[ I'm posting this blog a week and some after I started writing it, because shit happens. ]
Haven't blogged in a while. Didn't realize how long a while until I came back and noticed that Spike was now my age. I'm afraid if I don't start writing more often, I'll return to discover that he's started a family, and that his kids are my age.In that spirit, here's a blog. I'm actually on a plane, right now, bound for Minneapolis, where I'll be boarding a new plane for San Francisco. I still work for Google; this is another greater team get-together. Life's been mostly calm and predictable. I can't remember if I blogged since buying a house, but if not, I bought a house. Yay! Homeownership can be a pain, but by and large, I think it's something adults bitch about because they forget how stressful high school is.On that note, I remember all the old people telling me that every successive phase represented “the best years of my lifeâ€?; it could still prove true, but that's because people my age don't seem to know how to have fun. I used to have friends with whom I could play video games or talk about the games I was making; now I'm so good at programming games, the challenge is in all the art and creative stuff that I don't have, anymore. Well, it's not that I don't have it, anymore; it's just that it didn't scale with my growing understanding of the universe and of proper gameplay elements (it is overcome by the rationale that prevents me from making games like Sticker Star; see my other blog post for details).So sure, I miss childhood, but only because I miss the lack of inhibition that comes with it… and the time I had to spend with friends…. I'm drowning in free time, actually, but no one else seems to be, and I'm low on physical friends, and the "IRL" friends I do have don't really share my entertainment interests.Anyway, that’s enough for that tangent. The point is, life's a bit boring for this and that reason. As usual, however, I've kept busy regardless.Since programming is pretty much a solved problem, in my life, I've taken to doing more physical crafts. I've been playing with the laser cutter at the local TechShop. I've made a number of small things—utilities for my house, decorative nonsense, gifts for my father, and playground equipment for my hamster.Oh, that's another thing I should mention. My parents got me a little housewarming present when I moved out of my apartment. His name is Zen. Since I'm not fourteen and poor, anymore, Zen lives like some kind of rodent king. He has upwards of a hundred linear feet of Habitrail at his disposal, but because that's still not enough space for a hamster, I went ahead and gave him access to the rest of the room, too. Hamsters are funny: regardless of the space available to them, if they can map it all out in their heads, they want more. So now that Zen has access to my room, he spends most nights digging at my door, trying to get out. I'm not even certain how he identified the door as a structural weakpoint. He's basically a giant rat. Relevant pics enclosed.As for computers, work at Google consumes most of my patience for the woes of software development. I still program on my own from time to time, but not big stuff, like I used to. I have been considering rekindling ENIGMA, and have even written some code in the direction of doing so, but motivation is sparse because the space doesn't fascinate me like it once did (again, programming isn't really a challenge, anymore, and Google eats up most of my budget for putting up with stupid build systems and code health and blah blah blah).What keeps me going is the thought of the next generation of video games. I played Color Splash…. I'll be writing about that one, shortly. What I really want is the next installment of Metroid (and for said installment to not be made by Team Bouncing Boobies). For those who don't know from, eg, IRC, I was big on Mario Maker for a while. I'll be looking for a new installment of that, on Switch; I had a lot of fun with it on Wii U. I made a few of my own levels (like this and this), which look hard, but pale in comparison to even some of the easier levels by harder-core players (try Carl's Premature Detonation or U-Break, the original hard-as-balls level). I've been hanging around in the Mario Maker community on Twitch, and have made some friends there, but none that I actually hang out with off Twitch or play any games with. I think I'll be playing Breath of the Wild in another month or two, here—whenever I can get my hands on a Switch. That should be fun. In broader life news, I'm excited for UberEATS’ arrival in Pittsburgh—I hate driving, but I love eating out. I've been bitching at Taco Bell on Twitter about only backing DoorDash for delivery. They've ignored me outright. If you know anything about my personal life, you know I really can't stand driving (or travel in general). So Taco Bell delivery is something I'm pretty interested in, as is Google Express. Even more broadly, and looking forward a few years, Autonomous cars are a huge interest of mine, and I'm excited for the Pittsburgh-Chicago hyperloop—whatever becomes of it.I'm not sure if any of you pay attention to the autonomous car space, but it's really heating up. At least Waymo is; everyone else is basically just peacocking, and some of them are paying the price for it, now (the rest will pay the price for it, later). This is as far as I wrote on the plane—I only had internet over SSL, because I refuse to pay Gogo $8 for the shittiest connection on earth. This meant I could check email and send IMs, etc, but not view or post to 64D. Since then, I had a fucking hell of a time at the airport, but otherwise all was well. I enjoyed the offsite, even though I had virtually zero time to myself—I reserved like eight hours for showering and sleeping (except on the last day where I reserved more like five). While I was there, AlphaGo trounced Ke Jie (19-year-old world champion) at Go… I have mixed feelings about this. I felt completely fine about it until I realized the kid is 19; now I feel bad because he's spearheading a new age of intellectual obsolescence that I won't enter myself for another decade and some….I could probably rant for hours about machine learning and AlphaGo and self-driving cars, but there's not much sense in it. Just know that big things are always afoot in these fields, and it's usually not the loudest voice that's anywhere near correct.Thanks for reading. Cheers.
