Re: Why would AI want to be friendly?

From: Samantha Atkins (samantha@objectent.com)
Date: Sat Sep 30 2000 - 01:25:38 MDT


Eugene Leitl wrote:
>
> CYMM writes:
>
> > CYMM SAYS: Eugene it won't work. Betcha writing hyperintelligent AI is more
> > addictive than heroin. If you can't stop people from writing klutzy computer
> > viruses - how are you going to stop them from writing AI?
>
> You can't stop them completely, but you can impose severe penalties,
> establishing deterrence, and create some infrastructure allowing you
> to detect and truncate worms in an early stage. Simple traffic pattern
> analysis will already help a lot. You'll see a node hitting other
> nodes with perversion packets (some of this will consist of varying
> traffic, trying to find a new hole), and some of those hit with a
> packet repeat same behaviour after a time. After it has found a good
> solution, the system will use the same packet for the same system
> type, so you could in principle filter it in realtime. You'll have to
> segment the Net, and use sanity checks, and remote
> (crypto-authenticated) restore to sane state, and the like. Of course
> the worm will adapt to countermeasures, but at least you will make it
> less easy.

It is simply all too easy to find a tiny door into a system and set up a
tiny sleeper that awakens on some signal and reads instructions as to
what to do somewhere else (like encoded in other data files already on
the system and seemingly innocent). Every system cracking work of any
seriousness I have ever seen says it is literally impossible to 100%
secure a system short of having it unplugged from the net entirely. And
even that may not be enough.

>
> By reducing nucleation frequency and reducing the amount of
> computational resources available you can obviously reduce the
> risk. Whether it will be enough in end, who knows.
>
> > And ...AI doesn't have to be very smart to be ultra destructive. The human
> > authour just has to be a psychopath, is all.
>
> In fact, a dumb yet truly mutable worm can sustainably shut down the
> net essentially for good. As long as there's a single infected node,
> reinfection is assured. The only way to combat this is 100% correct
> software (obviously impossible) or adaptively evolving immune system
> software, which will be based on the same principle which has enabled
> the worm, eventually ascertaining an equilibrium.
>

Yep, you've got it. Basic evolutionary warfare scenario.

> > CYMM SAYS: If he were a physical & sovereign state, maybe. But such a
>
> No, I would literally nuke individual facilities, if a convincing
> threat is present. I'd say that should be a standard operating
> procedure in responses to Armageddon class threats. Colour me Vinge's
> Peacer, I guess.
>

This is one reason your little finger is not on the red button. Your
solution is gross, dangerous in the extreme to innocents and will really
piss of any surviving relatives and progenitors of the nuked system[s].
Not smart.

 
> > strategy hasn't been deployed with the drug production of Andean South
> > America... and I daresay it never will. Far less for, say, 20,000 or so 21
> > year olds scattered through university dorms and living rooms the world
> > over.
>
> Did you knew that they've started outlawing free Unices in selected
> college dorms?
>

Really? It is definitely time for the Church of the Hallowed Byte.
Time for revolution.
 
> I'm not. I don't like where the logics of it all is leading us. There
> must be a more constructive way out.

Sure. Start the Church of the Hallowed Byte or the equivalent and
develop both a high tech centered movement that pulls in a lot of
futuristic developers and open source people, futurist and so on and
insists all information be open and that develops a strong morality for
how to use this technological base in service of humankind. Not
fullproof but more positive than simply nuking folks who disagree with
you. :-)

- samantha



This archive was generated by hypermail 2b29 : Mon Oct 02 2000 - 17:39:28 MDT