LogFAQs > #952841405

LinkPizza posted...

Im still confused why people think everything will be free, tbh

Because robots don't need to be paid because - as I've said ad nauseum - money is a measurement of human labour.

When you mow the lawn, do you cut a cheque to your lawnmower at the end of it? When you rake up leaves, do you give your rake $20? Of course not - those are tools, they don't need to be paid.

What about a vacuum cleaner? You probably don't pay that either. But, of course, you still have to do the vacuuming, so it takes some labour from you (and if you had to get someone else to do it, you'd need to pay them for their time). But what if we made a robotic vacuum that would do all the cleaning for you? We could call it... I don't know, a "Vroomba" or something catchy like that. This "Vroomba" could do all the vacuuming for you, no labour on your part required. Would you pay it? I don't think you would.

But you'd still have to pay *for* it, right? And you'd still have to pay to replace it or fix it if it gets damaged, right? Well, what would happen if we had robots that could build you new Roombas for free and recycle the old ones when they get too damaged to continue? You'd have free vacuuming for life, with nary a bill to be paid!

Now take that and apply it to the entirety of all jobs everywhere and you'll see why I'm saying that everything in a fully automated world effectively has zero cost.

LinkPizza posted...

Even playing against themselves, they would have to know the rules.

Well, obviously - I thought that much was self-evident.

That said, just because a program knows *how* to play chess doesn't mean it knows how to play it *well*. That's where the self-teaching AI comes into play.

LinkPizza posted...

So, I was right when I said limited amount of moves. Theres still a finite amount of moves you can make.

I'm not sure you appreciate the sheer scale of the example I just gave you.

The number of potential moves in Go are so high they may as well be limitless. By the 30th move in the game, there are 1.5x10^76 different combinations that could be played. For reference, there are an estimated 10^21 stars in the observable universe. If you could record each combination on a single byte of data, you could use every single piece of digital storage in the entire world (computers, servers, phones, hard drives, etc.) and you would not only not have enough data to store the resultant database, you wouldn't even have scratched the surface. The total estimated computer storage space in the world is 300 exabytes, or 300 x 10^18 bytes. If you could compress the entire world's data storage power - all 300 exabytes - down to a single byte and then build a new data network of 300 exabytes of those new bytes... it still wouldn't be enough. You would need to do that four times - a world-network of world-networks of world-networks of world-networks to have enough storage space to store every permutation and combination. For reference, if you could somehow use a single atom to store a game of Go up to that point, you would use up somewhere between 0.01% and 1% of all of the atoms in the entire observable universe writing out all the possible combinations.

That's all for thirty moves. The average game is over 200 moves (which creates a number of potential combinations so high that Microsoft Excel literally just returns an overflow error when I tried to calculate it).

You see why I pointed out that this was not a game that an AI could solve with brute force?

LinkPizza posted...

And depending on how they play, they could very easily predict an opponents moves. So, no. I dont actually see a problem with what I said.

OK, person who knows nothing about Go, how exactly would an AI very easily predict an opponent's moves? I'm very interested in your answer, given that this was a problem the best computer scientists in the world couldn't solve until ~5 years ago.

LinkPizza posted...

But some are. Or different enough that people may not have played them before.

And why would an AI not be able to similarly create these new games?

LinkPizza posted...

But it probably wouldnt make the mistakes in the first place. To get a positive response, you would first need to make the mistake.

Which an AI that has studied historical human games would potentially do.

LinkPizza posted...

The other problem is that I dont know if mistakes that were made on purpose are as funny.

Goat Simulator was a game made on that premise and it did pretty well for itself.

---

Kill 1 man: You are a murderer. Kill 10 men: You are a monster.
Kill 100 men: You are a hero. Kill 10,000 men, you are a conqueror!