Computers to predict the unpredictable

Get ready for a rant. U of Arizona prof. Jerzy Rozenblit has been awarded a $2.2 million grant to study volatile 'political and military situations' using software that will predict the actions of paramilitary and ethnic groups, terrorists and criminal groups, etc.   I'm in constant amazement with the human race.  Why does it seem that every time we make any scientific progress in a specific area, our first instinct is to use it to make weapons.

$2.2 Million Grant Calls for Designing Computer Software to Predict the Unpredictable

The Asymmetric Threat Response and Analysis Project, known as ATRAP, is a massively complex set of computer algorithms (mathematical procedures) that sift through millions of pieces of data, considering many factors including social, political, cultural, military and media influences, said Rozenblit, who holds the Raymond J. Oglethorpe Endowed Chair in the Electrical and Computer Engineering Department at the UA.

The software can handle data loads that would overwhelm human analysts, while dispassionately exploring actions and behaviors based solely on the data, sidestepping human cultural biases that might prematurely rule out unorthodox or seemingly bizarre courses of action.

A couple points here, but first a little history lesson.  In 1905 when Albert Einstein published his special theory of relativity, there was very little understanding about the amount of energy that was tied up in the mass of atoms.  Over the next 20 years, as the experiments began to support the theory, it was debated about to use this great discovery.  No, our first real use was not medicine, or energy, or space exploration.  Our first use of this fantastic discovery was to make a bomb that could kill millions of people in seconds.

We seem to have done this throughout our history.  Our discoveries in biology and chemistry have led to many great accomplishments for mankind, but they have also created some of the most brutal and sinister weapons ever imagined.

Now its happening to our technology.  A couple notes about Dr. Rozenblit's grant.  Anybody in IT knows that $2.2 million is a drop in the bucket, and hardly anything to get very excited about.   And knowing what we know about predictive systems, we know they are extremely limited in their present capabilities, regardless of the "super algorithms" they contain.

More than anything, I was stuck by the pure desperation of Dept. of Defense for more thorough analysis of their existing data.  It doesn't make me feel very secure that these guys are now trying to use game theory models to predict the next terrorist attack.  When common sense should have told us not to let known terrorists onto fueled jets with piloting experience and box cutters.

As to 'predicting the unpredictable', I'm a little disappointed that this concept of 'unpredictable' is even being discussed.  So let me be very clear on this one, if I can write a program to predict things, those things are not 'unpredictable'.  We have not achieved artificial consciousness here.  The system is still limited by the people who programmed it.  So in essence, the system is really nothing more than an extension of the predictions made by its programmer, probably with some randomness and probability thrown in.

Lets talk about IBM's Deep Blue for a moment, because I think the author made a mistake by using DB as a comparison.  Yes, DB made predictions based on the moves of the opponent, and the available moves to the computer.  But DB was also pre-loaded with thousands of previous championship chess matches.  It could interpret the outcome of similar games, and the mistakes made by each player.  In other words, it was seeded with a rich history of chess, i.e., everything was PREDICTABLE based on knowledge of previous outcomes.

A chess board is an 8x8 square.  The boundaries are fixed, the game rules are unbreakable.  I cannot put a Queen on a non-existent board square or move in a pattern not allowed by queens.  So where does unpredictability enter into this?  Answer: it doesn't.  The system is only able to predict what the programmer has seeded it to predict.

So, why does the military waste money like this?  And will this system ever produce a prediction of any value?  It remains to be seen I guess.  Not knocking the work of Dr. Rozenblit here, I'm sure he and his research team have developed some exotic algorithms to find data correspondence based on cultural correlations.  I just wish our government would support this type of technology for a purpose other than finding more people to kill.