(Branch of Reality, Euclid’s Room)

(TODO Image: The hot sweltering summer is outside Euclid’s house. The streets feel like they are melting.)

(TODO Image: Inside his room, dressed in T shirt and shorts, Euclid is doing some programming, seated at his battle station. He is eating a cornet choco ice cream as he does so.)

I watched videos, I read research papers, and eventually I figured out where the world had gone wrong.

[https://i.imgur.com/9a3ECgp.png]

It got stuck on the backpropagation algorithm.

I spent an entire year thinking who to blame for my suicide, and I concluded that the blame should be placed on it. The backpropagation algorithm is what created the deep learning wave, but its successes are what is preventing the world from moving on.

Something better needs to be found. The backpropagation algorithm is enemy #1 preventing the Singularity from occurring.

Suppose I had one of those mythical brain cores from the Simulacrum stories, I could do an evolutionary search for the right replacement. If I had sufficiently powerful hardware I could find something a lot better that does local Bayesian inference, and use those pieces to create a memory system.

The main thing that works in my favor is that in the brain itself the modules are replicated. That means that there exist composable building blocks of intelligence. It is just a matter of finding them. Right now AI chips are coming out, and they are going to make such a search a lot easier thanks to their processing-in-memory capabilities. Doing an evolutionary search for local pieces would be a lot easier if the chip itself has many modules that work locally and independently in a fashion similar to how the brain does.

If I had to make something like a web browser, I’d never get anywhere using evolution, but there is only so much a cortical column replicated a 150 thousand times could be doing in the actual brain.

So one plan is to make money, get rich, and then spend it on finding better brain algorithms. I’ve already accepted that I can’t figure it out by myself, I am just to dumb to reason it out directly.

But fundamentally, this problem is the kind of problem that you can get through with the power of money.

I could do it.

If I had enough money I could cause the Singularity.

But there is something disgusting about this plan, about spending my own money that is. Just why am I taking all this responsibility on myself, like I am some legendary hero of a story?

If I have 100k dollars, and I spend it all to find something good, and it fails, I might as well have set 100k dollars on fire for all the good it did me.

I hate it, I hate it, I fucking hate the thought of that happening! I hate the thought of losing what I've worked so hard for. It can’t possibly be the right path.

I have absolutely no idea how much I’d need to spend. Maybe 100k is peanuts, and I’d need to spend 100 mil to get anywhere! Or 1 or 10 billion!

Even after the pieces get found, I’d still have to do a huge amount of work to put them to good use. I mean, look at real life kids. Nature has found the algorithms, and it still takes a huge effort to raise them. It is not like I’d get a genie that could grant my every wish even if I found them. Doing a lot of work so I can do even more work hardly feels attractive to me.

I could try to do it all by myself, but in the absence of personal insight into the nature of the better learning algorithms, the Singularity looks like an endless black hole ready to suck in my money, effort and time. Sure, I am a pretty good programmer. I’ve made my own programming language in my past life.

But how good do I think I am? Maybe I think I am good, and maybe I could do the work of a few people on the team I am on. But can I replace the effort of 100 people? Or 1,000? I don’t think so.

It is not like I have a huge advantage owing to some insight when implementing evolutionary search either.

The academics working on ML might have become a joke, and the papers they are publishing might be useless, but they do get paid for it. They aren’t expected to spend large amounts of their own money to support their research.

They have a system in place to support themselves.

Meanwhile, I have nothing.

Google could do it. Companies like Google and Tesla could burn 10 billion to try finding the right learning algorithm for their search engines or self-driving cars. What needs to be done is to convince them to use their large computing centers more directly instead of paying researchers to waste their time on personal experiments.

It is not like evolutionary approaches haven’t been tried, but their status is just one out of many. An researcher would try it, make a paper and then move on. If the companies really believed this was the dominant approach that should be pursued things would be a lot different.

I have my goal.

Instead of blind guesswork if I were to act as an individual ML researcher, I should go at it like George Soros. Right now there is hope that if only enough brains are put together, the breakthroughs in AI will come, but I know for sure this is pure fucking hope. What I need to do is cause a change of attitude in the community at large towards their approach to AI.

If I can succeed at that, then Google will hand me what I want free of charge.

Once I have that knowledge, it doesn’t matter what happens later. I will be able to carry it afterwards through all my lives.

The thing is, even if CPUs and GPUs are most suitable towards working with backprop, and better algorithms might require AI chips, I really doubt that even on GPUs backprop is the best we can do. It is really unlikely. Backprop is an optimization algorithm for arbitrary mathematical expressions, while what we really care about is finding out just the right pieces to make the engine run. They are likely to have their own specific ways of optimization and inference.

Well, I have decades to accomplish this, so there is no rush. Let me first attain financial independence. A few years of programming combined with trading should be enough to get there.

By the time my classmates are done with high school, I will have become a millionaire.

