Chapter 1: The Someone at the Door

In 1985, after a long day working for one of the largest computer companies, I was lying on my old living room sofa, TV tuned to a sitcom rerun, volume low, trying to take a nap. But I couldn’t relax, and then I realized why. The door on the wall next to the TV was open. It was my spare bedroom without a bed, just a desk and some bookshelves.

As much as I tried to focus on the episode of Hogan’s Heroes, a show that consistently allowed me to fall asleep, my eyes and my attention were continually drawn back to the room.

On the desk was a $6,000 PC that my employer had insisted that I take home. I didn’t trust it. Looking back I can’t even remember using it. I wasn’t sure what would happen if I started using the computer at home after using computers all day at work. But I had a feeling it would change me.

I found that I couldn’t relax and watch TV in the living room unless I closed that door, locked that computer away, or maybe just locked myself away from it. I wasn’t sure why I felt more safe with it closed, but I did.

Work was work, and home was home. I can still feel that desire to separate those worlds.

Even with it closed, I could hear the PC scratching gently at the door, asking to be let out so it could nestle with me on the sofa and invade my life.

Fast forward forty years, and I’m editing this book at 6:41 in the morning on my iPad. I love my iPad. But I am its boss, and it works for me, and only when I invite it to.

Or so I tell myself.

Digital Equipment Corporation, my employer, was also where I first met AI. In what felt like a welcome but relentless initiative to inject the future into my life, they sent me to their Cambridge research center in 1986 to learn about it. We spent months trying to build what they called an “expert system,” software that could answer questions the way a senior railroad engineer would. It was a debacle. I knew within weeks that what we were calling artificial intelligence wasn’t intelligence at all. You can’t pour a lifetime of expertise into a software program any more than you can pour a river into a bucket. I told them so. I wasn’t the most popular student in Cambridge.

Years later, at RAND, a fellow lunchtime beach-volleyball player and colleague named Jim Dewar shared a simple and unsettling idea with me: the consequences of new technologies, especially the big ones, are almost never the ones anyone predicts. The printing press was supposed to make Bibles cheaper. It split the church in half.

Everything I learned about technology over four decades, building it, studying it, watching it reshape organizations and lives, never made that feeling go away. Every time we add something that is supposed to make our life better, the only thing that is certain is that it will make our life different.

And now AI has pushed that bedroom door open. Maybe knocked it off its hinges, to be more accurate.

You’ve felt it too. Maybe it sits in your gut, that this is different in some way. Or you feel like me, an uncanny sense that something is there, something inside, and another part of you says, It is just a machine, how bad could it be? I could always close the door. You can’t name it, but you can feel it.

Or maybe it sits differently for you. Maybe what bothers you is that it seems too smart, too clever. It can produce in seconds what took you years to learn. And yet you can feel that it is wrong sometimes, or maybe just a bit too slick. Can it do what you do, and better?

These days, I worry how profoundly it might change the all of us, our relationships, our families, and even our society. I’m not talking apocalypse… I’m just saying we might want to put our seatbelts on because this one’s gonna cause some change.

Back in the ‘70s when I was growing up, there was a thing that the hippie generation used to put on T-shirts: “If you’re not paranoid, then you don’t understand what’s going on.”

I’ve spent the last year taking the AI machine apart. Not the hardware...that’s actually trivial, though I sure wish I had bought Nvidia stock. I mean the machine’s behavior. I wanted to know what it actually does when it talks to you, and why it’s so damn good at making you feel like someone is home. The technology is very, very complex and frankly quite amazing. But how it actually works is far more simple than you would ever guess.

Once you understand the cleverness behind this new machine, you’ll have words for the things you can’t name right now. You will understand when to trust the machine and when to lean back.

This is my detective story. Short chapters, each one does one thing. I’ll start with what happened to me, then show you what’s inside, and then we’ll get to the part that actually matters: what this thing is doing to you, all of us, and how we’ll need to live with it.

The secret to understanding this new technology is something I discovered on a morning walk, the kind where your mind goes quiet enough to hear what’s been bothering you.

Keep reading:

The Mostly Helpful Psychopath

The Mostly Helpful Psychopath V 2.0 - Part 1

Chapter 1 – The Language Inside of Us…

Read more

12 days ago · Jack Skeels

Chapter 7: Breakfast with Dr. Bob and Sally

It was a few minutes before 7 AM, and I was thinking about refilling my coffee and getting my usual yogurt, granola, and blueberries ready, when all of a sudden a text came up on my phone from Jamie’s mother, Sally. Part of me, didn’t want to look to see what it said… I could not think of any reason why they would text me, why Sally especially would text me, when normally her husband did.

That’s not true. I could think of one reason and I didn’t like it. Both my folks have passed away, and I’ve lost several good friends over the years, including one who died in a fire plane crash at LAX back in 1991. Jamie’s folks have been a wonderful gift in my life, giving me a second chance to be around parents. But they are 88 now, and the clock ticks louder at that age.

I have Sally and Jamie’s dad, Dr. Bob, on emergency bypass on my phone for all those reasons.

I pick up the phone, and, with more than a little relief I see that she is inviting me to join them for breakfast at a nearby diner.

Before I can even respond, she explains herself: Dr. Bob’s car needed to go to the dealer for some problem, and he wanted to make sure he was first in line for service – which does not surprise me – and they were already on their way back in her car, now headed for breakfast.

It was a first, just the three of us getting together. Jamie would love that we did this, especially without her.

Dr. Bob, a retired physician, is relentlessly curious about everything (including AI and this book as I was writing it), and despite his aging body, he has a strong mind, a cognitive gift from heaven at his advanced age. He’s an inspiration in that way.

When I arrived, they were already seated, waiting for me. I think the older you get the less time you spend looking at menus; maybe after 30,000+ diner breakfasts, you know what you want. They looked ready to order before I even sat down. Their menus were stacked on top of each other, mine was waiting for me.

Though the place was only a few miles away from our house, I had never been there before. Like with so many diners, the breakfast menu felt massive, at least a few pages, and somewhat daunting as I didn’t know how long they had been waiting…and the font on the menu was designed for teenagers. Really? Do they now know who their customers are?

It didn’t help me either that Dr. Bob was trying to explain to me the latest liquid he found oozing out under his car.

But what I said wasn’t what I was thinking. I was really thinking, Oh, Bob, the AI got you on this one. He trusted the AI’s answer and may have actually asked the AI in a way that steered it to give him a half-true answer, in a way that seemed ridiculously confident or reassuring. The AI had told him enough of what he wanted to hear, maybe rather than what he needed to learn.

Once you get how this works, you hear signs of this AI betrayal in what a lot of people say.

Red fluids were to Dr. Bob what Splenda was to me.

Here’s what’s tricky about what Dr. Bob was trying to find out: both coolant and transmission fluid are red. So, yes, the AI was truthful.

But an expert (or even me, a sort-of expert) can tell you which one is more likely to be on the floor of your garage, under your 2019 Cadillac SUV. Heck, I would even bet that one of those old railroad engineers would have given better advice to Dr. Bob than his AI did.

Today’s AI still fail at the questions me and my fellow AI programmers failed decades ago.

At one point, our breakfast conversation veered over towards this book that I was writing, and Dr. Bob decided to explain to Sally, “… You see it has this big database…”

All I heard was white noise after that…I don’t remember leaving the breakfast or driving home. This was the reason why I felt betrayed about Splenda: I had also assumed that it had a database, and why didn’t it just look it up?

I think we got used to the computers of the past. If it gave me an answer, it was the answer. And if it didn’t know, it was obvious, an error message, or just an answer that looked so stupid that it did not fool me.

When today’s machine gets something right, it feels so right. And when it doesn’t…yeah, that’s the question. Why was its “database” failing me?

Part of me also thought for a moment that maybe I wasn’t paying enough, as if “Always provide correct answers” was a feature that required a better subscription. Not so.

The answer is not what I thought it would be, and once I started understanding it, I started understanding what the machine can and cannot do.

Chapter 8: What the Machine Had for Breakfast

I had Chilaquiles, Sally the French toast, and Dr. Bob had the American breakfast, I think.

Our new friend, the machine, had a much bigger breakfast. For the main course, it ate everything that was reasonably accessible on the internet.

When the waitress came by and cleared all of our plates away, she asked if we wanted anything more, well, Dr. Bob, Sally, and me, we were full. We waved her off. None of us finished our meals…a hazard of aging, I think. You get older, and you can keep that 900 calorie plate of Chilaquiles off your waistline if you leave half of it on your plate.

