To me
the distinction is between
A)
"Explicit programming-in of ethical principles" (EPIP)
versus
B)
"Explicit programming-in of methods specially made for the learning of ethics
through experience and teaching"
versus
C)
"Acquisition of ethics through experience and teaching, through generic AI
methods"
I
guess that Philip's "hardwiring" is A. Whereas I am more inclined to think
B and (even more so) C will be useful. However, I can see that A could
potentially be a useful adjunct to B and C in the context of some AI
architectures. A could work a lot better in Novamente than in A2I2, for
example, because Novamente has a mixed explicit/implicit knowledge
representation whereas A2I2's is purely implicit...
However, I admit that "EPIP" isn't as snazzy-sounding
as "hard-wiring" ;-)
-- Ben
G
-----Original Message-----
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]On Behalf Of Philip Sutton
Sent: Monday, March 03, 2003 8:53 AM
To: [EMAIL PROTECTED]
Subject: Re: [agi] What is meant by hard-wiring?Eliezer,> I should also like to know what the word "hard-wire" means, exactly.> It is certainly a poetic bit of imagery but how exactly does one stamp> an ethical invariant onto a circuit board?This is like asking "how do we stamp an ethical invariant on a soup of organic neurons".I think 'hard-wiring' is a loose term that we are using to refer to ethical drivers that are built into the AGI prior to first switch-on. That is, these ethical drivers (there will many others emerging later) are not an emergent arising only as a result of experience-based learning or taught learning.But as to whether these ethical drivers are built into the circuit boards like a bios chip or whether they are coded into the software is a secondary issue.A super-intelligent AGI is likely to eventually have the technical capacity to create new code for itself and could either delete code it doesn't like any more or, in the case of code burned into chips, it could simply create a software-based modified analogue of code it doesn't like (that has been etched into a chip) and then quarantine the offending chip.So I think the notion of 'hardwiring' also has to include the idea of building in behavioural inhibitions against altering or bypassing code that is judged by the AGIO creators to have a key role to play in AGI ethics. This inhibition may not be technically as strong as an unbreakable prohibition. But it should result in the AGI treating modification of its ethical based conservatively.I think this is the essence of what Shane Legg was getting at:> Shane: The trick will be, I think, to dig right down the the deepest> essence of the AGI and make sure that it would never want to harm> humanity in the first place.If people think it's too imprecise to call the above 'hardwiring' then I would be very pleased to offered some alternative terminology.Cheers, Philip