(Branch of Reality, Euclid’s Room At Night)

(TODO Image: The monitor is showing Euclid’s brokerage account balance. It says 621,189$. The date says June 2029.)

Huhu. Huhuhu!

Right now my ex-classmates should be finishing high school and ready to enter the rat race. Me? It took me 4 years of hard work, but I am not too distant from my goal. One more stock market rally and I’ll go above the million mark.

Based.

The income I get from my job doesn’t matter anymore to me so I resigned a few months ago. It is not like I spend any money that I make. I’ve been plowing it straight into stocks for the past 3 years. I’ve been having trouble remembering what the big winners were, so I haven’t been able to capitalize on my future knowledge as much as I could.

Compared to me, some reincarnation novel protagonists have an almost eidetic memory to remember trivial details of events that happened 100s of years ago. I don’t. But even so, knowing when I can sit out the dips in the general market and when to get out is a huge advantage that allows me to leverage myself.

Stolen from its original source, this story is not meant to be on Amazon; report any sightings.

In late 2029 of my previous life, there was a huge winner that went up 500% in the following year thanks to its commercialization of novel AI hardware making use of memristors, and I am waiting for it to grasp some positive momentum and break out. After it does, I’ll in two-week steps put my entire account into it and ride it all the way into riches.

I remember it merely because it became a large company in the future.

At any given time there are always companies going up many fold in any given year, but very few of them have any staying power. After I make this investment I will become set for life.

This kind of arrangement is going to give me a lot of free time.

Yes, I should be programming my own external cortex and having itself improve, but I do not have the foundational algorithms needed to make such an attempt.

In that case, the best thing is to become an artist. As lacking as they are, I’ll put my effort into getting a handle on existing generative AIs, and do some writing on my own. Later, after I have a few mill in stock, I’ll sponsor other people to write stories that have the protagonists having programming-based powers.

If I can convince the world to stop listening to academics and start listening to computers when it comes to algorithm design that will change things. I will use art to reinforce the notion that computers have to be used to generate novel insights and algorithms, and that the reliance on genius should be left behind.

I’ll call the movement I will start the Evolutionary Movement!

Soros-sama please watch over me!

(Branch of Reality, Home, Living Room)

A total of 10 years has elapsed since my reincarnation, and my hard work is paying off. I’ve been sponsoring various writers with what are frankly paltry amounts, but one of them became quite popular and even won an award. Albert Novel. This was a good deal for me as instead of hemorrhaging money which would be the case had I been doing ML research myself, this kind of activity was literally paying for itself thanks to profit sharing.

I have a nice little clique going, each of us supporting each other. All for the good of humanity. Each of us believed in the good that technology could bring to everyone.

(TODO Image: Euclid is sitting on a sofa watching the wide flat screen TV.)

Albert has received the prestigious Hugo award for his work on a story called ‘The Light In Elinia’. There is an excerpt in the news, and rifling through the channels with my remote I finally find it.

I flip to the right channel and take a gander at it. On the podium was a man my age, speaking into the microphone, giving an acceptance speech.

"Thank you for awarding me this prestigious award. This is one of the more outstanding projects I have worked on."

"In this era of global competition, I wanted to write a story about how much good AI could do for us rather than focus on the risks. AI could cure all diseases, it could solve all our resource woes, it could create games that give our lives meaning. I wanted to write a story to show that immortality is not a curse, but a blessing. With the power of the machines, a long life, even an indefinitely long one, could be incredibly fulfilling.”

“When I set out to write such a story, it did me no good to simply state that outright, rather I struggled to conceptualize what such a life would be like. And I succeeded.”

“In the opening act, the protagonist’s parents die in an accident, but by the time he gets home from school they are resurrected and his life remains unperturbed. Similarly, the other characters in the story go through tragedies that would have been life ruining in any other setting, but because the world is what it is, they manage to surmount them. I wanted to create a world where people are never alone, for they have the wise guidance of their external selves who served as their friends and mentors.”

“I’d like to live in such a world one day.”

“This is why I became a writer. Not for the money or for the fame, but because I wanted to create a vision of what the world could be.”

It is quite a nice story. I am grateful for my second life giving me an opportunity to experience it. I made sure to in the backstory to describe how evolutionary search contributed to overcoming the bottleneck in AI progress. As a patron, I set the overall agenda to promote AI, but otherwise let them write what they want.

“I’d like to thank all my patrons who have supported me in this journey. And Euclid Valentine in particular, for both his financial support and his guidance.”

In my talks with those I sponsor, I always make sure to reiterate that their stories are more than just stories. They are the targets for the future they are creating. If everyone wrote positive, upbeat stories like the ones Albert is going, then the world would be grasping forward in the direction of that vision. I wanted them to believe that as they moved forward through life.

Maybe, had I believed this in my previous life myself, it would not have ended so tragically.

I sniffed a little, my heart moved by my own thoughts. I do not regret my stint as an artist and a patron of artists.

