Re: Robot survival instincts

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

That last point is the one which sums up the error:

Sooner or later our
machines will become knowledgeable enough to handle their
own maintenance, reproduction and self-improvement without
help. When this happens the new genetic takeover will be
complete.

This is just totally wrong. It's based on the false assumption that
"intelligence" = "a desire to survive". It's based on the assumption
that since humans use our intelligence to do all this complex stuff to
help us survive, that all intelligence will naturally work the same
way. I don't believe that. I believe "intelilgence" = "reinforcement
learning machine". Reinforcement learning machines have no desire to
survive. They only have the innate desire to produce behaviors which
will maximize their reward signal. The behavior the intelligent
machine ends up producing will be whatever it takes to make the reward
system produce more rewards.

Our DNA gives us a complex reward system that motivates us to produce
highly advanced survival behaviors. And evolution and natural
selection is hard at working making sure that continues to be what
happens.

We will build AIs that are as smart as we are, and once they get that
smart, they will be able to design and build other AIs for us, and they
will be able to maintain themselves. But they will NOT be built to
survive. [...]

Even Asimov's robots had survival instincts. The reason is obvious:
robots cost money.

Sure, that's true. But it will depend on the robot how that works.
Machines will need some concept of what they are worth (to us) so that they
can make good decisions about how to best serve us. A car-bot needs to
know that not getting into an accident is more important than being late on
the delivery of the ice-cream cone. So it needs to understand it's worth.

But an intelligent agent we use for research on the internet which exists
as computer hardware in a protected server room has no real ability to harm
itself, as such, we have little to no need to give it any concept of it's
self worth in order to prevent it from harming itself. If we decide to
turn it off, it should be just as happy with that as a TV is and not try to
protect itself from being turned off.

Something like a car bot might operate as a business. That is, you have to
pay it to do something for your, and there's a free market for all the
car-bots. They might have to pay for their own maintenance and as such,
they know that a $10 delivery job they are on is not important enough to be
worth getting into an accident that will cause $1,000 worth of damage to
the car. But if their only motivation is to operate as efficiently as
possible (no need or desire to make a profit), they don't need a survival
instinct.

It will be complex psychology problem to correctly designing and motivate
these machines so they do the best job while not being a threat to humans,
but I'm sure there will be ways to motivate them so they understand the
value of not harming themselves, while at the same time, value the needs of
humans far above themselves.

Instead of making a car-bot smart enough to operate as it's own business,
we might make it fairly dumb, so the car only has enough intelligence to
get from place to place without hitting things, and then give a remote
server-based AI the job of acting as the dispatcher.

Maybe the way we make the AIs work for us without the threat of them
turning into survival machines and puting their own value higher than that
of humans is to use only dumb AIs where their intelligence is optimized and
trained to do only one simple task - like car driving - or car dispatching,
or traffic optimization, or car body repair or tire changing, etc. So we
have a large army of machines, which might have the average intelligence of
an chimp, but which are highly optimized and trained to do only one very
specific job.

I think you had better find some other way to
express the point you are trying to make here in the future.

I'm always working on that. :)

Anyway, I deny that a machine takeover would necessarily be
based around humans giving robots survival instincts. We
could get to a 99% machine society simply by using robots as passive
tools, with no survival instincts.

I don't know, if I count the number of smart machines I own (aka have a CPU
in them), I think it's over 100. Which puts us in that 99% machine society
already. :)

The machines could become
numerous because they are useful - not because they have been
constructed with survival instincts.

Yeah, but if I replace my car with a smart-car that can drive itself when I
ask it to take me to the store, and the gas station I get gas at is smart
enough to fill the car up without me getting out of the car and without a
human to watch over it, and the car is smart enough to get itself serviced,
the world won't be very different from what it is now.

Yesterday, the check-engine light on my car came on, so I had to drive to
the service center and have them check it out. A vacuum hose had slipped
off. I spent 3 hours taking care of this. But the car was already smart
enough to know it needed to go get service. With a little more
intelligence, it could have been smart enough to ask me if I wanted it to
get service and then drive itself to the dealer and come back when it was
done. With a little more intelligence in the world, most of the service
could have been automated where the car just got itself in line at the
dealer and had a human diagnose and fix the problem so only the car
mechanic would need to be human and the rest of the system (all the people
that take care of the paper work) could be eliminated.

