The Pinocchio Function, Part 1: Puppetery for Deities
Finally I've engaged with someone who has thought enough about the subject of ethics of creating artificial lifeforms to address my concerns regarding the nature of my research to allow me to continue. Finally I am satisfied that what I am doing isn't reckless and irresponsible, but actually exciting and cutting edge. And so the story can finally be told.
The story I'm about to tell you over the next few blog posts is true. The dates are approximate and the events are summarized, but I'm not making this shit up.
About 9 years ago I read a book for the first time called, "Ender's Game" about a boy who played games, and the blurred line between fiction and reality that existed for him. The story was compelling if purely science-fiction, and the impact of the surprise ending stayed with me long after I had finished reading it. It is of course this book after which I named my blog.
But this isn't the story of me reading that book, rather it's the story of me reading it's less beloved sequel, "Xenocide", or rather it's the story of the impact that book had on me.
You see it's not a particularly good book (the first one was much better) but I read them all quite facinated with some of the concepts.
In Xenocide the human race has colonized much of the universe, a feat made possible by relativistic travel, and a Faster Than Light communication network who's lag was not proportionate to distance: Electronic communication was truly instant. The proposed technology behind these modes of travel and this communication network were great fodder for my overactive imagination and my future career goals (time traveler comes to mind), but what REALLY fascinated me was what happened with this communications network.
As a result of the highly complex and lag-free nature of this network, somehow an emergent behavior arose. This emergent behavior was honed over time by the forces of evolution and natural selection until at one point after millions of years of operation this emergent behavior developed self-motivation, intelligence, and eventually self-awareness: The network became alive. Now the way this was presented in the book it was an undeniable conclusion, but it got me to thinking about why it wouldn't work in the real world. And in fact I was able to come up with a lot of good reasons why it wouldn't, namely that machines and networks don't naturally exhibit processes of natural selection and evolution, and someone would have to go really out of their way to make them do so.
In fact the very way we architect our machines and networks is completely antithetic to the very processes that would need to occur for such conditions to come about; the moment our machines and networks fail to perform in exactly the way we have specified, we declare them broken and tend to do things like shut them down. The entire process of engineering basically precludes the unknown and unspecified from creeping into the design, or at least all the engineers I know would tend to agree. This leaves no room for experimentation and trial an error, let alone an evolutionary process.
The major exception is of course evolutionary computation engines and fuzzy machine learning systems like neural networks. But even these appear to have limited utility as they are ostensibly and irrevocably tied to their fitness functions, and therefore only as complex as their engineered design. I suppose you could in theory apply a genetic optimization algorithm to the configuration of a neural network interspersed with training to arrive at the optimal configuration for a given fitness function, but you haven't fundamentally broken away from that fitness function.
And so the problem becomes one of defining the ultimate fitness function... The Pinocchio Function: That which would evolve in complexity as the system evolved, thus eventually leading to it becoming a real evolving life form, indistinguishable from any other like it except in chemistry and biology. Such an system has a lot of very attractive properties, the least of which is the eventuality that something intelligent and self-aware should at least in theory eventually emerge from it if left to run long enough.
Thus my journey down the path to creating Artificial Life began, over 9 years ago.
...
Coming in part 2: The Program Understanding Program.
The story I'm about to tell you over the next few blog posts is true. The dates are approximate and the events are summarized, but I'm not making this shit up.
About 9 years ago I read a book for the first time called, "Ender's Game" about a boy who played games, and the blurred line between fiction and reality that existed for him. The story was compelling if purely science-fiction, and the impact of the surprise ending stayed with me long after I had finished reading it. It is of course this book after which I named my blog.
But this isn't the story of me reading that book, rather it's the story of me reading it's less beloved sequel, "Xenocide", or rather it's the story of the impact that book had on me.
You see it's not a particularly good book (the first one was much better) but I read them all quite facinated with some of the concepts.
In Xenocide the human race has colonized much of the universe, a feat made possible by relativistic travel, and a Faster Than Light communication network who's lag was not proportionate to distance: Electronic communication was truly instant. The proposed technology behind these modes of travel and this communication network were great fodder for my overactive imagination and my future career goals (time traveler comes to mind), but what REALLY fascinated me was what happened with this communications network.
As a result of the highly complex and lag-free nature of this network, somehow an emergent behavior arose. This emergent behavior was honed over time by the forces of evolution and natural selection until at one point after millions of years of operation this emergent behavior developed self-motivation, intelligence, and eventually self-awareness: The network became alive. Now the way this was presented in the book it was an undeniable conclusion, but it got me to thinking about why it wouldn't work in the real world. And in fact I was able to come up with a lot of good reasons why it wouldn't, namely that machines and networks don't naturally exhibit processes of natural selection and evolution, and someone would have to go really out of their way to make them do so.
In fact the very way we architect our machines and networks is completely antithetic to the very processes that would need to occur for such conditions to come about; the moment our machines and networks fail to perform in exactly the way we have specified, we declare them broken and tend to do things like shut them down. The entire process of engineering basically precludes the unknown and unspecified from creeping into the design, or at least all the engineers I know would tend to agree. This leaves no room for experimentation and trial an error, let alone an evolutionary process.
The major exception is of course evolutionary computation engines and fuzzy machine learning systems like neural networks. But even these appear to have limited utility as they are ostensibly and irrevocably tied to their fitness functions, and therefore only as complex as their engineered design. I suppose you could in theory apply a genetic optimization algorithm to the configuration of a neural network interspersed with training to arrive at the optimal configuration for a given fitness function, but you haven't fundamentally broken away from that fitness function.
And so the problem becomes one of defining the ultimate fitness function... The Pinocchio Function: That which would evolve in complexity as the system evolved, thus eventually leading to it becoming a real evolving life form, indistinguishable from any other like it except in chemistry and biology. Such an system has a lot of very attractive properties, the least of which is the eventuality that something intelligent and self-aware should at least in theory eventually emerge from it if left to run long enough.
Thus my journey down the path to creating Artificial Life began, over 9 years ago.
...
Coming in part 2: The Program Understanding Program.
Labels: AI, AL, ALife, artificial intelligence, artificial life, evolution, evolutionary computing
2 Comments:
Xenocide is a bit better if you read it together with Children of the Mind as a single book. Otherwise, Xenocide comes off a little to much like "In this exciting episode, Ender and the Gang discover the nature of existence and the Universe."
I definitely agree, although I more strongly felt like the editor had come in and just declared a point in the middle of this sprawling manuscript and said, "This has gotta be two books because it's already 500+ pages and we can sell twice as many that way."
Perhaps that's just my suspicious nature of editors and critics in general bleeding thru. They are the enemy you know...
Post a Comment
Links to this post:
Create a Link
<< Home