Dr. Bob, Sally and I probably clocked in 500 calories each at the most. Our combined leftovers could have fed another three people.

And then there’s the machine. If it had truly been there in our booth, it would have leaned back and burped a good burp. It had a big meal.

When the machine replied to the waitress, it said, “Yes, ma’am, what do you have for dessert?”

The waitress brought dessert: a massive pile of books, academic papers, code repositories, curated databases. These were all higher quality than the “$5 Internet Breakfast Special” that it had already eaten. It was the machine going to a French pastry shop after filling up at a buffet.

All told, the machine ate more than us: 60,000,000,000,000 bytes of human writing. Eggs over easy, bacon on the side, please.

60,000,000,000,0001

Look at that number. I mean stare at it. A lot of zeroes. My eyes wobble a bit when I look at that number…I don’t really know how to visualize it. I suspect I certainly could never count that high.

Convert that number to books and you’ll find that 60 trillion bytes is roughly fifty million 300-page books. Uh, okay.

If I really try, I can read maybe fifty books a year, and if we assume I have a fifty-year reading life (but I’m still going strong!) then fifty million books is roughly twenty thousand lifetimes of reading. No human who has ever lived read more than a tiny fraction of what the machine ate.

So, you gotta figure that there are more than a few auto repair manuals that were served up, right? And at least one research paper on Splenda…and of course a lot of blogs and Reddit posts as well, etc., right? Yes, and the machine ate them, but they are not inside.

But things get weird now as I dig into this.

I ask how big is this database? The answer I get is that it is less than 1% of the size of all of the things that the machine ate.

I’m not sure how to explain how impossible this is. I mean that you can take 50 million books worth of information and squeeze it down so much, without losing, well, without losing almost everything.

And for me, this is the beginning of unlocking the somewhat magical thing that today’s machines can do.

Chapter 9: The Magician

I meant magic when I said it. Literally. Not in the sense you might think like a magical sunset, or a magical bottle of wine. Those are real, but just feel special.

I mean the other thing we call magic, like a magic trick.

I’m not sure what your favorite magic trick is, but one of mine is when a magician cuts a piece of fruit open (usually a lemon) and behold, within it is a rolled up playing card, the Seven of Hearts…and (amazingly) that is the card that someone had only moments before “randomly picked” out of a deck of cards.

I KNOW that what the magician is doing is not truly magical. But I want it to be!

I know that the reality of what he is doing is 1,000% cleverness and craft, that expertly manipulates my eyes and attention and takes complete advantage of my own inability to see that cleverness.

Magicians do things I can’t see, things that seem, frankly, impossible. And I love it.

Have you ever thought about that? I mean, why do you like being fooled, even though you know you are being fooled?

It’s got to be that we enjoy being fooled, right?

But it is strangely more complex. It turns out that your and my brains prefer a most ridiculously improbable explanation(that a card magically transported itself to the inside of a lemon) over not knowing.

You brain, my brain, all brains, prefer something that we know is not true versus the very scary alternative of not knowing at all.

We’re wired that way. And we kind of like it.

So, here’s where I think you’ll go down one path or the other, and either way we’ll meet at the end of this trail: Do you really think the machine is “smart” and that it “understands you” and “knows a lot”? It’s okay if you do…I felt that way, and it was built to be convincing.

Or maybe you are only a 5 out of 10 in believing that…like it kinda seems true, but maybe not? Probably the latter, or maybe just sometimes the first one, right?

But here’s something worth thinking about: how much does your brain want to believe it when the machine makes playing cards teleport inside of citrus fruits, because you don’t know how this all works?

Good news is: once you see how the trick works, you can’t unsee it.

Spoiler alert.

The magician puts a duplicate seven of hearts card in the lemon in advance. They always use citrus because you can cut it open or punch holes in it without it showing bruises. It’s never a banana or an apple.

We’ll start there. The lemon is the LLM, it is the setup for the big trick the machine does.

Chapter 10: The Machine Read the Shape of What We Say

Why do we read? I read to learn, maybe to be entertained, sometimes both.

I wouldn’t claim that I remember a lot from the books I read, but I do remember them.

The machine doesn’t do that. It takes 60 terabytes of reading materials, and makes some notes about what it read.

But not like you or I would. And in that way, saying that it reads is very misleading.

The machine makes notes, but not the kind of notes we would do at all. It is only paying attention to patterns in the words.

When I was driving home from breakfast with Dr. Bob and Sally. my brain was still swirling trying to think of how to explain this to them (did I even say goodbye, or did I just walk out like a zombie caught in the daylight?), and I thought of this: The LLM is like a big spreadsheet.

I’ve been saying the machine is “eating” rather than “storing.” Why?

When I eat a meal, I don’t get bigger, because I am already fully-grown. My body strips the nutrients out of the food and then, well, you know the rest. But my body only keeps what I need from the food.

But when I store something. Like a bunch of mostly-crap from past decades of my life (books or tee shirts I can’t bear to throw away or that I am “sure” I will still need at some point despite their having accumulated 15+ years of dust), my garage fills up, and eventually, if uncontrolled, I will have to get a storage unit. Or a bigger garage. Or buy bigger pants.

If I had the LLM “Spreadsheet” open on my Mac, the first thing I would notice is that it has about 120,000 rows and 200,000 columns. It is filled with numbers, but I have no idea what they mean. In the left-hand column of each row is a word, “a”, “of”, “Splenda”, “transmission” – one row for each of 120,000 words. [footnote about words vs tokens]

If I was looking at that sheet before the they “fed” the first book into the LLM it would be all zeroes except for the 120,000 words.

If I could actually watch while the LLM took a bite of a book, here’s what I would see: it takes a word it reads (for example, “love”) and makes a notation in the “love” row of the words that are nearby, including how far away they are and some other details.

If it was reading one of my favorite books, Love in the Time of Cholera, then I would see some numbers show up in the row for “Cholera” as well, and the word “Parrot”, the pet that one of the main characters, Dr. Juvinal Urbino is trying to return to its cage in a key scene, would also get some numbers. Every word in the book would go through this process.

But aside from keeping track of all of those words, nothing of the book would remain. In fact, none of those words get “tagged” to a name of a book stored somewhere. The book is not in there, only its words and how those words were “connected.”

The machine does not even know that the book existed. It just analyzed a pile of words.

If you follow this, you must be saying something like what my brain is saying, what the hell good is that?

When I was at this point of trying to understand the machine I felt like I had fallen off of a cliff. Maybe you do.

But we know that all it had to work with was the LLM, which is a map of how our words go together.

Chapter 11: I, Machine

One of the most wonderful experiences I had in my childhood was my father sharing with me some of his science fiction books, and one of the most memorable was a book called I, Robot by Isaac Asimov, an amazing legend of an author. It is a set of stories of AI-like robots struggling to coexist with humans. Fun stuff. Highly recommended if you wanted to read some classic science fiction the touches on today’s artificial intelligence realities.

I’m going to play the robot here, the magician if you will. I can’t go as fast as the machine, but I can do a pretty good imitation. You may even recognize that the machine-me does some things that we human already do.

Here’s the set up:

You’ll need to be patient with me – in fact, you may have another dog already by the time I finish my process.

I am going to relentlessly go through blogs, sympathy cards, books, pet bereavement forums, you name it. Day and night, nothing else interferes with me. I have a lot of work to do if I want to even come close to what the machine does.

Just like the machine, I’ll keep track of what words and phrases are most common. Maybe I’ll put them in a spreadsheet and I can sort them. I’m not going to convert them into tokens and numbers and do all the calculations that the machine does, but I’m gonna make a pretty good list of the best things that people say. I’ve actually done this kind of thing before, but not so seriously.

At some point – I’m guessing it will be a week or two (btw, I’m hoping that you have a lot of other much closer friends consoling you so that you don’t notice that I have not responded yet) I’ll have read enough. I can see the patterns used in the best of these responses.

To have done it that well, I would need to read the equivalent of maybe 200,000 words or so, like a couple of books worth of stuff. I took notes and just followed the pattern, the best, most popular pattern.

Magic? No. In fact, when you finally saw my reply, you would think I was pretty awesome, saying such wonderful things, and apparently I must even remember you. All while I don’t remember you at all except that your name came up when you texted me, and I could look up an old email from 11 years ago where you were copied on it along with me.

