Introduction / disclaimer: Yesterday I was peeking at quora as usual and ended up finding this interesting company through this another also interesting philanthropic company all somewhat related to facebook. On deciding to apply for a job there, I was given this question, which I thought would be cool posting here! I'm far from being an expert on this subject, so I will probably sound quite dumb here if you, my dear reader, have any experience... I'll take my chances. :D
I know all the below might sound too unpractical, but I think it's where it should begin. Some well defined concepts. And this is where I stand today trying to achieve that.
There are several distinct points I would consider:
- Singularity and Vision
You (vicarious) seem to have done (so much I have to read there!) some amazing work focusing on "vision", which is probably the best approach to begin developing such a technological singularity. Yes, this "intelligent machine" would have to mark some kind of singularity, in which it would be so much smarter than us it would either destroy us all or help us all. That is to me the goal here. And yes, I think enabling it to have a "vision" is a great start.
- Intelligence and Examples
I like to define "intelligence" as "capacity to predict the future" (someone once told me that not too long ago, but I can't recall who). As much as we like to perceive the brain as the most intelligent machine ever built, it has 1 main function above all: generating movement.
I'm not sure those two things *must* be connected, and I don't think anyone can tell that for sure. So I'd have to take a guess and if, as I suspect, an intelligent machine indeed doesn't have to be able to generate movement, I think that's reason enough to ignore trying to replicate the brain at all to begin with. So forget neurology or biology! (which I believe are some of the current trends in this subject)
- Internet and Algorithm
Also, I think the best approach would be using the internet. Think SETI@home. The "magical singularity algorithm" would have to be able to identify itself anywhere and communicate with one another in unison.
At this point, for no good reason, I believe it is possible to develop such an algorithm. The "intelligent machine" could get "alive" growing from one single complex and rather small algorithm. I hope I'm not completely wrong here... This is where I should probably read more first, as I bet there are attempts already done both in thinking and applying here.
- Randomness and Choice
It won't ever know everything there is to know. And it will need to make decisions. It is my philosophy that choice and free will are nothing but an illusion in our deterministic universe, and something is only as random as we can't predict its outcome. So true randomness and chaos are both just some kind of pattern we can't recognize. (I'm actually writing more about this in another yet unreleased blog post)
The algorithm will need to include the best randomization method and that means to gather as much data as possible and summarize it to help in the decision making process. This means, when it takes too long to find an intelligent answer, it will get a random answer instead and learn from it later on.
- Asimov Laws and Maleficence
I wouldn't bother of consequences. Trying to add restrictions to it would most likely be useless, since it would outsmart us anyway. Too much effort for nothing. At least the kind of "intelligent machine" I'd like to see "alive" would have to be at minimum that smart.
There is, of course, a chance it will not benign. In case that happens, well, at least the world will be a much more interesting place than it is today and we would have a common universal enemy to unite humanity! (watchmen anyone?) In any case, trying to develop something smarter than us is building something we will eventually have no control upon.
I believe there's a good chance a real smart being which will eventually arise would ultimately be benevolent. Take my definition of "intelligence" there, a collaborative future is always better, anyone knows that. Right?
And we wouldn't be building robots here.
That's about it. Now, it's time to choose a platform and let the trial and error begin! Too much concept and too little work will not yield to any humanity killing machine.