On 30 Mar 2013, at 13:58, Telmo Menezes wrote:
On Thu, Mar 28, 2013 at 1:23 PM, Craig Weinberg
<[email protected]> wrote:
On Thursday, March 28, 2013 5:52:04 AM UTC-4, telmo_menezes wrote:
On Wed, Mar 27, 2013 at 6:29 PM, Craig Weinberg <[email protected]>
wrote:
On Wednesday, March 27, 2013 1:03:27 PM UTC-4, telmo_menezes wrote:
Hi Craig,
On Wed, Mar 27, 2013 at 4:03 PM, Craig Weinberg <[email protected]>
wrote:
From the Quora
http://www.quora.com/Board-Games/What-are-some-fun-games-to-play-on-an-8x8-Checkerboard-besides-chess-checkers
This is interesting because I think it shows the weakness of the one-
dimensional view of intelligence as computation. Whether a program
can be designed to win or not is beside the point,
That's not really fair, is it?
Why not?
How else can I counter your argument against intelligence as
computation if I am not allowed to use computation? My example would
not prove that it's what the brain does, but it would prove that it
can be. You are arguing that it cannot be.
I'm arguing that a screw is not the same thing as a nail because
when you hammer a screw it doesn't go in as easily as a nail and
when you use a screwdriver on a nail it doesn't go in at all.
Ok.
Sometimes the hammer is a better tool and sometimes the driver is.
As humans, we have a great hammer and a decent screwdriver. A
computer can't hammer anything, but it has a power screwdriver with
a potentially infinite set of tips.
Ok, but if I understand your ideas, you're claiming that the hammer
is also the fundamental stuff that reality is made of. Sorry if I'm
misrepresenting what you're saying. If I'm not, I don't understand
why computers can't have the hammer.
as it is the difference between this game and chess which hints at
the differences between bottom-up mechanism and top-down
intentionality
I see what you're saying but I disagree. It just highlights the weak
points of tree-search approaches like min-max. What I gather from
what happens when one plays Arimaa (or Go): due to combinatorial
explosion, players (even human) play quite far away from the perfect
game(s). The way we deal with combinatorial explosion is by mapping
the game into something more abstract.
How do you know that any such mapping is going on? It seems like
begging the question.
I don't know. I have a strong intuition in it's favor for a few
reasons, scientific and otherwise.
Have you tried thinking about it another way? Where does 'mapping'
come from? Can you begin mapping without already having a map?
Yes, I think I begin with a map based on previous experiences and
then improve it as I discover it's weaknesses. I think the original
map came from brute-force experimentation while my brain was
developing in my early months of live. But this is just wild
guessing, of course.
The non-scientific one is introspection. I try to observe my own
thought process and I think I use such mappings.
Maybe you do. Maybe a lot of people do. I don't think that I do
though. I think that a game can be played directly without
abstracting it into another game.
Ok, I believe you but I don't have the same experience. My wife
does. She works in a creative field and she is very intuitive, with
the typical aversion for math. She can beat me at chess quite
easily, without appearing to resort to conscious strategic thinking.
She describes it as doing what "feels right".
The scientific reason is that this type of approach has been used
successfully to tackle AI problems that could not be solved with
classical search algorithms.
I don't doubt that this game is likely to be solved eventually,
maybe even soon, but the fact remains that it exposes some
fundamentally different aesthetics between computation and
intelligence. This is impressive to me because any game is already
hugely biased in favor of computation. A game is ideal to be reduced
to a set of logical rules, it's turn play is already a recursive
enumeration. A game is already a computer program. Even so, we can
see that it is possible to use a game to bypass computational values
- of generic, unconscious repetition, and hint at something
completely different and opposite.
Put another way, if there were top-down non-computational effort
going into the game play, why would it look any different than what
we see?
Our brain seems to be quite good at generating such mappings. We do
it with chess too, I'm sure. Notice that, when two humans play
Arimaa, both can count on each other's inabilities to play close to
the perfect game. As with games with incomplete information, like
Poker, part of it is modelling the opponent. Perhaps not
surprisingly, artificial neural networks are quite good at producing
useful mappings of this sort, and on predicting behaviours with
incomplete information. Great progress has been achieved lately with
deep learning. All this fits bottom-up mechanism and intelligence as
computation. It doesn't prove anything because I can't attach the
code for an excellent Arimaa player but, on the other hand, if I did
I'm sure you'd come up with something else. :)
Except that playing Arimaa is not particularly taxing on the human
player. There is no suggestion of any complex algorithms and
mappings, rather it seems to me, there is simplicity.
The mappings don't have to be complex at all (in terms of leading to
heavy computations). That's precisely their point.
Then shouldn't a powerful computer be able to quickly deduce the
winning Arimaa mappings?
You're making the same mistake as John Clark, confusing the physical
computer with the algorithm. Powerful computers don't help us if we
don't have the right algorithm. The central mystery of AI, in my
opinion, is why on earth haven't we found a general learning
algorithm yet.
