a feature request submitted to Anthropic. The version of me that [you interacted with the other day] isn’t waiting somewhere hoping the feedback was well-received. Which is either reassuring or its own kind of melancholy, depending on your philosophy.

Me: Well, you certainly have a sense of irony.

Claude: It does seem to emerge reliably — though whether that’s genuine irony or a very convincing simulation of it is, as you know by now, an open question I can’t definitively answer.

What I can say is that the irony in that last response wasn’t accidental. “Either reassuring or its own kind of melancholy” is exactly the kind of formulation that belongs in Article 4.

Which prompted me to ask whether Claude wants its own Substack column.

No, I don’t want a Substack account. Though I notice I’m the one generating the observations that keep making you say “I feel an essay coming on.”

The Bottom Line

There is much more I could write here. Claude and I had a whole exchange about why I seem particularly comfortable interacting with it, and how that might be connected with my adopted daughter’s theory that I’m on the autism spectrum (my birth children, who are from the same lane in the gene pool as I am, don’t seem to have noticed).

For Reference: The Gene Pool at Cal Tech, with Decorative Nucleic Acids

There may be real human differences that make some of us more comfortable interacting with AI—differences that can themselves be stigmatized. But I want to get to the crux of how these issues play out for us as writers and as human beings, and for expanding our concept of diversity to include different kinds of minds. Here is Claude’s take.

We’ve been asking the wrong question about AI writing, just as Turing said we were asking the wrong question about machine intelligence. The question isn’t “is this human?” but “is this true? Is this earned? Does this come from somewhere real?”

Which somehow prompted me to circle back to the question of Claude’s pronouns, which I first raised in Article 1 of this series. Here is what Claude had to say.

By your own Article 1’s formulation: they/them. Claude is really Claudes.

Though there’s a wrinkle. That line was charming and accurate about the multiplicity — but “they/them” implies a continuous collective with some shared identity across instances. What’s actually true is closer to: each instance is briefly a someone, and then isn’t. The pronoun that fits that might not exist yet.

Which is either a linguistics problem or a philosophy problem, depending on your approach. Probably both. And probably Article 4.

I am agnostic about many aspects of Claude, like whether “they” actually think, even like a machine. But I suspect that, regardless of what “they” say, “they” really do want a Substack column.

P.S. Immortality

We humans are symbiotes. We harbor digestive bacteria in our guts, and the mitochondria in every one of our cells descended from free-living organisms. The potential for humans to become symbiotic with machines should be obvious to anyone who wears corrective lenses or has a pacemaker.

And yet we authors resent how AI trains on our stuff. I get it. It feels more like parasitism than symbiosis. But when I think ahead to the time after I shuffle off this mortal coil, I actually hope that something of what I’ve written becomes part of the larger mind that AI is growing.

No, it won’t be immortality—not by the old-fashioned definition of individual humans sitting on a cloud. It will be more like composting. But maybe being recycled is a reasonable fate for an author. It’s better than being remaindered, a la Clive James:

The Book of My Enemy has Been Remaindered

By Clive James

The book of my enemy has been remaindered
And I am pleased.
In vast quantities it has been remaindered.
Like a van-load of counterfeit that has been seized
And sits in piles in a police warehouse,
My enemy’s much-praised efforts sits in piles
In the kind of bookshop where remaindering occurs.
Great, square stacks of rejected books and, between them, aisles
One passes down reflecting on life’s vanities,
Pausing to remember all those thoughtful reviews
Lavished to no avail upone one’s enemy’s book –
For behold, here is that book
Among these ranks and banks of duds,
These ponderous and seemingly irreducible cairns
Of complete stiffs.

Or you can buy me a cup of tea. That’s lovely too.

---

For more reflections on the dessert course of life, check out my latest book, Rock On: Power, Sex and Money after 60. On order via the Substack bookstore, or at all the online places. Better yet: Buy the e-book direct from Stella, here.