The Computational Universe

(One million bonus points for anyone who picks up on the reference in the title.)

The debate between Searle and Brooks really cuts right to the heart of the issue of human identity. Searle defends an academic brand of human exceptionality by ascribing intrinsic semantic meaning to memory; Brooks criticizes this position as an anthropic and anti-scientific red herring in the tradition of the geocentric universe.

There’s not much middle ground between these positions, and the animosity between Brooks and Searle is clear. So, what are we, the laymen, to believe when the experts disagree so vehemently? Our class discussion was an interesting effort to try to parse the morass of “artificial intelligence”— that is to say, artificial consciousness or self-awareness.  Of course, computer systems have for many decades demonstrated astounding linear intelligence, such as huge memories and fast computational speeds. But there has never been any hint of a computer system independently synthesizing these hardware components into an overarching and spontaneous “mind.”

My take on this debate is that computer science, while an extremely useful discipline, really does not provide a way to approximate the function of the human mind in such a way that a computer could be considered conscious. While our understanding of consciousness is poor at best, it is fundamentally a unitary property; an entity is either aware of its own existence or not. Brooks’ argument seems to be that a sufficiently complex computer will “organically” become conscious, yet today’s supercomputers perform nearly as many calculations as the brain per second without demonstrating nonlinear behavior. If complexity itself is not a sufficient condition for AI, then consciousness cannot come about through brute force, and Searle must be correct that our brains have a fundamentally distinct system of computation.

The class discussion, especially Prof. Arora’s perspective, did cause me to reconsider one of my views. Earlier, I had thought that a possible future of computation could be the passing of Turing Tests without the recognition of AI; I have since amended that view to the position that the Turing Test will never be passed. Without a plastic consciousness, no algorithm will be able to meaningfully respond to the full range of possible conversations without betraying its artificiality.

I think there the public views computers as inscrutible and mysterious, and like all things that are poorly understood, there is a tendency to give them powers and potential beyond what is truly there. In my view, like Searle’s, artificial consciousness is a contradiction of terms and is beyond the reach of machines.

Sir Tim Berners-Lee’s public lecture concerning the future of the World Wide Web was sometimes enlightening, but more often than not, lost me in dense thickets of jargon. Berners-Lee, the developer of the web browser and the HTML formatting convention, is now hard at work marshaling support for what he terms the Semantic Web, a next generation directory technology which defines all on-line information and databases terms of a standardized and readable address. This algorithm makes searching far more efficient and all-encompassing, since the coordinate address contains a definition of the resource’s identity which can be directly understood by computers. With this information, computers can make much more intelligent decisions regarding the storage and retrieval of information, such as cross-referencing a work schedule with a time-stamped photograph in order to assign a likely location where the picture was taken. This is just one small example of the greater communication between programs and networks when Semantic Web protocols become more widely used.

Unfortunately, I had trouble following sections of Berners-Lee’s presentation due to the rather technical terminology he sometimes employed to make his case. For example, I have no idea what XML is, except that people should replace it with the Semantic Web’s RDF language posthaste. Had the good knight taken a second to define XML for the laymen in attendance, I’m sure his criticisms of it would have made more contextual sense. As it was, I found myself struggling to stay afloat in a rising sea of inscrutable acronyms at many points. Berners-Lee is clearly full of ideas for making the Internet an even more useful tool to humanity, but I found myself wishing he was speaking in the more “universal language” he advocates for networks.

When I first encountered the Turing-Church hypothesis, it seemed incredibly reductionary: after all, how could any computation be performed by a menu of only six major functions? Not to mention the somewhat unsettling existance of the ominously named “Universal machine” (”Skynet has become self-aware…”), which could perform the function of any other Turing machine, or replicate itself ad infinitum. Simply by defining the meaning of the strings of 1’s and 0’s and allowing a Turing machine to operate, had human beings made themselves as outmoded as a Commodore 64?

Fortunately, the limitations of Turing computation became clear in short order. Due to the Halting Problem, Turing machines are unable to look at a given string of data and give an answer for whether or not any Turing program will ever reach a halt command (i.e., find an answer to the problem). Therefore, not only can Turing machines not find answers for a certain set of problems, but there are questions for which there are no answers to be found, no matter the computational power thrown at the problem. For this reason, Turing machines cannot replace mathematicians in constructing proofs or disproofs of theorems.

