Thursday, December 30, 2010

Don't Be Surprised By Anthropomorphic AI

I've made my point, but let's see how I got there and back again.

Outside the Loews Theatre in Port Chester, NY, just after having the left the movie theatre for the opening screening of Tron: Legacy, the idea arrived within the first few minutes of analysis: why did the programs look like people? Why were they at a club and drinking?

Why's there anthropos in my artificial intelligence? 

My response is simple, "why wouldn't there be!?"

The phrase Artificial Intelligence alone is strange enough since it suggests that we could make a thing that thinks. In the construction of an artificial intelligence, two primary concerns should be the AI's ability to understand us and our ability to understand it. It makes sense to manifest the latter in a human form, since we have some understanding of the language of bodily expressions, and the former can be solved by teaching this AI our language. Binding this digital entelechia to the human form provides the benefit of understandability and the terrifying hazard of offering it everything that's wrong with humanity. Training, raising this intelligence to speak in any human language could spring the threat of it learning all the prejudices already encoded in our semantic web. 

Those biases are not a suggestion of hate in the language, per se, but they are suggestion of why, for example, the AI go clubbing in Tron: Legacy. The precise form of the stimulus regardless, the desire to get fucked-up may be a universal condition of a thinking being. We don't have a heck of a lot of evidence to go on, but--judging by the monkeys who steal drinks from beach side resorts--I tend to believe that there is something to  being conscious that, from time to time, makes one want to be slightly less than so. Though, it is fair to say that the relationship of the chemical alcohol to the brains of primates may help as well. Now, whether this impulse has something to do with a consciousness, particularly one aware of futurity, and the desire to alter it, or whether this has something to do with an intelligent being's ability to process stimuli and its voracious appetite there given, I find it hard to believe that we would develop an AI that didn't have some kind of reward processing. 

So it should make some sense that our attempt to create will likely be in and an imitation, made in our own image and crafted as we craft ourselves. Which is the real trick of the divine texts: the gods are always made in our image. We try to make the Other its own subject by subjecting it to ourselves, and the text of our history that we carry and that could Other ourselves as well. 

The primary concern, though, of Tron: Legacy and, similarly, Ted Chiang's The Lifecycle of Software Objects is not really the AI, Tron or the Software Objects, but about a somewhat more eternal human characteristic: legacy, lifecycle. T:L isn't about Clu and Kevin Flynn, but it is about Sam Flynn and Kevin Flynn, and how Sam had to live in the image of his father as opposed to Clu  who was made totally in Kevin's imagine. I don't have the time to properly riff on Chiang's work--and it's quite fresh so you should just go out and give it a read anyway, but I believe it becomes fairly obvious in the text that the devotion that the two main human characters, Ana and Derek, make to their digients* is no different than what may be displayed in child-rearing. Now, pulling "child-rearing" out is a bit of a loaded term, considering that Ted reflects in the text on the acuity of the digient's growing reasoning skills and in what ways they can and can not be considered "children" since they are totally new forms of intelligence, but the word stands since the emotions described around the relationship between the trainer-programmers and the digients does seem to be like that of a parent, or, perhaps, a pokemon trainer... if we didn't train pokemon to fight but instead, well... just educated them. 

Of course, there is a very serious difference between the two stories: one deals with a world where humans are uploaded into a digital world where AI look like humans, while the other story is about humans interacting with cartoon animals and cartoon robots that live within a virtual world a bit like Second Life. These differences in form are directly bound to the plight of their respective stories. While the programs of T:L are little more than stand-ins for humans, the digients of The Lifecycle at least suggest a space between some kind of pure intelligence that would just explode out of a computer, which would only be nominally artificial, and a mere duplicate of humanity. Clu's dictatorship sees him as being merely a better human, a more demonstratively human, and, therefore, monstrous human figure. The digients, like human children, do listen to and learn from their "parents," but they are also something a little more that is unrelated to their parents and is, instead, internal to themselves. Thus, the characters are bound together by language but they walk together in the shared pursuit of life.

However, T:L's programs are always running, like actual living beings, while the digients in Lifecycle can be "suspended," a full stop on their being opposed to the artificial sleep that the programmers include in them, but  can be accelerated so that they are conscious twenty-four hours a day. The digients suffer a number of metaphysical pains brought on by the fact that they experience the quandary of the cycles of life/death and awake/sleep far more acutely than humans can, never mind the fact that the digients are raised to be cloned, with some of their personality and memory intact, to be sold as a product. The process that is the digient can be halted, called up more or less in task manager or some such and then shut down, but it also can be restarted, and the digient would never know. Ignoring drugs and dramatic illness, human beings have only one skip in conscious and that's when they die and the record stops playing. 

The problem, then, is that Tron: Legacy isn't anthropomorphic, really, since the programs are just people in digital clothing and it doesn't really matter that they are digital beings. For example, the naturally occurring "life-forms" in Tron: Legacy, a new breed of programs that arise from the barren outlands beyond the city of The Grid, are miraculous simply for coming into being out of some wild, apparently unknown, while behaving essentially like any other program, perhaps being a little more curious and [pure/open]-minded, but we only have Olivia Wilde as evidence. However, since the behavior of the digients truly is unknown, given that there is a space that apparently exists between code written and ran, the digients need to be anthropomorphized so that they can be interacted with, in part because an image of them is to be sold and consequently formed into a pet by these consumers while the more hardcore users--mostly the programmers who originally made them--require a tactile set of responses that they can use to interact with, and so we need these characters to wear raiments that we can relate to.

So, going beyond the anthropic principle, perhaps now would be a good time for you to read Stars in my Pocket, Like Grains of Sand by Samuel R. Delaney.

*Which are essentially intelligent tamagotchi--but so much, much more than those words can stand.
P.S.: Yes, I do also appreciate that lifecycle sounds like light cycle.

No comments:

Post a Comment