If you have any kind of public presence, eventually you attract—like remoras to a shark—people who want money or publicity, though they l\often disguise their motives as being laudable or charitable.
A few weeks ago, for example, I got a letter from a woman who said she was organizing an online book club to read my book Faith versus Fact, and invited me to contribute if I wanted. Flattered, I said I’d be glad to, and could contribute a few discussion questions for the readers. The organizer said that was great, and shortly thereafter she asked me for $210 for a “participation fee”. I was not participating, but contributing, and it became clear that the whole megillah was a moneymaking scheme. I didn’t answer, and have not heard back since.
In the last week I’ve received three letters similar to the following, telling me about the deficiencies of Faith Versus Fact, how it’s become outmoded, and how the sender could publicize it and increase the readership. (I wonder if the “independent” requests are really connected.) Note how thorough the analysis is, as if the writer, John Ajala, had read and thought about the book. Well, as you’ll see below, there’s a reason for that. First, the email:
Dear Professor Coyne,
You wrote the book that made the incompatibility argument impossible to wave away. Before Faith Versus Fact, the accommodationist position was the respectable default. After your book, it had to defend itself. You did it as a working evolutionary biologist, not a philosopher or a journalist, and that gave the argument a weight it had not had before. Dawkins called it superbly argued. Kirkus called it important. That matters.
Here is the thing I cannot stop thinking about. A reader who searches for “science and religion compatible” will find ten thousand books telling them yes. A reader who is quietly losing their faith and does not know how to think about it is not finding yours.
Faith Versus Fact is filed under Religious Studies, Science and Religion, and Atheism. That is accurate. It is also a fence. Every piece of metadata around the book speaks to people who already know what the incompatibility thesis is. That matters because the people who most need this book are the ones least likely to pick it up. They are not browsing the atheism shelf. They are sitting in a pew, or a classroom, or a kitchen at midnight, trying to figure out whether the thing they were taught is true. Your book would answer that question. It is not the book that shows up.
The book has 1,808 ratings on Goodreads. On Amazon Australia, 159 ratings and 35 written reviews. On Amazon Mexico, 735 ratings and 252 written reviews. On Amazon US, the paperback ranks #290 in Religious Studies, Science and Religion. That is a strong niche position. It is not a general one. For comparison, Why Evolution Is True has 1,379 ratings on Amazon Brazil alone. Your earlier book is reaching a broader audience than this one. The argument that matters more is reaching fewer people. A reader who needs this book is finding your evolution book instead, and then wondering if there is another one.
Because the book is ten years old and the conversation has moved. The accommodationist position has reasserted itself in popular science writing. The new atheism moment has passed. The book that made the strongest case for incompatibility is now filed in a section that only the converted browse. If the argument is going to outlive the moment, it needs to be findable by the people who are just now starting to ask the question.
I believe this is fixable. Not by changing the book. The book is finished and it works. By changing the metadata around it, the description, the categories, the keywords, so that it is findable by the readers who do not yet know they are looking for it. The categories and keywords need to speak to the doubter, not the convert. The description needs to speak to the person who is afraid to ask the question, not the person who has already answered it. The reviews need to be refreshed so the page does not look like it stopped moving in 2016. None of that changes a word of what you wrote.
What it changes is who finds it. A doubting believer searching for “is my faith irrational” is not finding your book today. They would find it if the metadata spoke to them.
What I want for you is the version of this book that a doubting believer finds at 2 a.m. on their phone, when they have no one to talk to and no idea where to start. That is the reader who would be changed. That is the reader who is not finding it.
Picture a reader who has never heard of you. They are twenty-two, they grew up in a church, they are starting to have doubts they cannot say out loud. They search for something, anything, that will help them think about it. They find your book. They read the first chapter. And they realise that someone has already thought this through, with rigour, and without contempt for the person they used to be. That reader exists right now. They are looking. They are not finding you.