We're seeing the same thing in the world of Go, but this is especially striking to me because Ke Jie is just 19 (if he's even 19). It seems somehow different to me—as though he's less established, having been alive for less time than most of the people he beats have been playing the game. Even though the Go community is stronger than ever, now, I feel like he's going to feel pressure to find something new to excel at, since he's so young. But maybe this doesn't change anything.
You shouldn't feel bad for Ke Jie at all. Just because machines are better at doing a task doesn't mean he's lost meaning or enjoyment from it. People who do things at the highest levels find fulfillment in pushing themselves further and further to become better and better. It doesn't matter if there's someone (or something) out there that's much better than they are, the pleasure comes from the task itself as well as the journey.
One of the best examples of this comes from a fun story about a study conducted in the 1970's on memory. A tl;dr of that story is in regards to the subject of that study, SF, was constantly breaking what was believed to be the limit of memory. Eventually he became far and away the greatest at memorizing a random sequence of digits on the planet with his 80+ digits completely crushing the previous best of around 30. What motivated him to keep going? Just… seeing how far he could take himself, really.Cars won't have steering wheels or pedals in five years, so don't worry too much about that.
Go has been a big-ticket problem for a long time, in the AI space. It has vastly more numerous board positions than chess—brute forcing it just isn't feasible. It requires human intuition, which AI previously has been unable to replicate. Now, AI can… to an extent. This is big news. AI "experts" weren't expecting this for another decade. By contrast, I'm expecting an AGI in two.I'm interested to hear that you would want to use an old-fashioned map. When I want to go without technology, I don't replace it with junk technology—I just don't use it. You build up an intuition for how to cook well with a given palette of ingredients. You can ask people for directions if you get lost. No sense putting up with paper.Even in physical crafts, I'm not escaping computers—I had to program the laser cutter, for example. The most elaborate program, I generated with a bit of Processing. The result was this.And I completely forgot to link pics of Zen. Added.I'm not sure I have found a happy medium like you're describing, where an answer is known but I still feel challenged in obtaining it. Those are hard to come by. I know a couple cool riddles in computer science and information theory that I enjoyed, and Project Euler is full of some neat math and coding questions, but I think the hard part of finding a good challenge is really in finding one that isn't too hard for you, individually. I've personally looked at all the millennium prize problems, and I will attest… they're pretty fucking hard problems. Every time you discover something interesting, you realize it's just another occurrence of the same unanswered question (or, if you're lucky, one of the other millennium prize problems).
Back on the subject of machine learning, what really makes AlphaGo viable isn't the DNN alone—that is an important component, as is the MCTS implementation it uses, but the real magic is in how the two are used together. The DNN serves as AlphaGo's window into human intuition. It's a 60-layer network designed to decide how likely it is that a human would make a particular move, given the shape of the board. Using this in conjunction with existing technologies designed to conquer Go (namely, MCTS) is what suddenly makes the search space tractable. Now, don't get me wrong—the quality of DeepMind's DNNs works to its advantage enormously. Prior to AlphaGo, there was no neural network that could decide who was ahead in a game of Go with anything near reasonable accuracy. That's especially bad because a value function is basically the only ingredient in any successful minimax implementation.But if you're interested in AI, maybe that would be something to look into. Go reinvent the wheel, a little, and learn how experts in the field do it, the hard way. That sounds like a decent way to push yourself, and it's how I did most of my best learning, personally.