Re: of possible interest: interaction with autonomous agents

From: Eliezer S. Yudkowsky (
Date: Tue Jun 28 2005 - 14:12:31 MDT

justin corwin wrote:
> On 6/28/05, Eliezer S. Yudkowsky <> wrote:
>>That's an odd thing to say. Do you mean that all else being equal, you infer
>>that people in more powerful positions are probably more intelligent? That is
>>backed by the data. But do you mean that people in more powerful positions
>>are by definition more intelligent? That in virtue of having more options,
>>they are smarter? This would mean that I could stand outside an experiment
>>and control someone's intelligence with a dial.
> This is disingenous. No matter what intelligence is based on, if it
> has a physical basis, you could experimentally control it with a dial.

That was disingenous. It's not surprising if a dial that controls people's
serotonin levels can make them more or less intelligent. It is surprising if
a dial that has no direct effect on people's brains can make them immediately
more or less intelligent. If you expose people to different experiences they
might become more intelligent after the passage of time, as you point out.
The problem with a definition of intelligence that depends so sensitively on
environmental availability of options, is that it doesn't seem to capture a
concept of individual intelligence. Rather it seems to be talking about some
property of the whole system which is interesting but not a measure of
individual intelligence per se.

> It's also not true that powerful people simply have more options.
> They're also exercising more options, which means a host of other
> things, like owning good tools, getting better information, continuing
> education, and possibly having underlings to assist them. These things
> all have an effect on demonstrated intelligence, but perhaps not in a
> static IQ test.

These things have delayed effects on intelligence, not immediate effects for
twisting a dial.

Eliezer S. Yudkowsky                
Research Fellow, Singularity Institute for Artificial Intelligence

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:51 MDT