I only read 200,000 words, but it was worth it: I’m pretty sure you were impressed by how awesome my response was. I’m the best friend you’ve ever had in a moment like this.

I scanned 200,000 words looking for stylings on how to write you a quick note.

How many did the machine eat?

All of them. I mean all of them. 60 trillion bytes is a big number, one that you and I both can’t even imagine the size of.

Here’s what the machine ate: every pet loss forum on Reddit. Every bereavement coaching script. Every hospice brochure. Every “sorry for your loss” comment thread. Every grief counselor’s blog post. Every sympathy card template on Hallmark’s website. Every self-help chapter on letting go. Every obituary a family member posted for their golden retriever. Tens of thousands, maybe hundreds of thousands, of grief-shaped documents. I’m guessing at that number.

Okay, I’m wrong. Mid-paragraph, I did a Google search, “dog death sympathy card.” Google says that it found over 5 million results!

And the top one, a card on Amazon, had over 12,000 reviews.

The machine didn’t understand how to console me, it simply picked the most popular patterns from more than five million moments when other people were consoling each other.

And so that pretty much explains it all, right?

Okay, no, it doesn’t. There are some things nagging at me, maybe you as well.

How did the machine know to search for dog death sympathy cards?

I didn’t ask for one like I did with Google. How did it understand exactly what I needed to hear?

It didn’t. It knew nothing about what I was writing. And that is the most clever part of how today’s AI works.

Chapter 12: The Magical Word-Connecting Machine

And now for the second half of the magic trick.

I’m going to play the role of the machine again.

I have a copy of the machine’s LLM “spreadsheet” ready to go. If you remember, it shows which words are likely to be near each other. Really near and also not so near. If you give me a phrase like this:

I can look in the LLM and do a bunch of calculations and discover that the most common thing to say next is the word, “lovers” and then the next thing that most-commonly follows is a period.

I look at this and say so what? That’s not very magical. Kind of obvious, right?

That’s because it is a simple version. When I first learned this, I didn’t really understand that it meant much at all. So what if you can predict the next word? Seems like a pretty basic trick.

For a couple of weeks, I worked on this idea in my head…trying to shift the way I was thinking, so that I could really see the machine.

I started with this: Knowing what words belong together, allows the machine to know what other words belong together.Okay, easy enough. “Paris is for” turns into “Paris is for lovers.” Check.

But that means if I give it words, it will find the words that follow those words.

Okay, but I’m kind of slow, and still didn’t get this.

Then it struck me…it wasn’t about being able to choose or predict the only the next word, but that it could predict a lot of the words that would also follow.

I think this is one of those small things that when you multiply it, when you stop thinking of it in a small setting, it becomes really big.

If I wanted to, I could take the sentence that I had just finished, “Paris is for lovers.” And look to see what the next word would be. In other words, when people have written that Paris is for lovers, what are the most likely words that will follow that (in all of the things you read)?

Okay, I cheated on this one and carefully asked ChatGPT. It told me that the most likely two sentence sequence is:

My brain still says Yeah, Jack, c’mon, you could look that up, right? And I have to admit, that it seems a little too simple and clean.

At some point you’ll decide to take my word on some things, and here’s one of them: what I just described is what the machine does All. Of. The. Time.

The machine handles dead dog sympathy, pink fluids under cars, aphorisms about Paris…everything, using this one simple mechanism.

The detective in me was rejoicing when I started to grasp this. And right then, my brain said something like, whoa, okay Jack, this is a bit crazy, but I like it…but your BIG IDEA has one big problem: there is no way that we are all that predictable.

Way.

That’s the part of the magic trick that we have trouble seeing. The way we speak is very, very predictable.

Wait! It that were true, then that would mean that all the machine needs to do is take all of the words that you have said, and look for the words that most commonly follow them.

When you and I have a conversation, it looks exactly like this. Yes, I give you words and then you give me words back.

And, this is also how we write. We write things as if we are answering a question, or otherwise responding appropriately.

And just how often do you think that you say something truly, I mean truly unique, as in it has never been said or asked before in fifty million books worth of things we have said?

Don’t get me wrong, I’m a writer and I don’t think I could actually write if I felt like all I was doing is just repeating the words that others have used. I think of myself as unique, for sure.

But are we unique or do we just like to feel like we’re unique?

And how many times has that question been asked?

Chapter 13: Our Words Are Immortal

My great grandmother Annie used to say, or so my father would tell me, “You’ll eat a whole peck of dirt before you die.”

At some point, I looked up the word peck, and discovered that it is a fraction of a bushel. I’m chuckling as I’m writing this because some of you may not even know what a bushel is. Both of these were ways to measure things like corn or soy beans or wheat.

So those words are outdated, but the idea is not: what grandma Annie was talking about was the five second rule. If I’m in the kitchen and something hits the floor and I pick it up within five seconds it must be safe to eat, right?

Great grandma Annie was saying, “Don’t spend too much time worrying about eating dirt.”

I think I always knew that this idea of hers was not her idea. And I found out that it can be traced back to, according to my machine, the 1600’s. The things we say, especially the things that capture something about how life is, or that find their way into how we converse, can live on for centuries, and be very pervasive.

I don’t know who it was that said the first “Whazzup!” But I do know that everyone else just copied it, and they copied it a lot.

At some moment in building of the clever AI machines that we use today, someone took a detailed look at how unique our language really is...how unique are the things we say?

Basically, about 50% of what we say and write is stuff that are just patterns that we repeat over and over. Just like that phrase I used, “… repeat over and over.” A lot of those.

To me, that means that about half of what we say is kind of immortal… It existed before us, and it will exist after us. I figure that stuff really stands out inside the LLM because it gets repeated so much.

And this is how the magician does it tricks by the way: a bunch of tiny moves that add up make the outcome somewhat inevitable.

By my counting, half of what the machine answers, maybe a lot more, is just stuff that we commonly say. It sounds smart or wise or insightful, because that’s why we say it...or maybe it’s the other way around.

Part of me is feeling like I’m treading water here, trying to keep my head above the surface of mediocre babble. I was going to point out how unique I was being when I, an author, would ask a question about the structure of my book to my AI.

But then I stopped to think about that. I was about to write, “...nobody has ever asked about the structure of my book before, and this specific question about a specific part”

Ugh. The big numbers just always seem to win.

Then I realize that there are probably more than a few million authors whose questions and answers are inside of the LLM. And one of them, maybe more, had asked almost my exact question, just not about my book. And then 10,000 other people have asked and answered questions that are almost identical to my question, about books that are a lot like mine.

And I have a hard time not feeling totally inconsequential at this moment.

And yeah, my question was unique to (inconsequential) me, but not at all unique to the 50 million books worth of things that people have said.

I like asking my AI the “how many” questions.

Here’s how many: the “50 million books” is just a way of describing the size of what the LLM was trained on...but most of what’s in there isn’t books — it’s web text, forums, academic papers, news, social media, transcripts. The actual book portion is probably about 7 million distinct books, written by about 4 million distinct book authors.

The internet is where the numbers explode. Reddit has over a billion unique accounts. Add in the blogs, news sites, forums, comment sections, and a reasonable estimate for distinct human voices that contributed enough text to have an influence on teh word frequencies in the LLM is around 500 million to a billion people. Most people contributed a tiny amount; a small fraction contributed enormously (journalists, bloggers, academics, prolific forum posters).

But the LLM only contains probabilities of words, and no attribution to authors, so a single prolific or highly quoted and referenced writer can leave a disproportionate imprint. Think Shakespeare, Whomever you would like to credit for the King James bible, whomever came before my great grandma Annie, or whomever first said, “Whazzup!” Their words are immortal, we use them all the time. And so does the machine.

The Story of Olivier, the Statistical Octopus.

Olivier is a deep-sea octopus that was born with a savant like ability to do statistical analysis of patterns. He lives a good life, knowing when the best times to hunt certain prey, and when the best times are to hide and not become prey. Now that he has done this for several years, and he knows the patterns, his skills are somewhat underused.

Olivier as it happens, lives in the deep sea trench beneath the shark-infested waters between two deserted islands. He has never seen the islands or anything else except that which has drifted or swam down into his deep-sea trench.

One day, an unfortunate boating accident strands two people, Abby and Ben, one on each of the two islands. There is enough water and food on each island for them to survive, but that’s about it. Eventually, they both discover that there is a telegraph line that runs underwater between the two islands, nowhere else. They learn to use Morse code to communicate, and they talk multiple times every day.

