A small autobiography, with footnotes.

Four, then eighteen.

Returns in Chapter 8 (Bayes was right about your mother), where we will derive, from the events sketched here, a working prior that took the rest of my life to update.

My mother died when I was four. My sister was one. I will not dwell on this, because there is nothing useful I can tell you about it. I do not remember her, and the version of grief that comes from not remembering is a quieter thing than the kind people write books about. My body remembers something. A warmth, a smell, a sound. The rest is absence.

My father was alive, but only in the sense that he was technically not dead. He drank, gambled, and slept around, in roughly that order of priority, and in roughly that order of competence. My sister and I were handed to his younger brother, my uncle, in the manner one might hand off a difficult parcel one had not asked to receive and could not legally refuse.

My uncle was not a cruel man. I want to say this clearly, because the temptation to make him a villain is enormous and entirely unhelpful. He was simply a man keeping a ledger, and we, his brother's children, had landed in it firmly in the red. He fed us. He clothed us. He paid the school fees. He also reminded us, gently and frequently, that he was doing all of this, and that it was costing him. The reminder was the price of the keeping.

I was a terrible student. I want to say this plainly too, because it matters. The school I attended ran on rote memorization, which is the educational equivalent of asking a child to fill a sieve with water and grading him on how full the sieve looks at the end of the year. My sieve, for reasons no one was curious about, was extremely empty. I failed everything except one subject, which was computer science, in which I did not merely pass. I loved it the way you love a language that finally, for the first time, fits inside your mouth.

This was the 1990s in India. Computers were just arriving, blinking uncertainly into a country that had not yet decided what to make of them. Programming was a curiosity. It was not a respectable career. My uncle, having watched me ace one subject and fail twelve, drew the only conclusion a man with a ledger could draw. After the first year of engineering college, he closed the account. He told me, calmly, that I had one week to pack and leave, and that he would use his contacts to find me a job and a place to live before the week was up.

I did not use the week. I walked, the same afternoon, to a small computer training institute I had visited before. The owner offered me a job on the spot. Part time instructor in the mornings, part time lab assistant in the afternoons. The pay was sufficient for rice and not much else. I went home, packed my things, which consisted of a few clothes and two or three pairs of shoes, including a pair of chappals, and moved into the back of the institute.

I was eighteen. I owned what I was wearing and a small bag. I had, depending on how you counted, either lost everything or lost nothing, because I had never quite been allowed to have anything in the first place.

What I had instead, although I would not understand this for several months, was an entire room of computers and an entire wall of textbooks, both of which I now had unsupervised access to every night after the last student left. The universe had, by some strange administrative error, locked an anxious eighteen year old in a library and forgotten to retrieve the keys.

The Bayesian prior the school had installed in me, namely this kid is not going anywhere, would take the next twenty years to update. We will return to the math of why, in Chapter 8. For now I will only say that priors written before you are old enough to argue with them are extraordinarily difficult to update later, even with very large amounts of contradictory evidence, which I would proceed to spend the next two decades generating.

A self defined by minus signs.

Returns in Chapter 7 (Eigenvectors of the self), where we will discover that defining yourself in opposition to someone is, mathematically, still letting them define you.

The adult I became was, in retrospect, a piece of software written by a small frightened boy. The boy had observed his father carefully and had reached the only conclusion the data supported, which was that becoming his father was a fate worth any amount of personal redesign to avoid. The personal redesign duly began.

My father drank. I became, and remain, a teetotaler. I have tasted alcohol. I hated the taste. I hated the smell. I was, in this rare instance, handed a piece of luck by my own biology, because the revulsion was immediate and required no willpower to maintain.

My father slept around. I made loyalty the load-bearing virtue of my emotional architecture. I have, in adult relationships, been faithful to a degree that on at least one occasion alarmed my partner, who suspected that anyone this committed to fidelity must be hiding something. I was not. I was simply running the opposite of my father, at full optimization, on every available axis.

My father was lazy. I became ambitious to a degree that, viewed from the outside, looks like a personality trait, and viewed from the inside, looks like a man being chased. My father destroyed himself. I refused to destroy myself, which is not the same as being well, but which is, I will grant the younger version of me, a reasonable opening move.

It would take me a long time, and a great deal of mathematics, to notice the structural problem with this approach. The problem is geometric. If you define your entire personality as the negation of someone else's, the geometry is still theirs. You are standing on the opposite side of their coordinate system, but the coordinate system is the same coordinate system. They picked the axes. You picked the signs. That is not freedom. That is, at best, a very disciplined argument with a ghost.

We will return to this in Chapter 7, when we discuss the directions in which your personality does not change under transformation. For now I will only say that the man writing this paragraph spent approximately three decades being the inverse of a man he barely knew, and that the energy required to maintain a constant inversion is, as it turns out, exactly the energy required to feel exhausted all the time.

The dogs that satisfied the wrong instinct.

Returns in Chapter 8 (Bayes was right about your mother), where we will look at why my DNA produced my father, therefore my DNA will produce another father like him is a perfectly reasonable piece of reasoning with one fatal statistical defect.