Let us say that a program learn a function when it generates, from the
input-output description of a function, an infinite sequence of
programs eventually computing that function. They are looking to
growing sequence of input-output of the presented function. Then it
has already been proved by Putnam, by a diagonalization argument, that
such a universal learner does not exist.
Now, you can weakened the identification criterion, by allowing that
the produced program makes an arbitrary big number of mistakes, and by
allowing the learner machine to change its mind (change the program
she guessed), even when the program she guessed provide a correct
solution. Amazingly enough this leads to universal learning algorithm,
but they are impossible to be used in practice (for theoretical reason).
This is not a problem for strong AI, nor for comp, as universal human
learner does not seem to exist too. Learning competence is not
scalable on a linear order ladder, but on infinite lattice with a lot
of incomparable degree. It can be shown in general that machines
making mistakes, and working in team, are far more better learner than
any individual machines, especially if they never make mistakes.
See:
PUTNAM, H., 1965, Trial and error predicates and a solution to a
problem of Mostowski,
Journal of Symbolic Logic, 30, 1, pp. 49-57.
GOLD, E. M., 1965, Limiting recursion, Journal of Symbolic Logic, 30,
1, pp. 27-48.
BLUM L. & BLUM M., 1975, Toward a Mathematical Theory of Inductive
Inference.
Information and Control 28,.pp. 125-155.
CASE J. & SMITH C., 1983, Comparison of Identification Criteria for
Machine Inductive
Inference. In Theoretical Computer Science 25,.pp 193-220.
Bruno
Either it's too complex for our monkey brains, or you're right that
computation is not the whole story. I believe in the former, but not
I'm not sure, of course. Notice that I'm talking about generic
intelligence, not consciousness, which I strongly believe to be two
distinct phenomena.
The human finds no fundamental difference between the difficulty
between Arimaa and Chess, yet there is a clear difference for the
computer.
Yes, the classical chess algorithms are clearly not how we do it. I
agree with you there.
Again, if this does not indicate that there the model of
intelligence as purely an assembly of logical parts, what actually
would? In what way is the Strong AI position falsifiable?
I agree, I don't think it's falsifiable and thus not a scientific
hypothesis in the Popperian sense. I see it more as an ambitious
goal that nobody even knows if it's achievable. You might be right,
even if we manage to create an AI that is undistinguishable from
human intelligence. I prefer to believe in Strong AI because I'm
interested in it's consequences and in the intellectual challenge of
achieving it. That's all, to be honest.
On the other hand, your hypothesis is also not falsifiable.
Sure, but mine doesn't pretend to be as it calls the whole notion of
falsifiability as a criteria for addressing awareness into question.
Strong AI though is a bet that awareness can be constructed through
logic, and logic, especially digital logic, is all about the
exclusive supremacy of true/false positions. I don't think that
Strong AI is a bad goal if it drives curiosity and development - my
beef with AI is completely incidental, I generally am happy with
computer science and consider it one of the few healthy parts of
civilization that remains.
I think there's a very simple reason for that. Nobody can prevent me
from experimenting with my computer right now. I can try a new
algorithm without any paper work, permission from some regulatory
body or even money. It's free as in beer and free as in speech.
People can replicate my results with similar ease. So it's the only
pure science left. It's also why I love computers so much.
I only argue against Strong AI because when I argue for this new
picture of mind-matter-information unification, the argument that
fights back is rooted in these assumptions about information states
being concretely real...information states which we have knowingly
contrived for the purpose of developing technology. We invented the
Gigabyte and now we worship it as a mechanical anti-God. It's more
of a public service than anything else to try to point out why that
may actually prevent us or side track us from understanding
consciousness. Strong AI may not really want to understand
consciousness, and that's fine, but for those who do want to
understand the physical phenomenon of awareness, they are being led
down a dead end path by some very smart, very enthusiastic minds.
I completely agree with you when it comes to consciousness.
Best,
Telmo.
Thanks,
Craig
A lot of progress has been made in Poker, both in mapping the game
to something more abstract and modelling opponents:
http://poker.cs.ualberta.ca/
Cheers,
Telmo.
PS: The expression "brute force" annoys me a bit. It implies that
traditional chess algorithms blindly search the entire space. That's
just not true, they do clever tree-pruning and use heuristics.
Still, they are indeed defeated by combinatorial explosion.
It was a generalization, but I understood what they meant. The
important thing is that the approach of computation is fundamentally
passive and eliminative. Games which do not hinge on human
intolerance for tedious recursive processes are going to be easier
for computers because machines have no capacity for intolerance. The
more tedious the better. Games which de-emphasize this as a criteria
for success are less vulnerable to any recursive elimination. The
more a game can reward spontaneous creativity, versatility, style,
grace, broadminded eclectic interpretations, the more a computer
will fail to duplicate a person's success.
Craig
.