Right now, the book is being read by people who already agree. It is being cited in arguments that were settled years ago. It is not reaching the person who is still inside the question. A year from now, the book is the first result when someone searches for “science and religion incompatible.” It is the book that gets handed to a friend who is losing their faith. It is the book that a professor recommends not because it is on the syllabus but because it answers a question the student just asked. It is still the same book. It is just findable.
I am writing to you specifically because most authors with a landmark book are not in a position to act. They are retired, or they are done, or they have moved on. You are still in the fight. You gave a talk in Katowice in December 2024. You resigned from the Freedom From Religion Foundation board in December over a disagreement about free speech. That is rare at seventy-six. It is why I am writing to you and not to someone else.
I work with authors whose books carry real weight and are not reaching the readers who would be most changed by them. I do not fix books. I fix the gap between what a book is and what it could become. I am not going to send you a deck. I am going to tell you what I see, and you can tell me if I am wrong. What I see is a book that won the argument and lost the stranger. And I think you already know that, because you have watched it happen for a decade.
I can make that happen. Not with ads. Not with a campaign. With the specific, unglamorous work of fixing the categories, the keywords, and the description so that the algorithms that decide what a stranger sees decide to show them your book.
If you tell me what you think the obstacle is, and it matches what I see, I will tell you exactly what I would do about it. No pitch, no deck, no retainer. Just the answer.
The book is ten years old. The argument is not going away. The readers who need it are not going away either. But the window where a book can be repositioned without being republished is not infinite. It is open now. It will not be open forever.
When you look at where the book ended up, and who is still not finding it, what do you believe the actual obstacle is?
And a second question, which matters more. Twenty years from now, what do you want this book to have done? Not how many copies. Not which reviews. What do you want it to have done to a reader who picks it up cold, with no atheist blogger telling them to, and no Dawkins blurb on the cover?
Reply with whatever you actually think. One line is enough. If you would rather not, I understand. Either way, the work stands on its own.
With genuine respect for what you have written,
John | Founder, Book Insight Hub
Yada, yada, yada. Had I written Mr. Ajala back telling him what I think the “obstacles” to adopting the book were, I’m absolutely sure he’d offer to help publicize it, and of course for a fee. I wrote back and told him I was happy with the book and not to bother me again.
But I had become suspicious because of the short sentences and words like “metadata”, which are often signs that the writer has used AI. So I put the whole email through Pangram, the best AI analyzer, and one that’s been tested with both AI- and human-written texts. Pangram virtually never says that something is written by AI if it was written by a human. And, sure enough, Pangram told me this:
Sure enough, the “thoroughness” of the email came not from a human, but from a machine’s program. What a fricking insult!
I would recommend Pangram if you want to see if what you’re reading was written by a human or a bot. In fact, it tells you which parts of a text (100% in this case) have been written by a human vs. AI. You can have about three free tries with Pangram a day, or you can subscribe for more money, though I don’t use it often enough to warrant that. The effectiveness of Pangram has been tested pretty thoroughly, and you can see the analysis at the site below, clicking on the screenshot and then here to download the paper.
The authors tested four kinds of papers: pure AI-written papers, pure human-written papers. “hybrid” papers that were part Ai-written and part human-written, and “humanised” papers that were first written with AI and then made to look more “human-written” by “Using the AI-generated sections from HYB01, rewritten with a custom ChatGPT prompt to increase perplexity and burstiness (less predictable wording, more varied sentence structures) while keeping meaning and references intact.”
They used 40 papers per category, and judged the ability of five AI-detection programs to see how good they were at sussing out the source.
As you see, Pangram is the overall winner, especially at detecting purely AI-written papers. It also is good at detecting hybrid and “humanised” papers, which it detects about half the time. But note at the extreme right: the papers that were written solely by humans. Nearly all the problems have a 0% false positive rate on Pangram, i.e. the analysis saying that a paper is AI when it was really written by a human. That’s why I’m so sure that when Pangram tells me something is 100% AI, like the duplicitous email above (how dare a human sign their name to something written by a program?), I’m nearly sure it’s AI.