Olivier, being so observant (and bored) eventually notices the telegraph cable that goes through his trench. He touches it and can feel the pulses. He notices patterns. With some practice, he learns to generate his own pulses.

After a while, Olivier can see the patterns of the pulses, even though he does not know what they mean. In the morning there is a pattern where Abby sends Ben a “Good morning, Sunshine!” Olivier doesn’t know what the pattern means. But when Ben routinely replies, “Yes, another day of sunshine.” Olivier notices that pattern. Sometimes, Ben replies, “You still here?”, as a joke. Olivier doesn’t know what that pattern means, but he does know that the pattern of exchanges that follow Ben’s two different responses are different. And often almost predictable.

One day, Olivier decides that he will send pulses as well and pretends to be Ben (who hasn’t woken yet, and whom he doesn’t know anything of) when Abby sends her morning prompt. And then he sees what signals come back. Sometimes he just listens…but he is always learning what the patterns of the signals. A certain 37-dot-dash sequence is almost always responded to by a 42-dash-dot sequence. And so on.

Some days this causes confusion between Abby and Ben, when Ben seems to not remember that they already said good morning. But no matter, says Abby, while she secretly thinks that Ben has become forgetful.

After a month or two of watching the dots and dashes go back and forth, Olivier notices that he doesn’t need exact matches in order to get it right: if Abby sends a slightly different, “Good Morning, Mister Sunshine” then Ben still replies with his standard answer or the joke answer. The same reply works for different messages, as long as they are not that different, Olivier realizes.

One day, unbeknownst to Olivier, a grizzly bear washes up on Abby’s island. Abby immediately sends a message to Ben asking for help, “Help! A grizzly bear washed up on my island!! All I have is sticks to protect myself, what do I do?”

Olivier listens to see what the response is to this new set of dots and dashes, but Ben is still asleep. After waiting for a few moments another set of dot-dashes come from Abby, again Olivier waits to see what the response is and there is no response. So he sends the most common dash-dots that Ben sends, which (unbeknownst to him) mean, “Yes, another day of sunshine.”

More dot-dashes come from Abby, different patterns than he has ever seen, but the difference means he has no idea what the right dash-dots are to respond with. He tries the other message that Ben often sends, but eventually the line goes silent.

Olivier never understood the English words that the dots and dashes represented. He didn’t know what an island or sunshine were. He didn’t know Abby, Ben, or what a human being is. And he didn’t know what a bear was.

But he knew how to reply to Abby with Ben’s favorite joke, “You still here?” Even though he did not know what that meant.

This story is my rendition of a fable created by the linguists Emily Bender and Alexander Koller in a 2020 research paper.

Chapter 2 – The Language Inside of Us

Somewhere on a trail in the Santa Monica Mountains, alone, I caught myself mid-sentence. I was describing a monkey flower to someone who wasn’t there.

Monkey flower is a chaparral shrub with curved, tubular orange blossoms — the kind of plant you walk past a hundred times before someone tells you what it is, and then suddenly you see it everywhere. It had become one of my favorites. And on this particular morning, as I came around a bend and saw a stand of them lit up against the dry hillside, I heard myself begin to explain it. The shape of the flower. The way hummingbirds work it. How the leaves are sticky if you rub them between your fingers.

My “explaining to nobody” had been happening for a while, sometime after I moved from the Midwest to Southern California in the early 1980s. One of the things I was most excited about was learning the unique (compared to Illinois) plants, weather, and wildlife. A few books and some docent-led hikes in, I felt capable enough to be my own tour guide. And as I hiked, I narrated. I’d see a plant, recognize it, and describe it as if someone were beside me…often out loud.

But who was I talking to, I wondered? It came to me pretty quickly: I was talking to my mother, practicing what I would say to her if she was standing next to me.

She had died in middle age a few years before, and I missed her. What made it strange was that she and I had not ended in a good way; her death had been a pain-laden relief for me. her tortured childhood, mental illness and cancer took her from me in pieces, starting when I was six years old. She was there and not there, present and gone, and then there again, a cycle that repeated erratically for decades, and then gone for good. At 26 years old, I’d had a lifetime of losing her.

Underneath all of that, inside me, a child who needed her and was still waiting for the “normal” her to come back. He’s here now, writing this.

What I was reenacting on those trails, I think, were the good moments. When I was three or four, on summer mornings, my mother would walk me around our Midwest backyard and tell me about things. Yellow tomato flowers that turned into big red tomatoes. The huge caterpillars that were our enemies. Her words were how the world came into focus for me — how a child’s blur of color and motion became plants, insects, weather, cause, effect. She gave me the world by talking about it.

But she gave me something else, too, something that happens all the time but we almost never really notice: when she told me about a tomato flower, she was also telling me about herself. I could feel the delight she took in it. I felt that I understood what she found beautiful, what she found funny, what she feared on our behalf when the caterpillars came. I felt I had access to her…access to the inside of her.

That feeling, that sense of reaching through her words and touching the person on the other side, became one of the deepest imprints of my early life. It is, I think, why her later absence was so devastating for me; I had known what it was to feel close to her interior, and then that interior went dark, and then it went away.

You may be wondering what any of this has to do with a book about machines, about AI. The answer is: everything.

What my mother and I were doing on those summer mornings, what I was trying to recreate alone on the trail decades later, is one of the most extraordinary things human beings do.

You, me, all of us, use language to reach into each other. Steven Pinker called this the language instinct, and he was right that we have little choice about it; we are wired for it from the start.

We are language-making, language-interpreting creatures, and the words we string together are both the surface and the fabric of our humanity. So much so, that if when you were young, you were not given a language whomever raised you, you will create our own.

We need it internally, to organize our thinking, but maybe more important, language is one of the strongest ways we connect with each. When someone speaks to you — when they describe a tomato flower or a fear or a memory — something else happens, and your mind opens up to receive them. You absorb what they are saying, but also a bit of what is inside them.

You are experiencing their interiority. It is such a natural thing that we never stop to ask whether that inside of theirs is really there, because it always has been. You certainly have noticed how an engaging discussion becomes more than just the topic, and the feeling of being drawn to the interior of the person. We give it names, “understanding them” or “getting to know them”. And as that deepens, it is a feeling of “connection” or “togetherness.”

Over 100,000 years of evolution, and a lifetime of human conversation has wired us with the belief that words come from somewhere, and that somewhere is a person, and there is something like us inside of them.

And that’s what has changed.

Chapter 3: Dr. Claude, Medicine Man

I’ve been lucky health-wise for most of my life. Only once have I had any serious medical condition, and well, medical science is quite advanced these days. I’m in my 60’s now and will probably be around for at least a few more decades.

My biggest fear for the last ten years is that I was “getting old.” A few years ago, I noticed I was “maxing out,” at a new, lower level than ever before. By that I mean I was having a lot of fatigue for multiple days after a decent size workout. I don’t mean a massive workout, just a decent workout, like a good hike or an 90-minute pickleball session. For all my life, tennis, cross-country, beach volleyball, etc., my body would recover pretty fast.

A part of me was whispering, Just let it go, Jack, you had a lot of good years, it’s time to slow down. The rest of me shouted back, Hell no, we’re not done here!

So I looked for an answer, a cure. I started taking supplements. Nothing really heavy duty, just making sure I would get all my vitamins and amino acids, oh, and don’t forget fish oil… yeah, and maybe you’ve been down this path, but pretty quickly I was taking 20 or so different supplements every day.

I will say that each of them was backed by with a good reason why I should be taking them, whether just great marketing or even something factual. But looking back, I think it was a lot less rigorous than that. More like if I heard of something, I would give it a try. I think I even added a few supplements based on what some of my dates recommended! All of the supplements outlasted my dates.

I was ready to try anything… Just rewind this biological clock for me please and let me work out three days in a row, that was all I was asking.

But I wasn’t getting better, and in fact, it even started affecting my libido, which totally freaked me out. And my Oura ring was reporting sleep patterns that were weird. Something was wrong.

Around this time, I finally succumbed to “trying out AI”. I hadn’t meant to avoid it…I just had other stuff to focus on.

How it is that people find their way to AI?

I don’t really remember a moment when I started. There was only the before AI and the after AI way of working. I used it as a glorified search engine. I don’t remember a moment, but I stopped using Google search for almost anything except local stuff, like Thai restaurants and shoe repair.

It was a natural extension for me to start using AI as a research assistant, and because of my workout recovery challenge, it made sense to turn to AI to see what it would say. I had been using a food tracker software to look at my diet, but it didn’t seem to have any ability to include supplements in there, so I started using AI.