With a bit more intelligence, a robot does the maintenance work as well.

How many more "machines" would we have to add to the large set of machines
that already exist to make this happen? Not many really, because all the
people at the shop are already using computers to track all the work. And
my car already has multiple computers in it. And the tech I'm sure hooked
up a diagnostic computer to the car to find out what the car thought was
wrong with it. And the tech has a large collection of tools and machines
to do his job already.

It's only an issue of making the machines that are already there smarter,
and adding some robots to replace the arms of the mechanics and there we
have full automation of a process like car maintenance.

And in the whole cycle of car maintenance, nothing is done without the
permission of a human. My car doesn't go off to get fixed without getting
my permission, the AI mechanics don't charge me anything without first
getting my permission, and the AI mechanics have no motivation do anything
other than serve the humans.

And then we need to office workers who design and build these automation
systems to be replaced by AI's. But all those office workers are already
sitting in front of computers and talk to the other office workers using
computers and computer networks. So all we need to do there is upgrade the
computer so it doesn't need the human anymore.

So we end up with a world which is much like it is today except we have AI
software running in all the computers instead of just Microsoft Word, and
we don't need all the wasted space for office buildings, we just need lots
of server rooms.

But how do we get from this world of intelligent machines, which is
organized into a society which has one top level purpose - to do what
humans ask them to do - into a "machine takeover".

If I were to die, my car would simply sit in my garage and do nothing - it
would wait for me to return and ask it to do something. If it had a
maintenance issue, it would not go get it fixed on it's own, because it
couldn't. It wouldn't have permission to leave, and even if it did leave,
the mechanics couldn't work on the car without my permission.

We would build an army of intelligent servants - just like our machines are
servants today. If I go away, my computer won't decide on it's own to
start printing emails without me. If turn my computer off, it has no
instinct to deny my request (aka no survival instincts).

The smart machines of the future might be able to respond more
intelligently, but the end result is they will still be built as servants
to man and if there are no humans around the serve, they will just power
down and wait for a human to show up again.

Even with intelligence I don't think we will build the types of machines
that will continue on without us. They entire society of machines will act
like one huge intelligent vending machines for humans, and if the last
human dies, the machines will simply have nothing to do. They won't
suddenly decide to get drunk and party, or go explore the stars. They will
simply do nothing, or they will just keep themselves functioning waiting
for the humans to return - like the power stations will keep themselves
running - even though there is very little power demand now that there are
no humans asking for things. The security systems will keep monitoring for
dangerous activity, etc, but the machines won't "take over" any more than
our machines today will "take over" when we leave the house.

The "survival instincts" can stay with the humans and the
companies, for the sake of argument. Or perhaps more
accurately, with the genes and the memes.

IMO, the key to understanding all this lies with memes:

Think of something like Linux as a memetic organism with
complex adaptations designed to ensure its survival. If
Linux was not concerned with its survival, we wouldn't
see it around. It will be the same with robots.

I'm not seeing it. Linux survives because it exists in an environment that
allows it to survive. But if we killed the humans, Linux would no longer
have an environment in which it could survive, it would die with the
humans.

Change the environment of earth to one which had no atmosphere, and most
the life on the planet would die. Wipe out lower life forms, and the life
forms higher on the food chain die because they no longer have an
environment they can survive in.

The AI machines we build, will depend on us for their survival as much as
Linux does. They will exist only to support us. It would be against our
interest to design and build, or allow, the machines to become
self-motivated survival machines that value their own existence over human
existence. As such, their survival will depend on our survival.

The blueprint in the robot factory is the robot heritable
material, and it is the survival of *that* (and _not_
individual robots) which must be considered.

Sure, I see that.

That /will/
have complex adaptations to ensure its own survival.
The survival of individual robots is of low relevance.
They are like worker ants: mere phenotype - genetic
dead ends - disposable soma.

But what is going to _cause_ the blueprint to become a new robot? What is
going to cause the design of the robot to be improved and produce new
better robots?

The entire robot society will be built for only one purpose - do what what
humans want them to do. As such, they will exist, and do all their work,
only to serve us. If we aren't there to serve anymore, they will stop
building new robots, and stop trying to improve their design because there
is no purpose to it. They will just become depressed and sit around
moping.

But, there is another possibility.

