Saturday, January 2, 2010

I am a meat puppet

First off, most of the thoughts hear are inspired by this great post on the state of machine learning and what is ahead.

I think machines are already smarter than humans and are already coercing the course of human history in certain domains. They just lack consciousness and can't recognize or leverage this ability yet. This may seem like a ridiculous notion, but let us look at how this isn't as absurd as it sounds.

It is a given that computers have revolutionized how we humans do things. The human need to experience and discover information in the world has resulted in computational systems that curate information for humans. Computers have been allowed to provide input into the human existence under the guise that humans have taught them what information they would like to receive.

Examples of computational systems telling humans what to do:
  • Netflix: Deciding what movies humans should watch.
  • Pandora: Deciding what music humans should listen to.
  • Amazon: Inviting humans to exchange money for various objects.
  • Google AdWords (with AdSense): Showing humans things they are most likely to click.
  • Flight Caster: Suggesting to humans what airplane flights they should take.
These systems are self-sustaining (from the perspective of a human, not an age of the universe timeline). As long as the companies who host these systems exist, they will continue to provide input into our human existence. The more input they collect, the more human actions they can suggest. As long as they can suggest reasonable human actions, the company and the system will survive.

Assume these system can do this as humans have taught the system what is reasonable. The case of a competitor taking over is a moot point since it is one computational system replacing another.

The bias opinion of 'humans first' for these data driven systems is false. Once initialized, these systems are self-learning. They suggest actions to humans, and humans will provide input back regarding this action. (example: Rating a movie, that you rented based on a recommendation from Netflix).

In this way, machines tell humans what to do, humans tell the machines what they did, the machines tell the humans what to do again. We think we are in control because we can choose not to do something.

Allow me to demonstrate.
Go find a new book to read.
Do you have a topic or specific book in mind?
  • If you have something in mind, what made you interested in this topic or book. Did you read about it from a site like Digg, HackerNews, Reddit, Slashdot?
If you don't have a book or topic in mind, where do you go to find a new book?
  • The bookstore? (How did the bookstore decide what books to carry?)
  • Call a friend? (Where did your friend hear about this book?)
  • Amazon or some other computer recommendation?

The line between a human driven action and a computer driven action is already blurring. Humans think they are outside of the machines control only because these systems do not exist for all possible human activities.

The only other current limitation of these systems is the ability to receive new information to suggest. Currently this is a 'humans first' action. This too will be blurred as machines are given more ways to collect input from the world. Humans can always create new activities and information, but ultimately what humans learn, influences what they do.

What sources of information do humans use to learn?