Feb 14, 2012
This is a very personal account of thoughts and events that have led me to a very interesting point in my life. Please read it as such. I present a lot of points, arguments, conclusions, etc..., but that's not what this is about.
I've started reading LW around spring of 2010. I was at the rationality minicamp last summer (2011). The night of February 10, 2012 all the rationality learning and practice finally caught up with me. Like a water that has been building up behind a damn, it finally broke through and flooded my poor brain.
"What if the Bayesian Conspiracy is real?" (By Bayesian Conspiracy I just mean a secret group that operates within and around LW and SIAI.) That is the question that set it all in motion. "Perhaps they left clues for those that are smart enough to see it. And to see those clues, you would actually have to understand and apply everything that they are trying to teach." The chain of thoughts that followed (conspiracies within conspiracies, shadow governments and Illuminati) it too ridiculous to want to repeat, but it all ended up with one simple question: How do I find out for sure? And that's when I realized that almost all the information I have has been accepted without as much as an ounce of verification. So little of my knowledge has been tested in the real world. In that moment I achieved a sort of enlightenment: I realized I don't know anything. I felt a dire urge to regress to the very basic questions: "What is real? What is true?" And then I laughed, because that's exactly where The Sequences start.
Through the turmoil of jumbled and confused thoughts came a shock of my most valuable belief propagating through my mind, breaking down final barriers, reaching its logical conclusion. FAI is the most important thing we should be doing right now! I already knew that. In fact, I knew that for a long time now, but I didn't... what? Feel it? Accept it? Visualize it? Understand the consequences? I think I didn't let that belief propagate to its natural conclusion: I should be doing something to help this cause.
I can't say: "It's the most important thing, but..." Yet, I've said it so many times inside my head. It's like hearing other people say: "Yes, X is the rational thing to do, but..." What follows is a defense that allows them to keep the path to their goal that they are comfortable with, that they are already invested in.
Interestingly enough, I've already thought about this. Right after rationality minicamp, I've asked myself the question: Should I switch to working on FAI, or should I continue to make games? I've thought about it heavily for some time, but I felt like I lacked the necessary math skills to be of much use on FAI front. Making games was the convenient answer. It's something I've been doing for a long time, it's something I am good. I decided to make games that explain various ideas that LW presents in text. This way I could help raise the sanity waterline. Seemed like a very nice, neat solution that allowed me to do what I wanted and feel a bit helpful to the FAI cause.
Looking back, I was dishonest with myself. In my mind, I already wrote the answer I wanted. I convinced myself that I didn't, but part of me certainly sabotaged the whole process. But that's okay, because I was still somewhat helpful, even though may be not in the most optimal way. Right? Right?? The correct answer is "no". So, now I have to ask myself again: What is the best path for me? And to answer that, I have to understand what my goal is.
Rationality doesn't just help you to get what you want better/faster. Increased rationality starts to change what you want. May be you wanted the air to be clean, so you bought a hybrid. Sweet. But then you realized that what you actually want is for people to be healthy. So you became a nurse. That's nice. Then you realized that if you did research, you could be making an order of magnitude more people healthier. So you went into research. Cool. Then you realized that you could pay for multiple researchers if you had enough money. So you went out, become a billionaire, and created your own research institute. Great. There was always you, and there was your goal, but everything in between was (and should be) up for grabs.
And if you follow that kind of chain long enough, at some point you realize that FAI is actually the thing right before your goal. Why wouldn't it be? It solves everything in the best possible way!
People joke that LW is a cult. Everyone kind of laughs it off. It's funny because cultists are weird and crazy, but they are so sure they are right. LWers are kind of like that. Unlike other cults, though, we are really, truly right. Right? But, honestly, I like the term, and I think it has a ring of truth to it. Cultists have a goal that's beyond them. We do too. My life isn't about my preferences (I can change those), it's about my goals. I can change those too, of course, but if I'm rational (and nice) about it, I feel that it's hard not to end up wanting to help other people.
Okay, so I need a goal. Let's start from the beginning:
What is truth?
Reality is truth. It's what happens. It's the rules that dictate what happens. It's the invisible territory. It's the thing that makes you feel surprised.
(Okay, great, I won't have to go back to reading Greek philosophy.)
How do we discover truth?
So far, the best method has been the scientific principle. It's has also proved itself over and over again by providing actual tangible results.
(Fantastic, I won't have to reinvent the thousands of years of progress.)
Soon enough humans will commit a fatal mistake.
This isn't a question, it's an observation. The technology is advancing on all fronts to the point where it can be used on a planetary (and wider) scale. Humans make mistakes. Making mistake with something that affects the whole world could result in an injury or death... for the planet (and potentially beyond).
