back in uni I was playing around with artificial neural networks and genetic algorithms.
I’d found the AI classes interesting and wanted to play around with some of the methods.
I created some agents, a little 2d environment, a fitness function and a framework for breeding.
I left it running overnight and came back to little agents merrily trundling round my 2d environment and performing extremely well.
I had no idea how they were achieving it, I just knew that they were.
I could see the state of every node in their ANN but that didn’t mean I could easily decode the states and weightings to figure out their “thought processes”.
And that was just a little ANN with about 100 nodes.
Even if you can create something and see all it’s constituent parts doesn’t mean you understand everything about it.
This is precisely the type of simulation I’m taking about. I was also playing with genetic algorithms when I started thinking about this. So let me ask you this, if you got to a point where it was evident that your agents were conscience, intelligent, and examining their own environment, what sort of methods can you devise to communicate with them?
well.. it’s not likely to be an issue but I imagine the kinds of protocols proposed for talking with aliens would be applicable with the advantage that you can make sure you use some method they notice.
I’ve sat watching things like that improve over generations and you still only see the results.
Perhaps if I’d examined the ANN’s for every generation I’d have more insight but the important point is that merely being able to create something doesn’t mean you can automatically understand everything about it.
Aside from the fact that having godly power doesn’t necessarily correlate well with an ease of understanding life-forms...
The universe less like a carefully painted mural in which humans and other life forms were mapped out in exquisite and intricate detail, and more like a big empty petri dish set in a warm spot billions of years ago. The underlying specifications for creating the universe seem to be pretty simple mathematically (a few laws and some low-entropy initial conditions). The complex part is just the intricate structures that have arisen over time.
I find it implausible that a being with enough power to create a universe has a hard time understanding the life-forms that arise in it.
back in uni I was playing around with artificial neural networks and genetic algorithms.
I’d found the AI classes interesting and wanted to play around with some of the methods.
I created some agents, a little 2d environment, a fitness function and a framework for breeding.
I left it running overnight and came back to little agents merrily trundling round my 2d environment and performing extremely well.
I had no idea how they were achieving it, I just knew that they were.
I could see the state of every node in their ANN but that didn’t mean I could easily decode the states and weightings to figure out their “thought processes”.
And that was just a little ANN with about 100 nodes.
Even if you can create something and see all it’s constituent parts doesn’t mean you understand everything about it.
This is precisely the type of simulation I’m taking about. I was also playing with genetic algorithms when I started thinking about this. So let me ask you this, if you got to a point where it was evident that your agents were conscience, intelligent, and examining their own environment, what sort of methods can you devise to communicate with them?
well.. it’s not likely to be an issue but I imagine the kinds of protocols proposed for talking with aliens would be applicable with the advantage that you can make sure you use some method they notice.
Do you think you would have understood your creation if you had stayed to watch it grow?
only very slightly more.
I’ve sat watching things like that improve over generations and you still only see the results.
Perhaps if I’d examined the ANN’s for every generation I’d have more insight but the important point is that merely being able to create something doesn’t mean you can automatically understand everything about it.
Aside from the fact that having godly power doesn’t necessarily correlate well with an ease of understanding life-forms...
The universe less like a carefully painted mural in which humans and other life forms were mapped out in exquisite and intricate detail, and more like a big empty petri dish set in a warm spot billions of years ago. The underlying specifications for creating the universe seem to be pretty simple mathematically (a few laws and some low-entropy initial conditions). The complex part is just the intricate structures that have arisen over time.
Exactly. Which reminds me of the computational irreducibility of the universal cellular automaton a la Wolfram.