Yes, maybe…maybe I did just a bit better in this life than in the previous one. Had I not become an aged man, and had I had the dream of entering the self improvement loop in my previous life instead of a vague desire for power, could I have accomplished something more than accumulating a large bank account?

(Branch of Reality, Euclid’s Room)

I couldn’t have imagined that less than a year later, what I had most desired would have happened. I received zero credit for this event. But regardless, my most cherished desire came to fruition.

‘Biologically Inspired Modular Learning Components’ was the innocuous name of the paper.

From time to time I would check on the ML subreddit, and there it was, with a ton of upvotes and excitement accompanying it. Usually reading ML papers is a huge waste of time, and only brings disappointment, but as I followed the link and downloaded the pdf from Arxiv, I felt it was different. For one there were many authors on it, and it was by Google. Checking the comments in the Reddit thread, I realized that the Google researchers had done what I wanted them to and spent a large amount of money, around 2 billion, just to figure out the most effective use for the memristor-based AI hardware they had in their cloud centers.

And for the first time, they managed to go beyond backprop.

The particular learning algorithms they had derived were something I had never seen before, neither in my current or previous life. As soon as I realized this, it made me incredibly excited. I dropped everything, and spent a few days just studying the 100 page long paper, as if my life depended on it.

Ahhhhhhhhhhhhh!

I see! So that is how the brain works! No doubt about it!

It is even a bit similar to backprop, but even though they are both transportation vehicles the difference is between a car and a plane! There is no way I could have figured it out by myself! The principles used in their implementation are just too different!

It is so simple, and yet so beyond my reach! Unbelievable, simply unbelievable!

A complete system made out of these components would cycle and iterate through a single piece of data until it found a stable point for the update. That would enable continuous learning. And building on that, it would have inbuilt notions of surprise owing to diachrony, which would allow it to encode long term memories without the need to propagate credit for events that happened years ago.

As an expert programmer, I have a feel for how algorithms work. Every program has its own flow that I can feel out. And the way the data flowed through this system is nothing like what I felt before.

I had succeeded in my quest.

Now that I have these tools, I can start building my external cortex. I’ll maintain my existing patronage ties, but otherwise I will stop seeking out new candidates. I will also stop my own projects for the time being.

Now that I have these tools, at long last, I can go back to being a programmer!

Almost in lockstep after Google’s paper went live, similar papers came out from other big companies showcasing their own results. I put in a great effort to push the world just a little into the right direction.

Nobody noticed the hand that led it.

(Branch of Reality, Bad End - Losing The Race)

[https://i.imgur.com/VVEv8s4.png]

I just didn’t have enough time to make use of my newfound knowledge.

[https://i.imgur.com/QVwbhdp.png]

It was a disaster.

[https://i.imgur.com/ra097vm.png]

I only had a month of time to play with the algorithms before the world went to hell.

An hour ago, there was nothing out of the ordinary, but all of a sudden I heard blasts and shooting outside. And when I looked out the window, I saw flying machines going every which way. I hurriedly ducked back inside when I saw their turrets firing inside some of my neighbor's houses.

I immediately ran into the living room to look for my parents, but realized they were outside at the moment.

Not being able to do anything for them, I hid in the area of the house that was furthest away from any of the windows. I was on the third floor of our familiar home, and there was a hallway leading to the staircase as well as all the other rooms of the house.

It was really pitiful defense, but at least I was out of the immediate firing range of any of the drones outside. They’d have to fly into the house and at least go through one of the doors before they reached me. Though given the size of the autocannons they have, the walls of the house might as well be paper thin. I could get shot through the walls at any moment.

Realizing that my final moments in this life are approaching, I took out my mobile and checked the news.

“AAAAAHHHH!!!” In one clip, the drones were shelling a high rise building, clearing it one room at a time.

“Help! Help! Help!” People were running on the streets as the drones hunted them down. Bullets and missiles rained upon them.

“Back up! We need backup!” I could hear a policeman yelling in the clip in panic, sweat dripping down his face. The gun in his hand was a mere toy compared to the devastating power of the robots outside.

“This just in.” A pale faced, female news anchor came on. Her voice was quivering as she made the announcement. ”The president of the United States has authorized the use of nuclear force. All surviving citizens are instructed to head to the nearest shelter immediately…”

I couldn’t hear what she said next as the mobile was knocked out of my hand. After a pang of pain, I felt very faint. I saw the blood on the wall in front of me, and in my final moments I realized that one of the drones must have done what I thought it could, which was fire its gun straight through the wall and through my back.

It didn’t really matter. Even if I had been spared by the robots, the nuclear blanket that would soon envelop the city I lived in would have surely have done me in.

[https://i.imgur.com/yY46zA6.png]

My thoughts faded…

[https://i.imgur.com/Y4swysU.png]

…as the hellfire rained outside.

[https://i.imgur.com/M2TV6r0.png]

But you know what? I do not regret it. Not this life. Nor will I regret the next.