Re: Takeover
- From: "zzbunker@xxxxxxxxxxxx" <zzbunker@xxxxxxxxxxxx>
- Date: Sun, 4 May 2008 06:31:32 -0700 (PDT)
On Apr 28, 3:28 pm, c...@xxxxxxxx (Curt Welch) wrote:
Tim Tyler <seemy...@xxxxxxxxxxxxxx> wrote:
Curt Welch wrote:
> Over the very long run, you can believe it just will find a way to
> happen if the other forms of replication are better (which I'm not
> sure is true BTW because DNA is about as small and efficient as you
> can get so making it "better" seems to be hard to me). But in the
> shorter term, it won't happen just because we can let it happen any
> more than man would let a small group of aliens come in and take over
> without a fight.
Well, that's a discontinuity. The machines will be our descendants.
Or at least they will contain heriatable information that originated
with us.
During a takeover, there's often a period of symbiosis where
both technologies exist, and information can pass between the two -
and that looks to be what is happening in this case. So, this
won't be simply an invasion of machines. The machines will be
our symbiotes. They will be part of our civilization. I don't
really forsee a fight. The machines will love us - and will want
to take at least some of us with them.
In the long run, I can see that happening. I see it as almost required
simply because I don't believe biological forms can evolve fast enough to
stay ahead of engineered life forms. But it seems to me that anything like
that will slowly evolve over thousands of years, and not be something that
will happen any time soon.
Well, that's true, Since your example is the perfect of how humans
mistake A,I for metal boxes.
Since it has more to do with satellites, chess, nano-tech, lasers,
masers, and cruise missiles than it does idiots in cars.
The thought that keeps tripping me up however is the requirement for
intelligence to be part of the loop to make that work. That is, for an
intelligent agent to decide to reproduce, or to build new machines that
are better and more advanced than themselves.
Because human intelligence is the only example we have of strong
intelligence, I suspect most people trying to speculate about the future
effects of AI are using the working assumption that AIs will act like
humans with metal bodies.
But I suspect intelligence can be shaped in ways unlike anything we have
seen and unlike most can imagine. Humans today are for the most part, very
social creatures, with what seems to be an innate desire to advance, and to
nurture, and to survive. We work together to create a culture, even while
we compete with each other. I believe a lot of people associate all these
sorts of human traits with intelligence.
But I think intelligence at the core is just a raw power to learn complex
behavior, and nothing more. What behavior an intelligent agents learns is
a function of the environment - and most important - the critic hardware
which sends rewards to the learning system.
Humans can be raised in different environments, but they can't change their
critic hardware. That's built in us by our generics. For the most part,
all humans have very similar critic hardware. When you combine a similar
size and power learning machine, with a similar cirtric system, in a
similar environment, you get lots of people with very similar learned
behaviors.
When you raise a person in a very different environment, they can develop
fairly unique and different behaviors. But those minor environment effects
are nothing compared to what a different critic would cause the system to
learn.
When we master the technology of strong learning systems, we will have the
power to build any sort of critic into the system that works for us. As a
result, we will see first-hand what happens when you combine supper-string
learning (super inteligence) with a critic which motivates the machine to
things very different from what any human would do.
We will want, for example, to make our AI cars, expert drivers. What type
of personality ends up developing when you combine super strong learning
with a complex critic which motivates the system to be the best car driver
possible? I don't know. But I suspect it will be so much unlike any
intelligent human that it will shock us. I suspect it will be absolutely
obsessed with doing it's job of transporting people and goods from one
place to another in minimal time and with minimal waste of energy and
minimal impact to the other cars etc.
We might be able to create such a machine with 10 times the intelligence of
a human, and have no risk or desire on the part of that car to taking over
control of the earth from humans. All it would want to do is transport
stuff.
I suspect the world will be full of these super intelligent machines, all
custom designed, and motivated to do only the job they were engineered to
do. The motivations and drives will be completely unlike human
motivations.
As such, I think the idea that these machines will take over control of the
earth, is as far fetched as believing the vending machines of today will
"want" to take control of the earth from the humans.
We will reach a point however where humans fully understand, without
question, what they are, and where they came from. In the case of our
intelligence and consciousness, I think that's going to lead to an
understanding that we are just learning machines forced to follow our
motivations. When the day comes that every machine around us (our car, the
vending machines, the factories that make all our food and take care of
us), are all filled with the same type of intelligent machines, there will
be no doubt what we are. And we see how easily their behavior is shaped by
the critic hardware in them, people will fully understand that we are just
machines shaped by the critic hardware in us.
This is going to make people look at the issues of morality, and right and
wrong, very differently than most people seem to look at it today. What is
it right to have children and take care of them? Why is murder so wrong?
Why is survival important?
All these issues which are so important to humans, are important to us only
because we have hardware in us that makes it important to us. There is no
higher purpose here. We exist, and we have that hardware in us just
because natural selection caused it to form in us. But, as our technology
advances, we will fully understand what that hardware is in us, and how to
change it. We will have the power to pick any morality we want to pick and
build it into us, or build it into our children.
The AI machines will build will have this power before we do because we
will figure out how to engineer the AI machines before we master the
complex task of re-engineering humans at that level.
What will a society of humans, or intelligent machines, do, when they
realize they do what they do, and they care about the things they care
about, only because of how they are built. And given the option to modify
themselves, or to build new AI's with any type of difference they want to,
what will they do? How will this level of understanding, and power over
ourselves, and power over our future evolution, change our society in time?
I can't really decide what will happen. But I sense it's not obvious that
we will naturally continue to survive. I can't help but suspect that
intelligence and long term survival are mutually exclusive. I wonder if it
might be a natural response for a society of intelligent agents to stop
caring about surviving once they become intelligent enough to figure out
why they cared so much about it in the past. It might tend to work like a
big house of cards that all falls apart once the intelligent agents figure
out what's behind the curtain. I'm not suggesting that we become depressed
and commit suicide - but that intelligent agents are destined to follow
some form of the wire-head path and party till they drop. They figure out
how to create ultimate happiness for everyone, and if making more machines
is not part of that formula the society just dies out in time. They die
very happy, but the intelligent race dies out none the less.
It's ideas like this make me suspect intelligence is not a key to survival..
And that survival only happens when it's directed by low level
non-intelligent processes. As such, any path of evolution which gets too
intelligent, might just tend to burn out, leaving the less intelligent life
forms to continue. Which makes me suspect that the idea that AI machines
will takeover the earth is something that just can't happen. The discovery
of what makes us intelligent might just be one more step closer to the day
that the human race dies out and leaves the world to the lower animals and
the bacteria.
--
Curt Welch http://CurtWelch.Com/
c...@xxxxxxxx http://NewsReader.Com/- Hide quoted text -
- Show quoted text -
.
- Prev by Date: Teaching computers how to LIE (say untruths) to teach them LOVE / PAIN / FEELINGS
- Next by Date: Re: Existential risks
- Previous by thread: Teaching computers how to LIE (say untruths) to teach them LOVE / PAIN / FEELINGS
- Next by thread: Re: Majority rule
- Index(es):
Relevant Pages
|
|