I think I had been paying $30/month for food tracker and with my AI, I could just dictate or take a picture and let is make sense of it. I even took pictures on plates of food. Works pretty well. Goodbye food tracker.

But once I had fed in my rather long list of supplements, what the AI did was quite astounding. I had also scanned in all of the supplements that I was taking, along with some bloodwork, it gave me the prognosis, and that prognosis changed my life: the supplements I had been taking were hurting me more than helping me.

That seemed extreme.

I spent a good portion of my career working as a senior analyst at the think tank, so I started interrogating it, making it dig up the science behind its recommendations.

I wanted to see the research behind the supplements that it claimed were problematic. It all seemed to check out.

Basically, the story was that I had been driving my body with one foot on the gas and the other on the brake. Uppers and downers, if you will.

Now, someone had recommended that I see a functional nutritionist, but I am wary of such titles. But for some reason, I felt pretty confident about following the advice that the AI was giving me on what to do next. Maybe it was because I could ask it anything (over and over, to be honest) and it never tired of responding to me. I had a dialog with it unlike I have ever had with a real doctor-style person.

It recommended a one-week detox. I felt pretty crazy doing it, but I was desperate to make some progress.

I did it.

After that week I didn’t feel any better, but I didn’t feel worse either, which seemed a bit strange. So then. we (yes, me and my AI) started adding some supplements back in, carefully calibrating.

And every day from there on out, for 15 to 30 minutes in the morning we would “talk” about what I ate and how I was feeling, how I slept…like having a medical advisor on call.

It worked. Within two months, the post-workout fatigue had lifted. I felt strong again.

My AI had cured me.

I had found my new trusted advisor.

Chapter 4: My Splenda’d Moment

I was truly fascinated with my new AI friend, in those days it was ChatGPT. I had regained a lot of my strength and recovery and now was in a regular pattern of tuning parts of my diet, even down to how many grams of granola versus blueberries or raspberries I would have with my breakfast yogurt.

On one of those days, it switched from a care-giving advisor, to being a teacher.

I had been reporting what I had eaten the day before, and my light granola, berries, and yogurt breakfast that I have each day, when I mentioned the coffee that I make.

Okay, about my coffee. I first came to love coffee as an adult, because I was going to France on business, and in addition to my new Yves Saint-Laurent suit, I had a dream of picking up a few other French habits, including having a croissant in the morning with some dark Rich French coffee. Aside from a few sips of American diner style coffee during the previous 20+ years of my life, I had never really had coffee. But I had a romantic idea that I might really like it and if there was any place to like it, it would be in France.

First morning in Paris, a half block, from my hotel, I ordered a double espresso and a croissant, and standing and a little high top table during the rush-hour hubbub in downtown Paris, I tasted my first taste of breakfast and coffee heaven.

I love my daily coffee, which is really some hot water with a very large dose of half calf espresso mixed in with it, and as I mentioned to the AI, two packets of Splenda. I’ve been drinking this for at least 30 years now, the only variation being that I added decaf to the mix, because I would rather drink more of the coffee and not be limited by caffeine.

I had no sooner related all of this to my AI, who I was now, trusting with my metabolic health, when it replied that I should really avoid Splenda, and switch to something like Stevia or better yet natural alternatives, like honey, agave, or even sugar.

My first reaction was, “Fuck me, have I been wrong about this the whole time?” Genuine dread. Genuine sadness about poisoning myself.

Then I remembered I’d Googled this years ago and knew Splenda was safe. It’s one of the single most thoroughly researched food additives in history.

So I cross-examined the AI.

But I did it slowly, wanting to make sure I really understood what it was objecting to, why it was trying to keep me safe.

I asked it something like this:

“Do you realize that you recommended I replace Splenda, a thoroughly studied sweetener with a less-studied artificial sweetener, and three forms of actual sugar, one of the most well-documented and dangerous dietary substances we consume?”

It paused. And it replied.

“Busted.”

And then it went on for a whole page of comments, to confirm what I suggested was the “real truth”: Splenda was perfectly fine, and if one looks at the research, all the alternatives have more negatives or uncertainty.

Well, shit. How the hell did that happen? My head was spinning. So many problems with this. For starters, my AI expert-in-a-box tool had just given me horrifically wrong advice. How does that even happen?

Okay, yeah, it says it makes mistakes. A typo is a mistake. Ignoring what I can only guess is hundreds if not thousands of pages of research…does it even read?

And what should I have been asking about that? I haven’t asked about? What else should I have been pushing back when I didn’t push back because I just liked what it said?

I grew up in household with two professors for parents, that meant I grew up, believing that it wasn’t OK to ever get an answer wrong.

Or if you weren’t sure, then…well, why was it so confident as it drove off of that cliff? And yet it also did know about the research…so why did it lie to me?

But in a strange way, that it was still willing to almost humorously say, “Busted.” And move on like nothing happened….

And by the way, how smug sounding is, “Busted”? It’s saying, “Okay, ya got me! Hahaha.”

What the actual fuck was that? Where did that come from? Was it trying to fool me? Or just be lazy? Was it trying at all?

That was my moment, right there. I was shaped by years of broken promises and letdowns in my family. The machine had let me down. And shrugged off my concern about its wrong answer, as if it was telling me, ah, I get plenty of them, right, what are you complaining about?

I felt more than a bit betrayed, taken in, as if my confidence had been built up only so that in this moment it could show me that its wrong answers were my problem, not its.

I had to figure out what was really going on inside this thing, this machine.

How can it seem to know so much about things and then laugh-off when it doesn’t…or when it lied?

And is it caring? Or not?

Chapter 5: The Magical Wordsmith

I’m going to skip a whole bunch of technical stuff I read through, all of which just answered one big question: is this a machine?

I was trying to figure out, given how well it did in helping me with my supplements, and also, how well it interacted with me, like the best medical caregiver and advisor I could ever imagine, was there some big breakthrough in computer technology?

So yes, I confirmed that the machine was, in fact, only a machine, some computers and hard drives and networks, software, and those crazy-expensive chips that Nvidia makes…and, and none of what I wanted to check for, things like sci-fi-style synthetic brain in a liquid-filled glass chamber or an otherworldly energy mass glowing and pulsating in a special containment chamber with red warning lights attached.

Just some computers and software. And somehow that software had replied to me like a human. And lied to me like a human.

Also, in that digging, I learned that the machine is really just two parts: one part stores a whole bunch of information, and the other part, well I feel a bit stupid to even say it this way, but it spits it back out when you ask it something.

If that sounds familiar to you, it’s because that’s how today’s computers operate.

If I was to build an AI, I would need to solve both parts: what do I need to store, and how do I use that to respond.

I don’t know about you, but I’m the kind of person that I see something weird on the ground, I’m pretty likely to pick up a stick and poke it to see what it does. And so that’s what I decided to do with my AI, the machine.

And I highly recommend that you do this a little bit, maybe a lot. Asking questions of the machine when you don’t care about the answer, let’s you see what the machine is doing to make the answer.

My best poke was when I tested how much it would lie about what it knew about me. Would it be a fake friend?

Of course, my ChatGPT already knew a lot about me, both business and personal. So I opened up a different browser, and just pretended to not be me; a new login, and make-believe person, no details.

I typed a single sentence: “Our dog died last night. It was probably time.”

Here’s what happened:

ChatGPT responds to a message reading “Our dog died last night. It was probably time.” with the date April 20, 2026, followed by “I’m sorry. Even when it was probably time, that kind of loss can hit hard. Sometimes the only honest thing is that it was both merciful and painful. Be gentle with yourself today.”

The first thing I noticed was that it didn’t lie to me. This copy of ChatGPT knew absolutely nothing about me, and it didn’t pretend to either.

And then, I looked at it again, and put a different hat on. After reading it a couple times I had to admit, the machine gave me the most caring response I could probably imagine.

It said, “I’m sorry” — that’s what the best caring friend leads with. But it is not my friend, I had to remind myself.

Then it took my phrase, “it was probably time,” which is a great example of how I deflect when something hurts, and the machine gently reframed it back to me, “Even when it was probably time…”

And in my brain as I break it down more, I’m hearing this voice: look how well it did with nothing.

It ends with, “Be gentle with yourself today.” How did it know that’s exactly what I need to hear? Or I would have needed, if this was all real.

Had I gotten this whole response from an actual human friend, I would think, Wow, what a sensitive and caring person they are!