I married a woman on the explicit understanding that we would not have children. She did not want children. I did not want children. The agreement was clean, mutually held, and, I now understand, the kind of agreement two people make at the start of a marriage before life remembers it has opinions of its own.

My reasons were technical. I had observed my father. I had observed my own DNA, which was, on the relevant chromosomes, his DNA. I had performed the calculation any reasonable engineer would perform under the circumstances. The probability that I would replicate his failures, given that I shared half his genome, was, in my estimation, unacceptably high. The decision not to have children was, in my mind, a piece of responsible risk management. It was, I would not have admitted at the time, also the act of a man trying to prevent a small child from one day standing in a room and being told he was not the kind of child who would amount to anything.

Four years into the marriage, my wife's elder sister had a son. My wife held him. Something rearranged. The agreement, which had been clean and mutually held, became, overnight, a position one of us had begun to feel was no longer the right position.

The next four years were a slow protracted disagreement of the kind that cannot be resolved by being right, because both parties are right in different coordinate systems. She wanted a child. I did not. The disagreement spread, the way disagreements do, into every other room of the marriage. At one point I refused to have unprotected sex with her, on the grounds that I suspected I might be tricked into fatherhood, which is, in retrospect, not the sentence of a man who is doing well.

We went to a family therapist. The therapist, after listening to both of us at length, made a recommendation that on paper was inspired. She suggested we get a dog. The theory was that the dog would awaken my dormant paternal instincts, that I would mellow, and that I would, over time, come to see fatherhood as a thing I could risk rather than a thing I must prevent.

We got two dogs. I reasoned, with the cool logic of a man with no children, that one dog would be lonely while we were at work, so it would be cruel to get fewer than two. I will spare you the next paragraph, except to say that I loved those dogs. I loved them with a totality I had not previously known I was capable of. I loved them in the morning. I loved them in the evening. I loved them when they ate my shoes. I loved them when they ate the other shoes. I loved them with every paternal instinct I possessed, and possibly several I had imported from a future life.

The therapist's algorithm had executed perfectly. The therapist's algorithm had also, in technical terms, returned the wrong answer, because my paternal instincts were now, definitively, fully satisfied. The dogs had absorbed the entire budget. There was no remaining demand. I wanted children less than ever, because I no longer had the small unfilled space inside me that wanting children would have been filling.

My wife left after eight years of marriage, on the entirely reasonable grounds that her biological clock was not negotiable and mine had been replaced, several years earlier, by two dachshunds.

The day she walked out of our home I had a nervous breakdown that lasted twelve days, followed by eight months of clinical depression. I will not describe either, because describing them would not help you, and because the version of me who survived them is sitting at this desk and would prefer to write the next sentence.

The next sentence is this. The reasoning that led me to refuse children, namely I share my father's genome therefore I will repeat my father's mistakes, is a perfectly recognizable piece of Bayesian inference with one fatal statistical defect. We will get to it in Chapter 8. The short version is that DNA is the substrate, not the program, and the men who repeat their fathers are not the men who notice they have fathers. They are the men who do not. I noticed. The noticing was the protection. The math had been on my side the entire time. I just did not, at the time, know how to read it.

The founder who was a single point of failure.

Returns in Chapter 12 (Greedy algorithms make terrible life partners), where we will see why the locally optimal move of do it yourself, it is faster is so reliably a global catastrophe wearing a small confident hat.

I founded a company. I hired, with some care, the best designers and the best programmer I could find. They were, by every measurable standard, excellent at their jobs. Several of them were, in their particular specialties, better than me, which is the entire point of hiring people, and which is, on most days, a fact I could repeat aloud without flinching.

I then proceeded to review every line of code they wrote. I reviewed every design they produced. I flagged errors. I suggested improvements. I hovered. I kept everyone, including myself, on a permanent low simmer of vigilance, on the unstated theory that if I was not personally involved in every decision, the quality would, somehow, by some mechanism I could not name but profoundly believed in, degrade.

The reasoning was, in a technical sense, valid. The reasoning was also, in a technical sense, the reasoning of a four year old in a house that had not caught him. The child who has learned that the only safe operator in the system is himself will, given a company and a payroll, build a company in which he is the only safe operator. This is called, in distributed systems literature, a single point of failure architecture. It works extremely well until the single point fails, which it always eventually does, often spectacularly.

The function being executed was the same function. The arguments were different. As an eighteen year old, the function had been called on survival. As a thirty something founder, it was called on a Postgres migration. The function did not care which. It ran the same way, with the same energy budget, in the same fundamental belief that nobody else would do the thing properly if I let them. The colleagues, who were entirely capable of doing the thing properly, sensed the lack of trust the way colleagues always do, and adjusted downward to meet it. The micromanaged employee is, eventually, the underperforming employee. The founder, observing the underperformance, concludes that the micromanagement was justified. The loop closes.

We will return to this in Chapter 12, when we discuss why the locally optimal action, repeated long enough, produces the globally catastrophic outcome. For now I will only say that the man who could not trust his designers is the same man who could not trust his uncle, his school, his father, his marriage, or his own genome, and that the function being run in all six cases is one function, called with different arguments, all the way down.