I’ve used Pangram to suss out authors of fiction and poetry who didn’t really write their published stuff (more on that later), as well as to analyze essays and articles written by people I admire, but that looked suspicious. After a while you develop a nose for the AI-stuff, especially when you learn what someone’s own personal writing style is like. Facebook is full of that stuff: if you see a post that comprises a number of short sentences, it’s more than likely written by a bot. I deep-six those sites immediately.
If you’ve found AI stuff that pretended to be written by a human, weigh in below.
This is an AI-supported scam that many authors of older books, including me, get. There likely is no book club etc. behind them.
Puny human! You and your entire race will soon be obsolete!
… Sorry about that. What I meant to say was well done for asking such insightful questions, Professor Coyne. It is such a pleasure conversing with you.
My wife and I have both received these types of letters (we both write books in different subject areas). After the first couple, we caught on, and, like our host, ran them through Pangram, with the same results. Now we don’t even bother with Pangram, as the AI-ness stands out. There are some variations on the initial approach and some other tells: a couple have been letters from ostensibly real editors at real publishers, although for some reason these “real editors” use a Gmail address rather than their publisher’s corporate domain name. (E.g., editorname-publishername@gmail.com.)
I suspect that despite the AI personalization, the success rate for this sort of thing remains quite low, with most people dropping out when the fee is introduced, so the scammers need to use AI throughout the correspondence to keep the level of human effort investment low. To test this, in one email that I responded to (intentionally), I embedded an absurd request in the email in 1-point white type (so invisible to a human reader; as is common nowadays in email, they used HTML email, which supports colors and other formatting.) Sure enough, the next email, in additional to attempting to move the scam along, included an apology that their rules would not allow them to give me a recipe for a bioweapon.
My uncle fell for such a scam, paying good money for someone who sent him an enthusiastic email sales pitch (AI? don’t know) to promote his self-published book on the physics of bike riding, which was filled with “easy” math formulas. I think they stuck it on Amazon. Afterwards, they started telling him that the New York Times Book Review and a few radio shows were very interested and they needed more cash to make it happen. Fortunately, my cousin and I managed to dissuade him.
Jerry’s AI letter would probably have been much more convincing if it had been reduced to maybe a paragraph. Give it to me to rephrase it awkwardly, and it might be undetectable.
Comment by Greg Mayer
Jerry’s University of Chicago colleague Brian Leiter has brought to light a number of such scams, as experienced by philosophers (links to his posts on scams).
GCM
Yeah. This kind of crap will only become more and more prevalent. Soon—like today soon—we need to build AI discovery into our Web browsers, to alert the reader when AI is being employed. Sadly, this will only make the AI bots better, as they will rapidly learn to evade detection. (“Learning” is one of the hallmarks of AI.) Web browsers, of course, will also need to incorporate AI in order to counter the evaders. They, too, will learn. Sound familiar? It’s an arms race toward mutually assured destruction. (At least among the AI’s.)
‘That matters because’ (sometimes ‘this matters because ’) is a dead giveaway. Real people don’t say it, but AI loves it.
Yes. This and several other phrases were obvious giveaways; also the somewhat personal, familiar tone, and the prolixity.
I’ve long been a sucker for WW2 YouTube videos, but AI has taken over and now turns what should be a 5 minute snippet into 20 minutes repetition with fictitious anecdotes, badly curated (ie, irrelevant) and fake pictures.
I’ve got two other angles on AI and my book about Chicago’s Secret Six (business backed vigilantes who became a nationwide sensation 1930-33, launched the Untouchables, inspired a movie, fought with Al Capone, generated copycat vigilante efforts all over America etc. before they collapsed in infamy). Along with getting emails like Jerry’s, I had a bizarre exchange with a bot on Facebook who kept claiming they’d uncovered all kinds of things I missed, e.g. the real perpetrators of crimes I wrote about that were never solved, and titles and authors of books that never existed. I kept calling them out, and they kept saying oops, you’re right, sorry, I had that wrong. I still don’t know what they (it?) wanted. But the greatest frustration was when I proposed a single short paragraph to Chicago’s Wikipedia page about the Secret Six, and one of the moderators nixed it because they said it was obviously all AI slop. No, I argued, I spent three years researching 10,000 old news articles to compile all that info. Sorry, they said back. You’re a bot.