This touches on a general truth of computation: the difficulty that computers have in simulating the complexity of thought that is second nature to the human brain. Turing program logic is highly linear, almost like a decision tree of possible outcomes. In contrast, our consciousness is an emergent property, much more fluid, flexible, and harder to pin down- the kind of operation the Halting Problem makes difficult. So, keep your T101s in the garage, it’s not Judgement Day yet.

A strange thing happened in New Jersey on March 5, 2001: it didn’t snow (much). Unremarkable, until you consider that the best meteorologists were predicting over two feet of snow, a storm comparable to the greatest blizzards of all-time.

Despite inland snow totals of over 30 inches, March 5 is remembered as one of the most infamous forecast busts in history. New York finished with a scant 4” and Philadelphia barely managed 1”, triggering a public backlash against the weathermen whose apocalyptic predictions cancelled thousands of flights and cost billions in lost productivity. What went wrong?

In short, meteorologists were caught flatfooted by increased dependency on global forecasting models. Like the tornado simulations we discussed in class, these programs use inputs of windspeed, pressure, temperature, etc., and algorithms based on thermodynamic laws in order to construct a three-dimensional model of weather conditions, only on a larger scale. In fact, these models parameterize the initial conditions of the entire atmosphere in order to produce truly global forecast output. Starting a week before March 5, the most popular models strongly indicated potential for a significant Northeast Corridor blizzard. This in turn led to a wave of breathless hype in major media outlets.

But like any simulated world within a computer, forecasting models are only as good as the data fed into them. Unfortunately, large expanses of the Earth’s surface lack critical weather observations, forcing rough satellite estimations to be used instead. These initial errors compound with respect to time. Additionally, computational limitations limit the resolution (distance between forecast gridpoints) of global models. This truncation of output continually introduces new error into computer forecasts. These two divergences led the simulations to mishandle certain upper-level features, which caused the March 5 Nor’easter to track much farther north than originally thought. Result: busted forecast.

Moral of the story? Weather forecast models require critically-minded interpretation. They are a great tool, but due to exigencies of computation, they are not infallible!

Look ma, I’m blogging! Move over, pets.com sock puppet, there’s a new sheriff in town.

Hello! My name is Ryan Truchelut. While I defy attempts to be  “categorized, filed, and easily referenced” (in the words of Special Agent Fox Mulder), I’ll strive to provide you a few salient facts about me in the following 200-300 words.

The locus to my Princeton experiences thus far is, oddly enough, the weather. While weather is an ambiguous part of the cosmic background radiation in most of your lives, as long as I can remember I’ve been fascinated by the darn thing. I especially have a penchant for the extremes: one highlight was when the eye of Hurricane Charley passed over my house (28.7N, 81.3W) at precisely 9:48 PM, 13 August 2004. As I’ve gotten older, I’ve (partially) exchanged this youthful exuberance for a deeper understanding of the nuts and bolts of meteorological science- I actually have a job doing freelance forecasting for the commodities markets on the side (long story), which is going to be my career post-grad school. In the meantime, I get to write the Daily Prince’s weather column, which gives me a chance to simultaneously indulge another one of my passions, namely being a complete goofball.

Meteorology is actually a big part of why I’m taking COS 116, as a matter of fact. A lot of the basis of modern forecasting comes from the output of fantastically elaborate computer models, one of which is run on a supercomputer over at Forrestal Campus. What is weather if not a chaotic three-dimensional fluid dynamics equation solved with respect to time? Nothing, that’s what. Since I’m a complete neophyte when it comes to the intricasies of the inner sanctums of processing, and shouldn’t be, thus COS 116. Q.E.D.

Other than that, I’m down with a lot of various ephemera/effluvia, much of which it would be pointless to list here. A few high points: I run varsity track (800m, et al.), play the banjo (imcompetently, I might add), and enjoy spinning music on delicious vinyltastic audio fidelity stereo discs (analog media! sacrilege!).

In fact, at this point I’m going to quit haranguing the masses and keep figuring out how to play the Quizno’s spongmonkey song on banjo, because that’s just my way.

The title is a Simpsons quote, in case you were wondering.

Edit (1:18 AM)–   This post was written on a standard OIT issue Dell Latitude D600, running Windows XP Professional. Retransmission without the expressed written consent of Major League Baseball is prohibited.