Re: The wirehead problem

Tim Tyler <seemysig@xxxxxxxxxxxxxx> wrote:
Curt Welch wrote:

All the machines we use today are designed from the bottom up to be
controlled only by the will of humans. Where is your evidence to
suggest that will change one day?

It's not true today. Machines do all kinds of things against the
wills of humans:

"In 1981, Kenji Urada hopped a safety fence at a Kawasaki plant
to carry out maintenance work on a robot. While working on the
machine, the robot reached out and pushed 37-year-old Japanese
factory worker into a grinder with its powerful hydraulic arm.
Urada?s death is often said to mark the first recorded victim
to die at the hands of a robot, although Robert Williams was
killed by a robot two years earlier. Since both deaths, and
despite the introduction of improved safety mechanisms, there
have been many more gruesome industrial fatalities involving
robots crushing humans, smashing their heads and even pouring
molten aluminum over them."

Yeah, well, that's just silly semantics. Things get our of our control all
the time and hurt us. People have been crushed by a falling rocks for
millions of years. These industrial machines which are not called "robots"
are no different than someone getting crushed under a car when the jack
failed and that's been happening for 100's of years (if you consider a
carriage a "car").

The point is, every time an accident like that happens, we change the
design of the machine to reduce the odds of it happening again, and the net
result is machine designs that are generally safe for humans to be around.

An robot with inteligence built to protect itself from harm will be a very
dangerous machine to have around. They only way you can get it to do
things for you, is if you bargain with it, and give it something that helps
protect itself from harm (money), in trade of work performed.

My argument is that we don't need to build these types of machines in order
to take advantage of AI technology. Instead of giving it a prime
motivation of protecting itself from harm (like humans have) we can give it
the prime motivation of obeying and protecting humans. This means the
human will then be responsible for protecting the robot from harm. If the
human is stupid enough to tell it to jump off a cliff, the machine will
jump of the cliff just like a car will drive off a cliff if you tell it to.
So just like with all our machines, we have to take care of them and not
abuse them or else we will damage a very valuable machine.

As I said before, it's not a question of whether we will use the
machines to put everyone out of work - that will happen. It's a
question of whether their actions will always be controlled by humans.

FWIW, the main question I am interested in this area is whether
there is going to be a "machine takeover" - and I measure that
in terms of how much of the biomass is engineered (as opposed to the
product of random mutations and selection). When we get to 99%
engineered, I will be happy to claim that there's been a machine
takeover.

Well, then you consider "machines" part of the "biomass"? So if we just
turn all the rocks on the planet into machines, without changing the size
of the true biomass, you would call that a "machine take over"?

When I look around my house, I see 99% engineered structure, and 1% human
(and dog and cat and insect) biomass. Looking around me, I would say I'm
already at the 99% level. But yet, I'm still in control of all this stuff
around me. My chair isn't about to leave and go find a job down the
street.

Whether at that point there a few humans still on the scene seems
like a bit of a side issue to me. In the case above, I would guess
probably not - the remaining 1% seems mosly likely to be
some recalcitrant bacteria somewhere - not something human.

Yeah, that's all fine. I just don't believe humans are going to roll over
and die so easily.

If we can't build them as salves, we won't build them. Humans have no
reason to fill the world with machines they can't control and 1000
reasons not to do it.

So, can humans control motor vehicles? If so, how do you explain
all the traffic accidents?

Can you control a hammer and never make the mistake of hitting your thumb
with it? Can you control your leg and never make the mistake of kicking
something and hurting your toe?

The point is that we do far more good by moving our leg, than by keeping it
still to prevent kicking something and hurting our toe.

The control we have over the cars allows us to do far more good, than harm.
I've been driving cars for 35 years and I've never seen anyone killed by
one (except on the news). No one I know has even been killed in a car
accident. But yet, I've seen first hand millions of good things cars have
done for me and for others - that that would have been impossible without
the car. It's the ratio of good to bad that is important here. It's not a
question of absolutes.

If you allow supper intelligent robots motivated for survival by self
protection to be set loose on the world, the ratio of good to bad would
turn to the bad very quickly. They would harm humans in order to protect
themselves. Cars don't do that. If you hit them with a hammer, they don't
try to run you over - intelligent machines will do that.

Complete control over machines is not necessary to be able to make
money off them.

That's very true. But my real belief here is that we will never need to
build the type of machines you think we will build. You seem to assume
that intelligence means "will to survive". That anything which is
"intelligent" will act like a human and try to do things to protect itself.
If a human were to sit there and let something pound his face with brick,
we would say the human wasn't being very "intelligent". But a machine can
be intelligent without having a will to survive - or at least without that
being a prime motivation.

I think we can make intelligent machines perform all the tasks we currently
use humans to perform without having to give them self protection as their
prime motivation. To start with, much of what humans do don't require full
human intelligence by any measure. So you we can probably replace 99% of
all human jobs without having to build highly intelligent general purpose
machines. We given them only enough intelligence and training to do the
job we need them to do - like cook hamburgers without burning them, or
drive a car, or take an sales order, or answer simple questions about a
product or service.

If I want to build a machines I can't
control, that should be fun. No one does that however for very simple
reasons. They don't want to get hurt by a machine they can't control.

You keep bringing up these man-machine combat scenarios. I do not
think they are realistic. Sure, military AIs /might/ someday wipe
out a whole continent of humans - but basically I see a more peacable
future, where conflicts are mostly resolved under simulation.

I expect that machines will love us - and that we will love them.
Eventually, we may even love them more than we love each other.
We will want to be more like them - and for a few people, that
might even happen.

Well, this is the complex stuff that's hard to predict without having
working machines living with humans to see how these issues play out.

--
Curt Welch http://CurtWelch.Com/
curt@xxxxxxxx http://NewsReader.Com/
.

Relevant Pages

  • Re: "true" AI Hardware Development
    ... resistance to functional description at all. ... that we can build functionally equivalent machines. ... no two humans are the same. ... Do you understand that all computers are physical machines that work by ...
    (comp.ai.philosophy)
  • Re: Robot survival instincts
    ... would die fairly quickly if the humans died. ... be created by the smart machines. ... that will turn control of the earth, ... maximize the odds of our genes surviving and nothing else. ...
    (comp.ai.philosophy)
  • Re: The wirehead problem
    ... I love the idea of robots and many other people in AI or that like sci-fi ... When has evolution created friendly humans? ... independent survival machines that will end up competing with us. ... I just don't see it running out of control in only 100 ...
    (comp.ai.philosophy)
  • Re: Robot survival instincts
    ... But if we killed the humans, ... I don't really buy your premise that machines are going to take over ... that will turn control of the earth, ... maximize the odds of our genes surviving and nothing else. ...
    (comp.ai.philosophy)
  • Re: Merging with machines inevitable, scientists say
    ... I see the odds of the machines taking over being exceptionally slim. ... Humans are programmed by evolution with the goal of surviving. ... build AIs with the goal of surviving. ... advanced AIs for the sole purpose of taking over control from the humans. ...
    (comp.ai.philosophy)