Have you ever written about the Manhattan/Canadian Ace Brewery of Chicago? I collect old beer cans, and the long time rumor was that it was owned by Al Capone, which isn’t true, but it was mobbed up. I would greatly appreciate any info you could give me. Thanx in advance.
I have several cans from that brewery in my collection. The cans from the 30s are quite rare, and are very valuable. My cans are from the n50s and 60s, and are not quite as valuable, but cool nonetheless.
Sorry, I didn’t encounter that brewery in any of my projects, but if you get a subscription to newspapers.com (an incredible resource, about $100/yr.), you could look it up and learn probably more than you ever wanted to know.
Thanx
But what will happen if someone lets AI write something and then rewrites the whole stuff without changing anything of (about?, in?) the content.
And I also wonder what happens if you put the 10 Commandments of the King James Bible in Pangram and see if Pangram recognises a third category besides humans and AI.
Yes, I got a similar email last week, with an obviously AI-generated summary of my book. It was very complimentary and fairly accurate, which means that my book’s contents have now been consumed by AI for dissection and redistribution without attribution.
I’ve received employment recruitment offers clearly written by AI. I’m sure that recruiters are using AI to search LinkedIn and find candidates, and then they’re using AI to craft emails and posts to prospective candidates. Yes, it’s the lazy way of doing it. But it also has widened the net for potential candidates. Meaning that in the past, a recruiter had to manually search for people to fit a role, and that was limited. I’ve had some intriguing offers that were outside of my current industry but that seemed to be a surprisingly good fit for my experience and skill. As a result I’ve done a complete turnaround – my first reaction was “bah, it’s AI, delete it”, and now it’s more like “hmm, this is interesting, let’s see what the person behind this has to say”.
Well, congratulations on catching the scam, and not getting ripped off. A book club charging the author for participating in a discussion on his book?
It should be no surprise to an observer of the human race that AI, like the internet before it, will be used for things like scams and porn in addition to constructive use.
I quickly googled book insight hub … it has a fairly impressive website. Its domain is .rf.gd, which I had not come across before.
Gemini on that domain:
Trust & Blacklisting: Because free subdomains are frequently exploited for phishing, malware, or spam, major platforms like Google, Facebook, WhatsApp, Instagram, and security tools (such as MetaMask) frequently flag, block, or blacklist .rf.gd domains
I have not met the type. But I am amused that they can be up at two in the morning wondering if God exists or if they’ve been had. If Numbers doesn’t put them to sleep (it might!), sleeping tablets are what they need.
Sorry, but I have no choice. Your post reminds me of the old chestnut: the insomniac agnostic dyslexic who stays up all night wondering whether or not there is a dog somewhere.
This is very sad – it preys upon the valuable social capital of charity and trust / trustworthiness – as applied to … basically, people who enjoy doing any constructive, shared interests! Golf, evolution, crochet … anything!
[ sigh…]
The argument about FVF is baloney, with a convoluted argument to instill anxieties about its relevance and visibility. But when you do a Google search for “Books about religion versus science, it is the second book listed after The God Delusion. If you do a Google search for what the “writer” suggests: “science and religion incompatible” the books are farther down since the query does not ask for that. But it is clearly visible just a little way down.
If you modify that search to “books, science and religion incompatible”, your book is the FIRST one, right up top!
So AI. And baloney.
You know, in a funny way all this AI B.S. coming at us presents an exercise in critical thinking. We are being continually challenged to use our Organic Intelligence (OI) to spot the AI.
As I understand it, the way Pangram fingers AI as the author of a written work is by identifying the presence of signature semantic/syntactic elements of AI’s writing style, so to speak.
