Infinite Quack

Karl Popper made a big stink about falsifiability . If you’re making a statement about the real world, there had better be some way to disprove (falsify) your statement. Let’s say your statement is “It’s raining”. I should be able to open my window and stick my arm outside and see if it gets wet. Writing as a software developer, I might use the shorter word “testable”. If your assertion is testable, it may be right or it may be wrong. I and others are empowered to figure that out. If the assertion is that life continues after death, how does one render that assertion testable? We must say more. “Ghosts exist”, for example, is testable, as long as your definition of “ghost” includes that they have some observable impacts on the living world. It should be possible to take a picture of a ghost, record a ghost saying something or moving an object. People make ghost hunter TV shows, of course, but the wider scientific community remains unconvinced because this “evidence” doesn’t meet the hi...

Here’s the problem with Superman as a character: He’s boring. Sure, all the Superman movies are packed with exciting drama. He’s facing off against world-ending threats, the odds always seem stacked against him, and though he ultimately triumphs, it always seems to have been a close call. Dramatic plots, no question. Superman movies are exciting, but Superman himself, as a character, generally isn’t. The trouble is that Superman is, traditionally, perfect: a demi-god, who always turns out to have whatever power he needs to defeat whatever enemy he faces. He can fly, he brushes off bullets and explosions, he can survive in the vacuum of space, and he’s always optimistic and kind. The trouble with a perfect character is that it leaves no room for character growth. We may admire people who seem to have it all, but (as Superman literally is) they also strike us as rather alien. They aren’t relatable. We look up to people who can face any challenge. We love people who have flaws. Making us ...

This post was inspired by a talk called Being Human , given by Jason Wodicka, someone I worked with at Amazon. This is a great talk, and you should watch it. Jason is a smart guy, and while he and I may not have arrived at exactly the same conclusions (or perhaps we "violently agree" and are just looking at similar ideas from different directions), I absolutely appreciate where he is coming from. What follows are my personal reactions and responses to the talk. Also I should note that this is hardly my first blog post on AI; if you'd like to read the others, see Never Is A Long Time and Buckle Up and... well, any of these . Yes, Jason: You're quite right about the boom-and-bust, summer-and-winter cycle of AI. It's looked promising before, and we've been disappointed before. I, too, have been around for a while, and I too used to read Wired magazine in print. 😄 We must approach reputed AI developments with a big ol' grain of salt. Turing's "imi...

I’m writing this in 2023. I figure, if you’re too young to be President of the United States (less than 35 years old), you probably can’t imagine what it was like before the Internet. I’m a 1974 model. When I was six, my grandparents left their TRS-80 home computer at our house, to watch it for them while they went on vacation. They also left the manual. I read it. I learned how to use BASIC to write a program. That was it. I was hooked. My first computer was a Commodore VIC-20. My parents bought it for me shortly after my encounter with the TRS-80, because they could see how entranced I was. Obsessed, really. For whatever reason, I got computers. The goal was always to push the machine’s boundaries, to see what I could make it do. Even on these very early machines, whose capabilities were laughable by today’s standards, their possibilities seemed endless to me. There was a universe contained inside a small box. I would invent a project for myself, then another. Most went unfinished, ...

chatGPT can never replace programmers "Never" is a long time. In a narrow sense, yes: maybe you can say that ChatGPT will never replace programmers entirely. But that's a straw man. We don't really care about the narrow statement, do we? Even if you expand it to "Large Language Models will never replace programmers entirely" (and I grok the argument that they can only ever mimic human-produced writing), that's not what we care about, either. What keeps us up at night is this: Can we never build any machine that can think well enough to replace programmers? There's no way to avoid the philosophical, with questions like this, so let's not mince words or dodge the key issue. Is there a magical component to the human brain? You, reader, can believe what you like. I can only describe my own beliefs: I'm a physicalist. I believe that matter and energy exist, the stuff that physics tries to study and understand... and that's it. If you believe t...

Based on lyrics of  Yours Truly, 2095  by Electric Light Orchestra (1981) 2025, 2025, 2025, 2025 I love you, sincerely Yours truly, yours truly I sent a message using just my mind But as the days unwind This I just can't believe I built a world inside a VR game Maybe my life's the same  But this I just can't conceive Can you see me? I ride inside the latest EV car I don't know where we are But it whisks me around the bend I met someone who sounds a lot like you She says the things you do But she is an LLM (2025, 2025, 2025, 2025 I love you, sincerely Yours truly, yours truly) She's built to answer anything I ask And she can multitask Her words always seem so wise She promises me I'm her superstar But when I go too far She spins a world of lies I parrot every phrase that she invents She wins all arguments And she doesn't break a sweat She always knows exactly where I am She aced the bar exam And her brain is the Internet Is this what you want? (Is it what you...

Part 1: The AI Revolution: The Road to Superintelligence Part 2: The AI Revolution: Our Immortality or Extinction Written by Tim Urban, based in large part on the work of Nick Bostrom. Urban does a good job explaining the concepts in an engaging way that should be accessible to anyone. (Bostrom's work is amazing, but his book Superintelligence  adopts a dry style.) Note, these were written back in 2015. So we're five years closer to those estimates for when Artificial General Intelligence (AGI) might arrive. 2045? 2075? Nobody's really sure. More recently, we've seen specialized  AI (called Artificial Narrow intelligence, or ANI) defeat expert Go players, and devices using Natural Language Processing (like Alexa devices) have become common. These are all stepping stones along a path of an unknown length. The arrival time, though, doesn't matter nearly so much as what could happen afterward, an outcome which depends entirely on whether we've prepared. "Our i...