The Daily Pennsylvanian :: Sharon Udasin | It isn't worth trying to predict the future

Here at Hogwarts - I mean Penn - Professor Trelawney's class in Divination has come to life. Suddenly, we are sorcerers, and crystal balls have the power to predict the future and ward off approaching danger. We may not be able to trap Voldemort, but mere Muggle murderers stand no chance against the prowess of Philadelphia.

Now that's wishful thinking.

As the Penn community struggles to protect its students from a sexual assaulter and a screwdriver-bearing robber, safety remains one of our primary concerns on campus. I'm certain our parents would breathe a lot easier if Philadelphia homicide levels significantly decreased. Quite admirably, two of our very own professors have been personally working to accomplish just that.

This coming spring, the city probation department will begin to implement a type of modeling software that estimates the likelihood that a probationer will become a murderer. One of the scholars involved in the program's development is Richard Berk, a Criminology professor at Penn. Approached by Philadelphia officials, both he and Penn professor Lawrence Sherman have been leading the project's development since April 2006.

According to Berk, the ultimate purpose of the software is "to forecast which individuals on probation are most likely to try to kill someone, whether they succeed or not."

Performing a trial run of their model, the group entered two years worth of data into the program, including probation cases from the years 2002 to 2004. Berk defended this seemingly arbitrary time frame, by explaining, "A large fraction of the people on probation who are going to try to kill someone will do so within two years after they start probation." Statistically, he continued, "About one person in a one hundred on probation will try to kill someone within two years."

Interestingly, Berk explained that "there is no checklist for the Philadelphia study," and "the variables come from records that the Probation Department keeps on each probationer."

According to The Philadelphia Inquirer, some of these traits might include gender, age and long-term criminal history.

Yet if we can't establish a reliable list of common variables, then we may face difficulties using the software effectively among a diverse, citywide group of suspects. Also, because the probationers don't necessarily share any of the same traits, there is no standard of comparison among multiple individuals' models.

"We are currently working on forecasting other kinds of criminal activity committed by individuals on probation," Berk said. "And most important, we are also trying to forecast the probationers who are very low risk so that supervisory resources can be shifted from them to the high-risk probationers."

Berk believes that by focusing our efforts on these individuals, we may be able to reduce the amount of crime in both the Penn area and the entire city of Philadelphia.

It's difficult to believe that this idealistic equipment will actually be effective anytime in the near future. Essentially, the project seems to be an attempt to engineer a crystal ball - a mortally impossible feat. Even if the program identified a suspect as a likely murderer, we must remember that a positive result does not necessarily indicate guilt - and most likely, it does not.

"For many of the models we tried, you get about 10 false positives for every true positive," Berk said. "We can reduce this ratio, but doing so would increase false negatives, which present an even bigger problem - murderers on the loose.

Right now, the 10-to-1 ratio is the most viable option, and according to Berk, "the high risk probationers we find who do not try to kill someone are often involved in other serious crimes."

However, we must proceed with caution and remember that this modeling process by no means guarantees accuracy. Perhaps there truly is no clear-cut formula to determine a person's homicidal tendencies, because the motivations behind such horrific actions are often not so black-and-white.

I'm certainly grateful for anything that improves our safety. Unfortunately, however, this new innovation in computer software just isn't going to significantly improve our currently frightening crime levels. Yet catching one murderer out of a slew of probationers is better than catching none, and perhaps for now, we should just be satisfied with our mortal limitations.

And after all, predicting the future didn't work too well in Minority Report.

Sharon Udasin is a College senior from East Brunswick, N.J. Her e-mail address is udasin@dailypennsylvanian.com. Shed a Little Light appears on Mondays.