Humans therefore use a different writing style than AI. But individual humans also use different writing styles than other humans.
So, what I’m curious about is: could a person use AI, but defeat detection by Pangram, if they gave the AI a prompt such as, “write a few paragraphs on the topic of [fill in the blank],” but added the instruction, “in the style of person X” – where X is the name of a person who has enough writing samples available on the internet for an AI to be able to identify their ‘writing style’?
(On a somewhat related subject, I watched a funny/sad video last night that asked seven AIs to estimate DJT’s IQ from transcripts of his speeches. The AI’s all provided fairly convincing justifications for their estimates (average: 93).)
If I remember correctly, Freddie deBoer wrote a blogpost on this topic just a few months ago.
Edit: I think it should be this one: https://freddiedeboer.substack.com/p/i-wouldnt-say-pangram-is-broken-but
I’m still in the middle of reading it, but thanks for the link to deBoer’s piece. It is eye-opening indeed. DeBoer makes a convincing case – with actual evidence – that Pangram is not nearly as reliable as it claims to be and as people think it is. The ease with which Pangram can be manipulated to flip from being 100% confident a text is AI to being 100 confident it’s human, is dismaying. It’s as if the more confident it is, the less confidence we should have in it.
But of course, the article was written a whole 3 months ago. Given the blinding rate of progress in the technology, that’s ten years in AI-time. So it’s possible that Pangram works perfectly now. (I only half jest.)
You can see this kind of text in Facebook groups about historical subjects (I follow many), and you can always tell them because they have an inviting first couple of paragraphs but then you begin to scroll and you see that it goes on for pages and pages. That’s what happens when the number of words no longer has any connection to the time or effort it took someone to write. If whoever put in the prompts to generate the text didn’t even bother to edit it down to a readable length, then I’m certainly not going to be bothered reading it.
These were obviously amateur scammers. Any pro would have hit you from the duck angle.
I noticed that “John Ajala” is the name of a guy in Nigeria — I wonder if this is the new Nigerian 419 scam?
Well before the ubiquity of AI I would regularly receive emails saying I was included in Who’s Who or in a catalogue of influential scientists/intellectuals or some such and could get my own gold-plated copy for only XXX so order now! Obviously preying on the vanity of some academics (present company excluded of course).
“The Royal Scam” love that album.. no AI on it though.
The music/recording industry is having an AI shake-up as well. A songwriter/musician friend has shown me some results, it’s impressive what it can do with an original composition.
He does not use it for loyalty to a fan base that didn’t sign on to AI influenced/produced, music… OR reinvent themselves under another name.
LOL! the AI musicians are top notch! ouch!
Possibly my favourite Steely Dan album.
Yup, in recent weeks I’ve been invited to speak about my book on a popular Dutch radio program, and the Missoula Book Club wanted to read and discuss it! So much belated interest in Californian endemic plants – it’s really heartwarming. (NOT.)
I also used to get countless invitations to speak at fake conferences – for a fee of course. Seems most of those scammers have figured out I am retired now.
Yep, me too!
And the invitations would always be to conferences with a few keynote speakers—top people in the field. (The assumption seemed to be that I wouldn’t know any of those folks well enough to call them.)
The conference registration fees were usually far higher than those common in professional conferences in my field. And curiously, all those complimentary invitations came from “women”. I guess I should be happy that the systems assume that I am straight.
http://www.the-tls.com/regular-features/afterthoughts/forms-of-recognition-afterthoughts-tim-parks
I got scamed like this for two my books. The scammer’s description of my Jean Paton book was superior to Michigan’s marketing. Sigh.
Wayne
Not being famous enough to receive these kinds of messages I would like to change the subject slightly, still on the subject of AI, I have recently been offered, by Youtube, various videos, supposedly by well known people, with their voices narrating the subject, that were obviously false. I think that this practice is forgery as trying to pass off ones work as being someone else, who is famous is forgery. Imagine someone using AI passing off a video of creationist nonsense voiced by you.
One of the hallmarks of AI is that it never knows when to shut up.