To be honest, I don't have a strong visceral negative feeling associated with all humans becoming extinct. It doesn't feel that bad, but then again I know better than to trust my feelings on such a scale. However, if I had to simply push a button to make one person's life significantly better, I would do it. And I would keep pushing that button for each new person. For something like 222 years, by my rough calculations. Okay, then. Humanity injuring or killing itself would be bad, and I can probably spent a century or so to try to prevent that, while also doing something that's a lot more fun that mashing a button.
We need a smart safety net.
Not only smart enough to know that triggering an atomic bomb inside a city is bad, or that you get the grandma out of a burning building by teleporting her in one piece to a safe spot, but also smart enough to know that if I keep snoozing every day for an hour or two, I'd rather someone stepped in and stopped me, no matter how much I want to sleep JUST FIVE MORE MINUTES. It's something I might actively fight, but it's something that I'll be grateful for later.
There it is: the ultimate safety net. Let's get to it?
Having FAI will be very very good, that's clear enough. Getting FAI wrong will be very very bad. But there are different levels of bad, and, frankly, a universe tiled with paper-clips is actually not that high on the list. Having an AI that treats humans as special objects is very dangerous. An AI that doesn't care about humans will not do anything to humans specifically. It might borrow a molecule, or an arm or two from our bodies, but that's okay. An AI that treats humans as special, yet is not Friendly could be very bad. Imagine 3^^^3 different people being created and forced to live really horrible lives. It's hell on a whole another level. So, if FAI goes wrong, pure destruction of all humans is a pretty good scenario.
Should we even be working on FAI? What are the chances we'll get it right? (I remember Anna Salamon's comparison: "getting FAI right" is like "trying to make the first atomic bomb explode in a shape of an elephant" would have been a century ago.) What are the chances we'll get it horribly wrong and end up in hell? By working on FAI, how are we changing the probability distribution for various outcomes? Perhaps a better alternative is to seek a decisive advantage like brain uploading, where a few key people can take a century or so to think the problem through?
I keep thinking about FAI going horribly wrong, and I want to scream at the people who are involved with it: "Do you even know what you are doing?!" Everything is at stake! And suddenly I care. Really care. There is curiosity, yes, but it's so much more than that. At LW minicamp we compared curiosity to a cat chasing a mouse. It's a kind of fun, playful feeling. I think we got it wrong. The real curiosity feels like hunger. The cat isn't chasing the mouse to play with it; it's chasing it to eat it because it needs to survive. Me? I need to know the right answer.
I finally understand why SIAI isn't focusing very hard on the actual AI part right now, but is instead pouring most of their efforts into recruiting talent. The next 50-100 years is going to be a marathon for our lives. Many participants might not make it to the finish line. It's important that we establish a community that can continue to carry the research forward until we succeed.
I finally understand why when I was talking about making games that help people be more rational with Carl Shulman, his value metric was to see how many academics it could impact/recruit. That didn't make sense to me. I just wanted to raise the sanity waterline for people in general. I think when LWers say "raise the sanity waterline," there are two ideas being presented. One is to make everyone a little bit more sane. That's nice, but overall probably not very beneficial to FAI cause. Another is to make certain key people a bit more sane, hopefully sane enough to realize that FAI is a big deal, and sane enough to do some meaningful progress on it.
I finally realized that when people were talking about donating to SIAI during the rationality minicamp, most of us (certainly myself) were thinking of may be tens of thousands of dollars a year. I now understand that's silly. If our goal is truly to make the most money for SIAI, then the goal should be measured in billions.
I've realized a lot of things lately. A lot of things have been shaken up. It has been a very stressful couple of days. I'll have to re-answer the question I asked myself not too long ago: What should I be doing? And this time, instead of hoping for an answer, I'm afraid of the answer. I'm truly and honestly afraid. Thankfully, I can fight pushing a lot better than pulling: fear is easier to fight than passion. I can plunge into the unknown, but it breaks my heart to put aside a very interesting and dear life path.
I've never felt more afraid, more ready to fall into a deep depression, more ready to scream and run away, retreat, abandon logic, go back to the safe comfortable beliefs and goals. I've spent the past 10 years making games and getting better at it. And just recently I've realized how really really good I actually am at it. Armed with my rationality toolkit, I could probably do wonders in that field.
Yet, I've also never felt more ready to make a step of this magnitude. Maximizing utility, all the fallacies, biases, defense mechanisms, etc, etc, etc. One by one they come to mind and help me move forward. Patterns of thoughts and reasoning that I can't even remember the name of. All these tools and skills are right here with me, and using them I feel like I can do anything. I feel that I can dodge bullets. But I also know full well that I am at the starting line of a long and difficult marathon. A marathon that has no path and no guides, but that has to be run nonetheless.
May the human race win.