Re: ITSSIM (was Some new ideas on Friendly AI)

From: Tennessee Leeuwenburg (
Date: Wed Feb 23 2005 - 14:22:44 MST

Hash: SHA1

Phil Goetz wrote:
| --- Tennessee Leeuwenburg <>
| wrote:
|>I read your idea on actions based on proven
|>theories, subject to a
|>blacklist against which unsafe options are blocked.
|>You seemed to be
|>worried that such a system would be susceptible to
|>improve itself to a
|>local minima - that is, it would be hill-climbing to
|>a potentially
|>limiting endpoint.
|>Let Mag(OPTIONS) represent the number of elements
|>contained in the set
|>Mag(OPTIONS) is proportional to IQ(AGI).
|>Here's my argument :
|>X = Max(OPTIONS)
|>If X != now & V(X) > V(now) then IQ(AGI,X) >
|>IQ(AGI,now) which implies
|>Mag(OPTIONS,AGI, X) > Mag(OPTIONS, AGI, now).
| This doesn't show that Mag(OPTIONS) is proportional to
| IQ(AGI).

No, that was an assumption, which you are welcome to query. What it
shows is that Mag(OPTIONS) increases with each iterative improvement, if
it improves at all, and that the maximum is not limited by the an
interative model...

|>Corrolary : Regardless of the speed at which the
|>intelligence of the
|>AGI grows, the options available to the AGI increase
|>with the
|>intelligence of AGI. Incremental increases to IQ do
|>not result in
|>local minima, because the horizon of the AGI is
|>pushed wider with each

What I have shown, I believe, is this :

Under an iterative model, AGI may escape any local minima at which the
peak IQ(AGI) is sufficient to see beyond the local minima. What I
proposed which I think was new, was that local minima can be escaped AT
ALL. This simply is true, and quite defensible.

Ben's point was that he thought that an AGI factoring in Safety would be
much more susceptible to this than ordinary - my response is that in
fact, I think the difference will be in extent, not in kind. Let me know
what about my argument you disagree with, and I will speak to that.

- -T
Version: GnuPG v1.4.0 (MingW32)
Comment: Using GnuPG with Thunderbird -


This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:50 MDT