Re: Mindless Thought Experiments

From: Matt Mahoney (
Date: Wed Mar 05 2008 - 10:56:35 MST

--- Lee Corbin <> wrote:

> Matt writes
> > The fact is that experience doesn't exist. Of course you will disagree.
> > So will the robot.
> "Experience doesn't exist"? You mean to say that you don't have an
> experience right now of sitting in front of a monitor? That you have
> not the experience of reading an email or watching TV? You mean to
> say that the experience of skiing down a slope doesn't exist? You
> are using the word in a way that I've never heard before.

Of course I believe in my own experience or qualia. But there is a difference
between believing something exists and it actually existing. You can't prove
that anything exists. I can't make any argument that a robot couldn't make.

When I suppress my evolutionarily programmed beliefs and analyze the question
logically I have to conclude that either a robot experiences or I don't. But
in the common sense meaning of the word, humans experience and machines don't.
 But do animals? Embryos? Terminally ill Alzheimer's patients?

> Whether a robot has an experience depends on its complexity. We
> can safely say that the very nice robots the Japanese have recently
> built have no experiences, any more than a car has experiences.
> (Granted, at some level, it may be argued that everything has
> experiences, but those of ants, viruses, and stones are infinitesimally
> rudimentary.)
> Eventually robots could have "inner lives" as complex as ours or more,
> when one day they have enough intelligence/consciousness.

How do you measure complexity? How many bits are required for consciousness?

A program like autobliss ( )
simulates reinforcement learning. Would it be moral to run a version of the
program capable of learning more complex functions? Suppose I add episodic
memory, i.e. it can recall the training sequence? Where do you draw the line?
 And note that I can teach it the exact same function using only reward or
only punishment. Does it make any difference?

-- Matt Mahoney,

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:01:02 MDT