What if we humans want robots that act more like humans than slaves? What
if humans start to build robot best-friends or even lovers, because with a
robot, you can have anything built you want - any look, any personality,
and they can be hard wired to love you if that is what you want, or hard
wired to be independent thinkers with their own needs and wants - aka, we
build "friendly" survival machines. We can build them as robot children as
well (like the AI movie) - that is, they will be given to us as if they
were babies to raise and shape as we want - and after years of raising and
training them, they will become friendly independent adult robots. We
could choose to do that simply because we enjoy raising and caring for an
intelligent agent and helping it grow and learn. What if the nurturing
instinct given to us by evolution was used to raise robots instead of
humans?

If we choose to build machines like that, then they would have the power to
be part of a society, and not just part of the big robot vending machine
system that did all our work for us. And those machines would have the
power to take over control of society from the humans - if we choose to let
them.

But will that happen? I can't tell. It seems unlikely to me, but I can't
tell.

It seems likely that we would build entertainment robots and robots that
acted a lot like humans for many reasons. The idea of having smart
robot-friends and robot-children would seem to be popular for many people.
But I can't see where human society would every allow those robots to
become part of human society and be allowed to vote, and be given equal
rights with humans to control the allocation of resources.

This question of control over allocation of resources is a key issue. We
already are seeing the problems of over population by humans. Because of
the number of humans we have allowed to be born, combined with advancing
technology, the earth is being changed in ways that is irreversible.
Hundreds of species are going extinct every year because of human
overpopulation and we will never be able to bring them back.

Human wealth was limited mostly by what a human could produce in their
life. Our work not only kept us alive, but it allowed us to transform our
environment into a better place to live. In the past, there were so few
humans, and so much environment, that there was more than enough to go
around. You could spend your whole life chopping down trees and building
homes and stuff and not make a dent in environment. But now, with
population growth and technology, resources are becoming scarce, and each
human alive is only allowed some small part of the environment to shape as
their own. As technology continues to advance, and population continues to
grow, the real limit on what we can have, and do, will be mostly a function
of human population. If we have smart machines doing all the work for us,
then we don't need a large work force of humans to do stuff for us (like
build our homes and our cars, and make food for us). The only thing humans
will do for us, is consume resources that we would rather have for
ourselves. If we kill off half the population of the world, then the
remaining half could become twice as rich - twice as much energy, raw
materials, and space, to do with as we want.

On the other hand, if we let the population continue to grow without
bounds, all those new people will just be taking resources away from us,
and from our children.

The end result I'm sure will be forced population control (if it doesn't
regulate itself that is). Once we have a huge society of smart machines to
take care of us, humans won't want billions of other humans around because
they do nothing for us except mess up the environment and limit our
resources.

In a world like that, where we can't even allow more humans to be born who
will be given the right to control the allocation of resources, why would
we allow people to add robots to the society? One robot-loving old lady
could raise a thousand robot children and thank you, but no, I won't allow
them to add to the population of intelligent agents trying to get a slice
of the control of the resources of the earth. I won't allow them to dilute
my share of control over the resources. I wouldn't even allow the 1000
robot-children to share the slice of control that belonged to the old lady
when she died. I want those resources returned to the pool of resources
that only the living humans have control over.

Maybe people will like the robots so much, that this won't be how it plays
out. Maybe the people will just turn control of society and control of the
government over to the robots in time. Maybe, people will start to like
the robots more than they like the other people. Maybe, since we will get
to a point where we will depend on the machines instead of depending on
other humans for our survival this will happen and humans will start to
fight with each other, and the end result is that we will give control to
the robots because we don't like humans anymore. But it just doesn't seem
possible to me because no matter how much we don't like each other, I think
we will always do better living under the control of the majority of
humans, than under the control of a majority of AIs.

So my vote is still that humans will remain in control, and the machines
will be slaves to us, just as much as they are slaves to us now. They will
just be much smarter slaves then they are now. There just won't be a
machine take over - at least not for thousands, or millions, of years.

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

Relevant Pages

  • 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: "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
    ... 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)
  • Re: The wirehead problem
    ... > humans or otherwise harm the things that humans love. ... But a super intelligence that can build as many robots as it wants ... on how many machines you can have. ... primarily built by genes - not memes. ...
    (comp.ai.philosophy)