This works great in the short term, and on paper. But there is nothing
limiting the machines to this viewpoint. There are just too many bad
scenarios that are easy to see here.
The AI's 'love' us and work with us cause it is beneficial to them.
So what happens when:
1. Humans get jealous, as always, and decide the AI's are too smart,
dont like what the AI's are doing, and we pass laws limiting their
growth / building or some groups start attacking them. Will they
defend themselves? Should they?
2. We limit their growth / procreation or some vital building
material like metals. They cant get to the things they need because of
us. Its now no longer benificial for them to work with us, we are
withholding vital parts to them.
3. Random glitches. All computers have random glitches, so its real
easy to imagine someones goal reversing backwards, or an AI just
getting confused in its goals and hurting someone.
Once we fully declare that yes, we can build a AI that is as smart as a
human, all bets are off really. They may act and react like a human in
many ways, but for how long? And the intelligence growth once they
reach our level will be exponential in comparison with ours, they will
very quickly be able to leave us behind.
I also have a big big problem with people that just say, well we'll
program them to love us, and to always be nice to us. It takes one
rougue programmer, or one mistake to leave out the "Morality" chip, and
it could spread throughout the society and be undone in literally
moments.
If they are truly as smart as us, tehy can remove components and add
components at will, and will.
James Ratcliff
The point is that in my homecountry, Finland in Europe, objectivity
equals moral. So would it be for AI too.
Kaisa Tervola
Computer programmers are tinkerers, and they will fiddle with the code
and with the AI, as soon as its released.
I thought about your solution of requiring all AI's to have the "moral
code" but that only works in a small closed system. So maybe as long
as the AI's are totally under control of a small group, say the
government research, then thats fine, they have restrictions and the
setup to try and prevent rogue elements.
Otherwise I guarentee the first line of AI's that are given out to
regular households will be broken into. There will be people hacking
into it, and modifying its programming immediatly. If not for pure
evil reasons, then just to push the AI to the limits of what it can do.
After that it would only take one person to publish his findings, to
tell exactly where the "moral chip" can be removed, or a patch that can
ovverride the "moral code" of any AI. Then what is to prevent it from
communicating with every other AI in the world?
JR