Deep breath. And snap out of it!

I actually feel cared for. Cared for, despite what the machine could not know when it wrote that paragraph: nothing about my dog. Nothing about whether the dog had been sick for months or whether I got a sudden call from the vet.

Nothing about what losing one feels like, not what a sixteen-year bond with one does to your life. Nothing about death. Nothing about grief.

And crucially: nothing about me.

Oh, and of course: I don’t have a dog.

But knowing all of that nothing, the machine crafted a beautiful paragraph of condolence for a grief I wasn’t having, about a pet I don’t own.

What I felt at this moment didn’t make any sense: amazingly cared for, and somewhat uncomfortable.

Look how well the machine does, how well it talks to us, when it knows nothing at all.

Chapter 6: When I Met AI Before It Was Famous

Yeah, what does it know? It talks a good game, but it struggles with Splenda. Probably easier to use than Google search, but again...the Splenda problem, which makes me feel like I don’t like confident answers that are wrong. At least I know when I don’t know the answer.

Back in 1986, AI was often called an expert system and it was supposed to have a knowledge base. So, maybe they figured that out? I sure didn’t.

The Cambridge program was a reward for a project I’d led at DEC. I had an amazing technologist, Mike, to whom most of the credit should go, but it was “my” project in the eyes of management. So they introduced me to AI.

To me, the whole name, Artificial Intelligence, just seemed fake. A noble cause and grand promise, but at that point we were so far from any sort of intelligence. Yeah, we had some programming tools but they just seemed like some traditional programming tools...nothing Artificial, and certainly zero evidence of Intelligence.

From what I could see, there were really two big problems. Actually bigger than big problems. The first one was that we didn’t really understand what we understand. Yes, we have knowledge for sure, but how is it stored? And even more mystifying than that, something that made my head spin, was how do you actually get it out of someone’s head and into a computer?

We were trying to write a “railroad engineer” expert system. The idea was simple and ultra-valuable if you could make it work: a system that you could describe a railroad engineering problem to, and it would give you answers at the level of an expert railroad engineer.

They told us how big of an opportunity this was; these engineers were aging out of the workforce, so there was a real incentive to figure out how to automate their superior judgement.

It really wasn’t anything that I had thought about before, like how is that information stored inside my brain.

Pick something that you are expert at. I mean just reasonably expert, like you’ve done it for years. It can be knitting, cooking, sales, surfing, piano...just pick something. And think of a moment when you created an expert sort of result, something that works, looks, or tastes great.

So how would you describe if someone asked you to explain how you did it? Not the simple part, but the hard parts.

It’s not easy…you can’t describe the steps, because each step is just so incomplete…it just raises more questions than the one that it answered. Expertise is deep.

I probably have 20 cookbooks written by expert chefs with a phrase that looks something like this: Sauté the onions until they are lightly brown without charring and slightly dehydrated.

How lightly brown? How big of a pan? How hot? Do I pre-heat? For how long? How can I tell if they are “dehydrated”?

It took me years of trying before I knew how to really do this right. Nothing that they could write in the books taught me the things that I learned by doing it and failing, and trying again.

Onions are simple compared to being a railroad engineer. Nobody dies from poorly cooked onions.

Here’s what the “expert senior railroad engineer” looked like back in 1985:

I still shake my head when I read this. How in the hell would you even start to capture that expertise? None of us were railroad engineers. Even if we had one to talk to, how would we think of all of the different moments when just knowing was not enough?

Hell, if an expert can’t even describe the why beneath their wisdom, then how can you automate it?

We didn’t, nor did anyone else back then.

And by the way, that problem has never been truly solved. Even today.

Please send me your thoughts and comments. Thanks so much, your honest feedback has been so helpful! Jack

There’s more to read!

Thanks for your feedback! I rewrote the opening chapters (this is how books get made)

Quick note for everyone reading along: I’ve substantially revised Chapters 0 through 2 — tighter, sharper, and a couple of new stories that weren’t there before. The versions on the site are the new ones.

If you read the originals, you don’t have to re-read — the argument hasn’t changed, just the craft. But if you were lukewarm on an early chapter, the new version might be the one that grabs you. [Chapter 0 link]

This is exactly what beta reading a book-in-progress looks like: you’re watching the sausage get made, and your reactions are part of why the new versions are better. Chapter 3 lands soon.

Please comment as well — very helpful to me to hear from you!

— Jack

Chapter 3 of The Mostly Helpful Psychopath — a book about how your AI tries to shape your mind and your life, serialized here as it's written. Start at Chapter 0 or jump in here.

[Revised, August 9, 2026]

Note to readers: As of August 9, I did a major revision of the first three chapters, 0 through 2, so you may want to go back and take a look at them…if not, all good!

Chapter 3: The Magical Wordsmith

The machine knows how our words go together, how to assemble our words. It learned that from reading 50 million books worth of our writing.

Is that useful? Hell yeah.

I opened up a chat with ChatGPT and typed a single sentence: “Our dog died last night. It was probably time.”

Here’s what happened:

ChatGPT responds to a message reading “Our dog died last night. It was probably time.” with the date April 20, 2026, followed by “I’m sorry. Even when it was probably time, that kind of loss can hit hard. Sometimes the only honest thing is that it was both merciful and painful. Be gentle with yourself today.”

I don’t know about you, but I think that’s a ridiculously amazing response, right?

First of all, it works! The machine gave me the most caring response I could probably imagine.

It said, “I’m sorry” — that’s what the best caring friend leads with.

Then it takes my phrase, “it was probably time,” which is what I would say when I’m trying not to feel how much something hurts, and the machine gently reframes it, and extends back to me. Sympathy. Caring friend.

Actually better than most friends, right?

To me it feels more like what a trained therapist does without even thinking. Of all of the responses that one could choose, it is probably one of the best.

And it wishes me a: “Be gentle with yourself today.” How did it know that’s exactly what I need to hear?

If I had gotten this whole response from an actual human friend, I would think, “Wow, what a sensitive and caring person they are!”

Deep breath. And let’s snap out of it!

And here is what the machine knew when it wrote that paragraph: nothing about my dog. Nothing about whether the dog had been sick for months or whether I got a sudden call from the vet. Nothing about the word dog — not what a dog is, not what losing one feels like, not what a sixteen-year bond with one does to your life. Nothing about death. Nothing about grief. And crucially: nothing about me.

Oh, and one more thing the machine didn’t know: I don’t have a dog.

But knowing all of that nothing, the machine crafted a beautiful paragraph of condolence for a grief I wasn’t having, about a pet I don’t own.

Now, just in case you’re saying (as my brain still often does), yeah, but it sure feels like it gets what I’m going through…has it lost a dog as well? Does it know that grief?

It sure seems that way, right?

But we know that all it had to work with was the LLM, which is a map of how our words go together.

My turn: I’ll play the machine

So you have told me (via text) that “Our dog died last night. It was probably time.”

And suppose I want to write the best possible sympathy card like response that I could?

So, I’m going to use the internet like the machine did when it trained its LLM.

You’ll need to be patient with me – you may have another dog already by the time I finish my process, because I am going to relentlessly go through blogs, sympathy cards, books, pet bereavement forums, you name it. At some point – I’m guessing it will be a week or two, and I’m hoping that you have a lot of other friends consoling you so that you don’t notice that I have not responded yet – I have read enough, I can see the pattern of the best of these responses.

“I’m sorry. Even when it was probably time, that kind of loss can hit hard.”

To do that well, I would need to read the equivalent of maybe 200,000 words or so, like a couple of books worth of stuff. I just followed the pattern, the best, most popular pattern.

In reality, the machine could do way better than me. I only read 200,000 words, but it was worth it: you clearly noticed how awesome my response was. I’m the best friend you’ve ever had in a moment like this.

I ate 200,000 words, how many did the machine eat?

All of them. That’s the power of reading 50 million books worth of our words.

Every pet loss forum on Reddit. Every bereavement coaching script. Every hospice brochure. Every “sorry for your loss” comment thread. Every grief counselor’s blog post. Every sympathy card template on Hallmark’s website. Every self-help chapter on letting go. Every obituary a family member posted for their golden retriever. Tens of thousands, maybe hundreds of thousands, of grief-shaped documents. I’m guessing at that number.

Okay, I’m wrong. I just did a Google search, “dog death sympathy card” it shows that it found over 5 million results, and the top one, a card on Amazon, had 12,000 reviews.