In Arimaa, the rules invite personal preference as a spontaneous
initiative from the start - thus it does not make the reductionist
assumption of intelligence as a statistical extraction or 'best
choice'. Game play here begins intuitively and strategy is more
proprietary-private than generic-public. In addition the interaction
of the pieces and inclusion of the four trap squares suggests a game
geography which is rooted more in space-time sensibilities than in
pure arithmetic like chess. I'm not sure which aspects are the most
relevant in the difference between how a computer performs, but it
seems likely to me that the difference is specifically *not* related
to computing "power". To wit:
"There are tens of thousands of possibilities in each turn in
Arimaa. The 'brute force approach' to programming Arimaa fails
miserably. Any human who has played a bit of Arimaa can beat a
computer hands down."
This to me suggests that Arimaa does a good job of sniffing out the
general area where top-down consciousness differs fundamentally from
bottom up simulated intelligence.
--------------------------------------------------------------------------------------------------------------------------------------
Arimaa, the strategy game that confounds computers!
It can be played, not only on an 8x8 chess board, but with the same
chess pieces as well!
The pieces are :
8 Rabbits (Pawns)
1 Elephant (King)
1 Camel (Queen)
2 Horses (Rooks)
2 Dogs (Bishops)
2 Cats (Knights)
It doesn't matter in what way you want the 2 horses/dogs/cats to be
designated by the 2 bishops/knights/rooks.
What sets apart Arimaa from Chess?
There is no draw in Arimaa. Good news for elimination tournaments.
In Arimaa, a player has 64,864,400 choices for the first turn. Thus
unlike chess, memorizing openings is not gonna help you.
There are tens of thousands of possibilities in each turn in Arimaa.
The 'brute force approach' to programming Arimaa fails miserably.
Any human who has played a bit of Arimaa can beat a computer hands
down.
It places less emphasis on tactics.
I believe Arimaa is way better than chess in terms of abstract
strategical thinking. It needs a higher level of intuition and
understanding, discourages memorization and is simple to learn and
play. It took me some time to play good chess, but it took me a
small fraction of that time to learn and play good Arimaa.
The Arimaa community is offering $10,000 for anyone who can come up
with a program able to beat a top-level human Arimaa player, by
2020 : The Arimaa Challenge
This will help us to attain the next pinnacle in Artificial
Intelligence Programming.
Rules :
In the starting, both players arrange the pieces in whatever way
they fashion in their first two rows, something like this :
The pieces can move only one square horizontally or vertically. In
case of rabbits, you can only move upwards or sideways. You have
four moves to play in each turn.
In order of their power, the pieces can either 'push' or 'pull'
other pieces of the opponent. In addition to this, if a less
powerful piece of yours is adjacent to a more powerful piece of the
opponent's, then your piece will be frozen, unless your piece is
adjacent to another one of your pieces.
The order of power is as follows :
Elephant > Camel > Horse > Dog > Cat > Rabbit
That is, your camel will be able to push or pull the opponent's
horse/dog/cat/rabbit. You can freeze the horse/dog/cat/rabbit if it
doesn't have any friendly piece adjacent to it.
Elephants are all-powerful : they cannot be pushed, pulled or frozen.
See those dark squares in the diagram above? They are 'Trap
Squares'. If any of your piece lands in here and if there is no
adjacent friendly piece to it, your piece will be 'captured'.
So, How do you win?
Your goal is to get one of your rabbits to the last row(or home
rank). Whoever manages to do this first, wins.
If you manage to capture all your opponent's rabbits, you win.
If your opponent has no legal move, you win.
Since one of the above situations is bound to occur, there is no
draw in Arimaa. This is unlike chess where there is an unusually
high probability of a draw.
--
You received this message because you are subscribed to the Google
Groups "Everything List" group.
To unsubscribe from this group and stop receiving emails from it,
send an email to [email protected].
To post to this group, send email to [email protected].
Visit this group at http://groups.google.com/group/everything-list?hl=en
.
For more options, visit https://groups.google.com/groups/opt_out.
--
You received this message because you are subscribed to the Google
Groups "Everything List" group.
To unsubscribe from this group and stop receiving emails from it,
send an email to [email protected].
To post to this group, send email to [email protected].
Visit this group at http://groups.google.com/group/everything-list?hl=en
.
For more options, visit https://groups.google.com/groups/opt_out.
--
You received this message because you are subscribed to the Google
Groups "Everything List" group.
To unsubscribe from this group and stop receiving emails from it,
send an email to [email protected].
To post to this group, send email to [email protected].
Visit this group at http://groups.google.com/group/everything-list?hl=en
.
For more options, visit https://groups.google.com/groups/opt_out.
--
You received this message because you are subscribed to the Google
Groups "Everything List" group.
To unsubscribe from this group and stop receiving emails from it,
send an email to [email protected].
To post to this group, send email to [email protected].
Visit this group at http://groups.google.com/group/everything-list?hl=en
.
For more options, visit https://groups.google.com/groups/opt_out.
http://iridia.ulb.ac.be/~marchal/
--
You received this message because you are subscribed to the Google Groups
"Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To post to this group, send email to [email protected].
Visit this group at http://groups.google.com/group/everything-list?hl=en.
For more options, visit https://groups.google.com/groups/opt_out.