Comments (9)

To be honest, this entire article reminds me of the arguments for intelligent design: "It is too complex for the human mind to map and understand - it must have come to be because of some outside force we can not possibly grasp"

how did you get sharon is terrible from this article? sorry, i dont think that your comment has any value or makes a point. you basically quote her and just say nothing. [QUOTE id="6b259b6a-2439-41d6-b996-e6ec5800a867"]"However, we must proceed with caution and remember that this modeling process by no means guarantees accuracy. Perhaps there truly is no clear-cut formula to determine a person's homicidal tendencies." Please, just stop writing.[/QUOTE]

I thought 'sharon is terrible' was just showing how absolutely terrible that quote was. In terms of writing. Sharon isn't necessarily terrible, but some parts of this column definitely are.

While I disagree with the previous poster's criticism of Sharon's writing, I must say that I am quite offended by her impugning my trade.

I need to respond to the three of you. I think it's funny that "Sharon is terrible" thinks she should stop writing because he disagrees with her. I think he's a communist. Maybe he should move somewhere that there isn't free speech and he'd be happier. Josh Stanfield is a little better. He actually wrote about her column, but he misses her point. She is not being critical of the intent of the program, just saying being skeptical of future results. And she expresses a real concern that these mistakes could be dangerous. Maybe you communists should read her column instead of just either reading the title or taking a random paragraph and copying and pasting it.

Zach Parise, I'll agree with you that she expresses concern. I read her column, multiple times. From what I understand she uses two main arguments throughout the column: The first is being skeptical of the workings of the model. However, the details she brings up shows that she has no understanding whatsoever of the actual statistical methods used. The second is that she claims that the future is to o vast and mysterious to be forecastable. The first argument is faulty - she simply doesn't understand the concepts. The second is illogical - it is not more of a real argument than the arguments brought forth for intelligent design.

I feel like you're unnecessarily pushing the "crystal ball" analogy. Why exaggerate the goal of the program? The Inquirer article [11.27.2006] makes it clear that the software takes 30-40 variables and produces a score for each individual, which relates to possible future lethality. Often juveniles who are convicted of drug/alcohol possession must complete a survey which, when entered into a software package, produces a score that's used as a predictor of likelihood of addiction/future substance abuse. These types of programs enable, as the Inquirer article suggests, the city to focus limited resources on individuals who are most likely to act a certain way: in this case, commit murder. And a fact you didn't mention: this program is free to the city. Nobody is claiming guaranteed accuracy, but I find it hard to be so critical of a new program with such ambitious goals, provided free to the city on a trial basis. The first paragraph and especially the last line should go: Harry Potter is lame, the whole analogy exaggerates the goal of the project, and the last line doesn't make sense and is out of place. Didn't predicting the future work extremely well in Minority Report, save Tom Cruise's exception? To whoever writes the headlines: the paragraph before the last line directly contradicts the headline with "Yet catching one murderer out of a slew of probationers is better than catching none, and perhaps for now, we should just be satisfied with our mortal limitations."

I assume they have tested the model extensively - especially with different recursive testing methods. That would mean that it produces at least one true positive for every false positive - much better than the 1 to 100 ratio the police worked with previously. This is of course so long as there has been no recent parameter change. The statistical model combined with the intuition and knowledge of experienced police officers has the possibility of greatly reducing homocide in Philadelphia. As for the arguments of the article, simply calling something black magic and claiming that it is impossible to predict the future is not a valid argument, and that seems to be what this article boils down to. The actual problem of the model is that a false positive might unjustly target an innocent person, and he runs the risk of being treated as guilty before proven guilty - although that problem is not so much related to the program as to the general attitudes of the officials. The model is not about indicating guilt, it is about identifying people with a high risk of committing murder. There are other fallacies in the article - especially whith regards to the workings of the model. Sharon, please talk to someone who understands regression analysis properly before you try and attack the details of the model: "Yet if we can't establish a reliable list of common variables, then we may face difficulties using the software effectively among a diverse, citywide group of suspects. Also, because the probationers don't necessarily share any of the same traits, there is no standard of comparison among multiple individuals' models." The idea behind the model is to take data on as many previous individuals on probation as possible and find which traits indicate that they might commit homocide. There are no models for individual people - there is one model which will be used to compare the traits of the individual to see where he would place in the distribution of probabilities of likelyhood to commit homocide.

Comments are closed for this item.