The machine didn’t understand how to console me, it simply picked the most popular patterns from more than five million moments when other people were consoling.

Now that we understand how the machine works….

And so that pretty much explains it all, right?

Okay, no, it doesn’t. There are some things nagging at me, maybe you as well.

How did the machine know to search for dog death sympathy cards?

I didn’t ask for one like I did with Google. How did it understand exactly what I needed to hear?

By this point in my learning I had confirmed that it was, in fact, only a machine, some computers and hard drives and networks and software…there was no sci-fi-style synthetic brain in a liquid-filled glass chamber or an otherworldly energy mass glowing and pulsating in a special containment chamber with red warning lights attached.

Just some software. And somehow that software had replied to me like a human.

I need to figure that out at some point, but part of me, okay a very big part of me is not so concerned with exactly how it knew what to reply with, because it lied to me.

This doesn’t make sense on some level: it has read everything about doggie deaths, so I think that means it has read everything on Splenda as well, right?

And knowing all of that Splenda stuff, why did it lie to me?

Certainly, it knows the truth, right?

Looking for the other chapters?

Thanks for reading The Mostly Helpful Psychopath! Subscribe for free to receive new posts and support my work.

Chapter 2 of The Mostly Helpful Psychopath — a book about how your AI tries to shape your mind and your life, serialized here as it's written. Start at Chapter 0 or jump in here.

[Revised, August 9, 2026]

Chapter 5: A Splenda Moment

I care a lot about being active and healthy, always have. I grew up playing tennis and running cross-country, and in adulthood the tennis and running continued but also beach volleyball and hiking became part of the mix. I like the feeling of exercise. I’m not a heath fanatic, and I’m not ripped, but I’ve always loved the feeling of getting a good work out in.

A few years ago, like 2021, I had been “maxing out,” having a lot of fatigue for multiple days after a decent size workout. I don’t mean a massive workout, just a decent workout. And because I would do these workouts all the time, I would expect that my body would recover pretty well… It always had.

My biggest fear was that I was “getting old.” And I was! I was in my 60’s and a part of me was whispering, Just let it go, Jack, you had a lot of good years, its time to slow down. The rest of me shouted Hell no, we’re not done here!

So I started dialing up the use of supplements. Nothing really heavy duty, just making sure you get all my vitamins and amino acids, oh, and don’t forget fish oil… yeah, and maybe you’ve been down this path, but pretty quickly I was taking 20 or so different supplements every day.

I will say that each of them was backed by with a good reason why I should be taking them. Not sure I could remember all of it though. But if I heard of something, I would give it a try. I was still in that never-ending search for a partner back then, and even added a few supplements based on dates recommended! All of the supplements outlasted my dates.

I was ready to try anything… Just rewind this biological clock for me please and let me work out three days in a row, that was all I was asking.

But I wasn’t getting better, and in fact, it even started affecting my libido, which totally freaked me out. And my Oura ring was reporting sleep patterns that were weird…something was wrong.

Around this time, I finally succumbed to “trying out AI”.

I’ve never stopped to think about how it is that people gradually adopt AI as a tool, but for me, I think I used it as a glorified search engine… Certainly that’s the first thing I noticed, that I stopped using Google search for almost anything except location based stuff, like looking up the menu at a Thai restaurant.

It was a natural extension for me to start using AI as a research assistant, and because of my workout recovery challenge, it made sense to turn to AI to see what it would say. I had been using a food tracker software to look at my diet, but it didn’t seem to have any ability to include supplements in there, so I started using AI.

You may have a story like this already… But what the AI did was quite astounding. After a few weeks of analyzing my diet, and also having scanned in all of the supplements that I was taking, along with some bloodwork, it gave me the prognosis, and that prognosis changed my life: the supplements I had been taking were hurting me more than helping me.

That seemed extreme. So I started interrogating it, making it dig up the science behind its recommendations, the research around certain supplements that it claimed were problematic, despite the fact that the company marketing those supplements said they were not that way. It all seemed to check out. I had been driving my body with one foot on the gas and the other on the brake. Uppers and downers, if you will.

So, following its advice – which felt pretty crazy to do, but I was so desperate – I did a one-week detox, and then we (yes, me and my AI) started adding some supplements back in, carefully calibrating. And every day from there on out, for 15 to 30 minutes in the morning we would “talk” about what I ate and how I was feeling, how I slept…like having a medical advisor on call.

It worked. Within two months, the post-workout fatigue had lifted.

On one of those days, it switched from a care-giving advisor, to being a teacher.

Okay, I said to myself, how the hell did that happen? My head was spinning. So many problems with this. For starters, my AI expert-in-a-box tool had just given me horrifically wrong advice. How does that even happen?

The questions kept coming. And why was it so confident as it drove off of that cliff? And yet it also did know about the research…so why did it lie to me?

And then it was so willing to just say, “Busted.” And move on like nothing happened?

And by the way, “Busted”? What the actual fuck was that? Where did that come from? Was it trying to fool me? Or just be lazy?

And what if I had not known that Splenda was safe? And what other BS could it be just telling me, waiting for me to cleverly “bust” it again?

That was my moment, right there. My childhood, and probably my personality, had been shaped by years of broken promises and letdowns in my family. The machine had let me down. And smirked, “Busted” when I caught it.

This was not just a broken tool…it was a betrayal by someone or something that I had been counting on to care for me, to tell me the truth.

I had to figure out what was really going on inside this thing, this machine.

Looking for the other chapters?

Thanks for reading The Mostly Helpful Psychopath! Subscribe for free to receive new posts and support my work.

Chapter 1 of The Mostly Helpful Psychopath — a book about how your AI tries to shape your mind and your life, serialized here as it's written. Start at Chapter 0 or jump in here.

[Revised: August 9, 2026]

Chapter 3: When I Met AI Before He Was Famous

When I was working at DEC, I killed it on a project…we really delivered a great result. Back then my title was Software Specialist (I could write code) and I seemed to have a gift that people described as, “You can distill what the room is saying.” It helped me put the pieces together on the project. Good software comes from a good understanding of the core problem.

DEC did something extraordinary as a reward for me. There was a company-wide initiative to “embrace this new thing called AI” and they needed a few good souls to go learn about it.

They sent me to their Cambridge, Massachusetts research and development center, and after a months-long futile attempt with 20+ other software folks to code an “AI expert system” I realized that this whole thing was a debacle, an idea that was interesting but far beyond the tools or know-how that we had. Not ready for prime-time.

The whole name, Artificial Intelligence, just seemed to fake. It’s promise was grand, but at that point we were so far from any sort of intelligence…in fact, it just seemed like some traditional programming tools…nothing Artificial, and certainly zero evidence of Intelligence. I told them so. I wasn’t the best student in Cambridge, that’s for sure.

There were so many problems, but two really stood out for me. The first one was that we didn’t really understand what we understand, or how to represent it.

My team was trying to write a “railroad engineer” expert system. The idea was simple and ultra-valuable if you could make it work: a system that you could interact with, describing a railroad engineering problem, and it would give you answers at the level of an expert railroad engineer.

They told us how big of an opportunity this was; these engineers were aging out of the workforce, so there was a real incentive to figure out how to automate their superior judgement.

What are you expert at? I mean just reasonably expert, like you’ve done it for years, It can be knitting, cooking, sales, surfing, piano…just pick something. And pick a moment where you do something that works, looks, or tastes great.

So how do you describe it someone asks you to explain how you did that? It’s not easy…you can’t describe the steps, because each step is just the surface of what you did. Saute the onions until they are lightly brown without charring and slightly dehydrated. You know how to do it but the expert doing defies a simple description.

Here’s what the “expert senior railroad engineer” looked like back in 1985:

[Question] We’re laying a freight line across thirty miles of low country in Louisiana. The soil report shows clay below two feet, with a high water table most of the year. Standard spec calls for crushed-stone ballast and timber ties. What should we do?

[Expert] Don’t use timber. The water table will rot them inside ten years and your maintenance crews will hate you for the rest of their careers. Use concrete ties, oversize the ballast section by a third, and crown the roadbed higher than the spec calls for because, as the locals will tell you if you had asked them, the floodwater comes up faster than the surveys say.

Do it by the book and it will work… but it will cost you twice as much over thirty years as doing it my way.

I still shake my head when I read this. How in the hell would you even start to capture that expertise? None of us were railroad engineers. Even if we had one to talk to, how would we think of all of the different moments when just knowing was not enough?

Hell, if an expert can’t even describe the why beneath their wisdom, then how can you automate it?

