HuffPost Live webcast: Feds Predict the Future

I appeared on Feb 27 on HuffPost Live webcast, discussing US intelligence and other government agencies that try to predict the future. What are the implications of the big data revolution for intelligence, privacy, and "pre-crime"?

Gregory Piatetsky, March 2, 2013.

Last week I appeared on Huffington Post webcast, which discussed the government agencies attempts to use analytics to predict the future. The webcast was hosted by Alyona Minkovski and included

  • Thomas Malone @twmalone Director of the Center for Collective Intelligence at MIT
  • Ginger McCall @gingermccall Director, Open Government Project Electronic Privacy Information Center
  • Gregory Piatetsky-Shapiro @KDnuggets (Boston, MA) President of KDNuggets
  • Christopher Westphal (Frederick, MD) Co-founder and CEO of Visual Analytics Inc

Here is the recording.


Feds Predict The Future, Feb 27, 2013

My comments regarding the use of analytics for prediction of behavior:

I am not concerned that governments or companies use analytics to predict people behavior. People have tried to predict behavior of other people, animals, and objects, for as long as human society existed.

At present, judges consider how dangerous a person is likely to be to the society in sentencing, or how likely is a person to flee in setting bail - both examples of behavior prediction.

Analytics can probably do as well as judges in predicting behavior, if there are enough examples.

However, I am concerned that analytics models are brittle. The models rarely specify their own limits and users of analytics can rely too much on them, and not realize when analytics no longer apply. That is not a problem when models incorrectly predict behavior of a few customers, but can be a big problem when models apply to large scale systems like the economy or the society or predicting revolutions.

Brittle models can lead to huge failures, like what happened with the financial markets in 2008. Nissim Taleb "The Black Swan" provides plenty more tales of caution.

Some of the links referred to in this program

Here I am explaining my data mining hat to the host.

Gregory Piatetsky in a data mining hat on HuffPost Live webcast