Re: a little thought experiment


J.A. Legris wrote:

bob the builder wrote:
J.A. Legris wrote:
bob the builder wrote:
Since this is comp.ai.philosophy, a thought experiment about AI would
seem to be appropriate. Keep in mind that this is a thought experiment,
all the strange things inside it are all true :)

Now suppose you would have made some sort of ingenious architecture
made out of positrons (they are a bit like neurons). And this
positronic network gets initialized by a genetic algorithm of some
sort. Further more you accept that it has the following
behaviour/feautures:

*The network easily scales up. If it takes 5 minutes to solve a
problem with difficulty 1, it takes 10 minutes to solve a problem with
difficulty 2. The architecture doesnt fail to scale up at some
threshold of complexity.

*The Genetic algorithm often finds a solution a little better than the
best in the previous population. So it never gets stuck.

Now ofcourse you are very excited and you want to show the rest of the
world that your architecture can deliver strong AI. You decide to
illustrate its potential by putting the network in a simulated ant
situated in a simulated landscape. Because its so obvious that your
positronic network can scale up with great ease you decide to let the
network solve some small problems (NO BIG ONES!).

Now my question to you: Which small problems would you choose to be
solved by your simulated ant? Keep in mind that you want to keep things
as simple as possible. Your goal is to let other people think: "if it
can solve THAT problem(a) , and it can scale up with great ease(b) ,
than its capeble of strong AI (c)". The people already believe
that (b) is true.


I will kick off: The ant has to be able to walk.

The ant has to be able to walk on two legs, leaving two to hold the
instruction manual open and two more to do the work.

Well thats at least a try, in some way, i suppose.
Walking on two legs.

OK, I'll get serious now. Just get your artificial ant to do everything
a real one does. Forget about the scaling up part, we already know that
works. But if you can make a realistic artificial ant you'll know so
much that the quest for strong AI will look boring and irrelevant.

So you think a real ants Intelligence stands to human intelligence as
a the first IBM pc stands to the latest Intel Core Duo 2 machine?
Its just the same, scaled up, architecture?

I personally think that if the network can let the ant behave just like
a real ant, then you are done.

.

Relevant Pages

  • Re: a little thought experiment
    ... Keep in mind that this is a thought ... *The network easily scales up. ... The architecture doesnt fail to scale up at some ... illustrate its potential by putting the network in a simulated ant ...
    (comp.ai.philosophy)
  • Re: RISCOS 4 basic networking
    ... > Yes this puzzles me too as a long time Ant User. ... > Given a RiscPC running 4.02 and a broadband router, ... All you have to do is use the Universal Boot and, in the Network config: ... If you can't get the browser to work, ...
    (comp.sys.acorn.networking)
  • Re: a little thought experiment
    ... *The network easily scales up. ... The architecture doesnt fail to scale up at some ... illustrate its potential by putting the network in a simulated ant ... simple reinforcement learning problems and show that it is better at ...
    (comp.ai.philosophy)
  • Re: a little thought experiment
    ... Keep in mind that this is a thought ... *The network easily scales up. ... The architecture doesnt fail to scale up at some ... illustrate its potential by putting the network in a simulated ant ...
    (comp.ai.philosophy)
  • Re: a little thought experiment
    ... Keep in mind that this is a thought experiment, ... *The network easily scales up. ... The architecture doesnt fail to scale up at some ... illustrate its potential by putting the network in a simulated ant ...
    (comp.ai.philosophy)