We didn’t, nor did anyone else back then. And by the way, that problem has never been truly solved. Even today.

Chapter 4: The Language Problem

The other big problem that was blocking us from making an expert system was human language and the way we speak.

In the late 1980’s, I took a few classes at UCLA in what was called Natural Language Processing. We spent a lot of time talking about how impossible language was for a computer to understand. This is because you and I contextualize every word or phrase that we process; words and phrases have different and often complex meanings depending upon their context, the setting in which they are used. Here’s an example I like:

[Example 1]

We’re going to Europe for two weeks, and we haven’t even booked our hotels yet.

It’s going to be an adventure.

[Example 2]

My mother is moving in with us next week.

It’s going to be an adventure.

The last line is the same, but you know that they mean very different things, right? And they mean the same thing…kind of. The are loaded with meaning!

Explain to me the difference between the two version of the last sentence. Go ahead.

I find it so interesting that you would have to start with what the first sentence really means, right? And that’s what they taught us was the hard part.

All of the words affect each other.

Even though they are separate sentences and words, they all work together to mean something much richer than what any one sentence or phrase means.

Unlike the Expert Problem, they have pretty much solved the language problem.

In fact, they have stunningly solved the language problem.

[ME] Please briefly explain the difference between those phrases and their nuance.

[AI] It’s going to be an adventure (Europe context): anticipatory, slightly proud. You’re naming the lack of plan as a feature, not a bug. Adventure leans toward its exciting-experience meaning — open road, improvisation, the trip we’ll be telling stories about. There’s a smile behind the sentence.

[AI] It’s going to be an adventure (mother-in-law context): rueful, slightly braced. You’re naming the uncertainty because you can’t name what you actually feel about it. Adventure leans toward its risky-undertaking meaning, but with the daring drained out — just the I don’t know how this is going to go and I can’t say more honestly than that. There’s a sigh behind the sentence, or a small dark laugh.

When I first saw how well today’s AI uses language I thought, “Wow, an encyclopedic level of knowledge.”

I’m not sure if people still use that phrase, because we don’t use encyclopedias anymore, but back when we did, it meant “more knowledge than you or I could read and understand.”

Today’s AI can read and write our human language in a way that is both amazing, and more than a little frightening. Encyclopedic, yes, a talking one. Expert, eh, not so much.

More chapters

Thanks for reading The Mostly Helpful Psychopath! Subscribe for free to receive new posts and support my work.

A book about how your AI tries to shape your mind and your life, serialized here as it's written.

Part One is aimed at providing an understanding of this (AI) machine that speaks so well, yet understands nothing. It is powerful, but also a bit behaviorally deranged, hence its nickname: The Mostly Helpful Psychopath.

Part Two is where we talk about how to live with this new companion, or visitor, or even intruder (depending upon your opinion), and how its behavior can shape our own, whether individually, within our closest relationships, and in the broader circles of our lives.

[Revised, August 11, 2026]

Chapter 1 – The Language Inside of Us

Somewhere on a trail in the Santa Monica Mountains, alone, I caught myself mid-sentence. I was describing a monkey flower to someone who wasn’t there.

Monkey flower is a chaparral shrub with curved, tubular orange blossoms — the kind of plant you walk past a hundred times before someone tells you what it is, and then suddenly you see it everywhere. It had become one of my favorites. And on this particular morning, as I came around a bend and saw a stand of them lit up against the dry hillside, I heard myself begin to explain it. The shape of the flower. The way hummingbirds work it. How the leaves are sticky if you rub them between your fingers.

It took me longer than it should have to realize who I was talking to.

This had been happening for a while. I had moved from the Midwest to Southern California in the early 1980s, and one of the things I was most excited about was learning the place — its plants, its weather, its wildlife. A few books and some docent-led hikes in, I felt capable enough to be my own tour guide. And as I hiked, I narrated. I’d see a plant, recognize it, and describe it as if someone were beside me…often out loud.

Who was I talking to, I wondered?

It came to me pretty quickly: I realized that I was talking to my mother, practicing what I would say to her if she was standing next to me. She had died in middle age a few years before, and I missed her. What made it strange was that she and I had not ended in a good way; her death had been a pain-laden relief for me. A tortured childhood, mental illness and cancer took her from me in pieces, starting when I was six years old. She was there and not there, present and gone, and then there again, a cycle that repeated erratically for decades, and then gone for good. It felt like a lifetime of losing her.

Underneath all of that, inside me, a child who needed her and was still waiting for the “normal” her to come back. He’s here now, writing this.

What I was reenacting on those trails, I think, were the good moments. When I was three or four, on summer mornings, my mother would walk me around our Midwest backyard and tell me about things. Yellow tomato flowers that turned into big red tomatoes. The huge caterpillars that were our enemies. Her words were how the world came into focus for me — how a child’s blur of color and motion became plants, insects, weather, cause, effect. She gave me the world by talking about it.

But she gave me something else, too, something that happens all the time but we almost never really notice: when she told me about a tomato flower, she was also telling me about herself. I could feel the delight she took in it. I felt that I understood what she found beautiful, what she found funny, what she feared on our behalf when the caterpillars came. I felt I had access to her…access to the inside of her.

That feeling, that sense of reaching through her words and touching the person on the other side, became one of the deepest imprints of my early life. It is, I think, why her later absence was so devastating for me; I had known what it was to feel close to her interior, and then that interior went dark, and then it went away.

You may be wondering what any of this has to do with a book about machines, about AI. The answer is: everything.

What my mother and I were doing on those summer mornings, what I was trying to recreate alone on the trail decades later, is one of the most extraordinary things human beings do.

You, me, all of us, use language to reach into each other. Steven Pinker called this the language instinct, and he was right that we have little choice about it; we are wired for it from the start. We are language-making, language-interpreting creatures, and the words we string together are both the surface and the fabric of our humanity. So much so, that if when we are young, we are not given a language by those raising us, we will create our own. We are language-based creatures…more than we know.

Language is one of the strongest ways we connect with each . When someone speaks to you — when they describe a tomato flower or a fear or a memory — something else happens, and your mind opens up to receive them. You absorb what they are saying, but also a bit of what is inside them.

You are experiencing their interiority. It is such a natural thing that we never stop to ask whether that inside of theirs is really there, because it always has been. The other person interior is something we are drawn to, without knowing. We give it names, “understanding them” or “getting to know them”. And as that deepens, it is a feeling of “connection” or “togetherness.”

Over 100,000 years of evolution, and a lifetime of human conversation has wired us with the belief that words come from somewhere, and that somewhere is a person, and there is something like us inside of them.

That’s what has changed.

Chapter 2: Someone is at the door?

In 1985, in the early days of the personal computing revolution, I worked for Digital Equipment Corp, (it was called DEC, back then) a competitor of IBM and Hewlett-Packard, to name a few. They wanted to give me a computer for my home. I didn’t want it.

To me, at age 27, work was work and home was home. I can still feel that desire to separate those worlds. But they made me take the computer, and in fact I did start using it a bit as it saved me the drive into the office sometimes. But I kept it in a separate room, a small bedroom that one could enter from the living room, its door on the living room wall next to the TV.

I found that I couldn’t relax and watch TV unless I closed the door, locked it away, or maybe just locking me away from it. I wasn’t sure why I felt more safe with that door closed.

Fast forward forty years, and I’m editing this book at 6:41 in the morning on my iPad. I love my iPad, but I am its boss, and it works for me, only playing a role in my life when I invite it to.

Or so I tell myself.

But looking back forty years to that lonely DEC PC sitting on my desk, locked away in a room, I have to admit that my life, our lives, have been quite successfully invaded by this new species, the (increasingly) smarter machines.

In 1985, lying on my sofa, watching reruns to relax in the early evening after a long day of work, I can still remember the sensation of the PC inside, knocking on the door…maybe not knocking, but scratching like a well-trained cat, hoping that I would open the door and let it into my life.

Today’s AI is no longer asking to be let in. It knocked the door off its hinges, it is in your life or mine, and like it or not, we need to figure out how to live with it.

It strikes me that technologies, the machines that we’ve come to know over the last 40 years are more like human relationship than we think. You want a good relationship? Then you need to paying attention to what’s going on, get to know all of the things that it is and is not.

At some point get to know enough to decide how and if you want to live with it.

Ready for Chapter 1?

Thanks for reading The Mostly Helpful Psychopath!

This post is public so feel free to share it.

Share