Hi Bruno,
On Sun, Oct 22, 2017 at 4:47 PM, Bruno Marchal <[email protected]> wrote:
> Hi Telmo,
>
>
>
> On 22 Oct 2017, at 09:58, Telmo Menezes wrote:
>
>> Hola Alberto,
>>
>> On Sun, Oct 22, 2017 at 9:16 AM, Alberto G. Corona <[email protected]>
>> wrote:
>>>
>>> Neural networks are not about artificial intelligence, but about
>>> artificial
>>> intuition. As you said, AlphaGo -a neural network application- can not
>>> answer the question why you did that move?.
>>>
>>> If they could answer, the answer would be ever the same: " I don´t know,
>>> I
>>> moved this because if found some patterns that are very close to this new
>>> one, so I did this move that produced a win at the end within those
>>> patterns".
>>
>>
>> The neural network is used to prune the minimax search tree. AlphaGo
>> can tell you exactly what it hopes to achieve with a given move. What
>> it cannot tell you precisely is why it decided that a certain branch
>> of the tree could be ignored.
>>
>> Under your terminology, one could say that the minimax search tree is
>> the intelligence part, while the neural network plays the role of
>> intuition. This is not so different from what a human player does.
>>
>> (I am basing myself on the original AlphaGo paper, and assuming that
>> nothing fundamental changed in this incarnation)
>>
>>> That does not qualify as intelligence. For me, the appropriate name is
>>> intuition.
>>
>>
>> You are a bit on the side of Chomsky on this -- something that you
>> might not exactly like :)
>> In any case, I can see value in yours and Chomsky's position. Neural
>> networks and statistical learning in general are great, but we should
>> not lose sight of understanding intelligence. However, I would not be
>> surprised if a given mind cannot fully understand the mechanisms
>> underlying itself. Maybe there's a threshold of complexity that must
>> be somewhat below the complexity of the mind itself.
>
>
> The machine can bet correctly on its syntactical, mechanical level, and
> reason correctly with respect to that bet, made in practice with respect to
> some bet on some universal environment. That amkes transhumanism consistent.
Ok.
I suspect that transhumaning is our only hope of surviving our own
creations in the future. On the other hand, maybe it doesn't matter.
Perhaps we are too attached to being homo sapiens. Nothing ever stays
the same. If the future belongs to Jupiter-brain entities, I guess
they are people too...
It is interesting how we are worried about the existence of homo
sapiens in the far future, but we are not bothered by the fact that
they did not exist in the distant past. I find this reminiscent of
fear of "not-existing" after death, but no problem with not having
existed before birth. I think you will agree that it comes from a
misunderstanding of reality.
> But no machine can name or circumscribe its own semantic, that is what
> incompleteness is about.
> Any semantic requires some act of faith on the par of the machine (probably
> in large part instinctive for the animals).
Agreed, I did not forget about Gödel :)
I love the challenge of reverse-engineering how our brain works. There
is some learning algorithm that it runs that we haven't cracked. There
are some clues:
Artificial neural networks use activation and inhibition (equivalent
to glutamate and GABA), but what about the other neurotransmitters?
For example dopamine, which appears to be strongly related to the
reward system -- and thus learning. In our current artificial models,
learning (mainly backpropagation) are still mostly "top-down". They
are something we impose on the network, as opposed to something that
emerges from local network behaviors.
Further, while glutamate and GABA appear to be mostly topological --
they propagate in cascades of neural activation across the network --
dopamine and others seem to work by diffusion. So the topology of the
brain networks is not the only thing that matters, it's spacial layout
also does. This is another layer of complexity. Let's not even get
into gene expression.
Can we figure out how to design such a powerful learning algorithm? It
has been tried for so long with no success. Current AI-hype comes from
enormous computing power and big datasets. There is nothing new about
AlphaGo. Can this wave be ridden all the way to human-level
intelligence (or competence :) or is there something fundamental we
are still missing? I bet on the latter. But then, could it be that the
reason this algo has not be found yet is that its intrinsic complexity
is beyond the grasp human intelligence itself? If you agree with this
idea, do you feel it is related to the Gödelian limit?
>
>>
>> On the other hand: historically, what you call "intuition" has been
>> the hard part...
>
>
> Yes, it is the soul, the knower, the feeler, the "hard" part of the
> mind-body problem, fogetting that the "matter" is as much hard.
>
> But the canonical theology, when understanding that incompletenees make
> provability into a type of rational belief, is clear enough, I think.
>
> p is the truth
> []p can be used as the mind, the ideas, (Plato's Noùs). [] is sigma_1
> complete, so we get all programs and all halting computations, and all
> initial segment of non halting computations.
>
> []p & p gives the non nameable soul. You cannot define it by a predicate
> like bew('p') & True('p') because of Tarski theorem. Scott-Montague)
> generalization shows the same for []p & p. It exists, obeys to S4Grz, but
> the machine cannot know that as such, but can prove it for correct machine's
> in general. That gives an arithmetical interpretation of intuitionist logics
> (by result due to Grzegorczyk, Boolos, Goldblatt).
I feel this is consistent with what I find by introspection --
something that AI researchers perhaps do not do enough. Intuition is
crucial, but impossible to grasp. You cannot use it without also
introducing doubt.
With AlphaGo, it is curious that the heuristic half of the system is
also the one that becomes a black box.
Best,
Telmo.
> Best,
>
>
> Bruno
>
>
>
>
>
>>
>> Telmo.
>>
>>> 2017-10-21 3:46 GMT+02:00 John Clark <[email protected]>:
>>>>
>>>>
>>>> Google reports in the current issue of the journal Nature that it has a
>>>> new greatly improved Go program called "AlphaGo Zero" that is now the
>>>> most
>>>> powerful GO program in the world. And the program isn't good because of
>>>> brute force, it needs to make less than one tenth as many calculations
>>>> as
>>>> the previous best GO program "AlphaGo" that defeated the world's top
>>>> human
>>>> GO player in 2015 4 games out of 5; and yet AlphaGo Zero just defeated
>>>> AlphaGo in a 100 game tournament 100 games to zero.
>>>>
>>>> Even more interesting is how AlphaGo Zero got so smart. The older
>>>> program
>>>> AlphaGo had to start by analyzing hundreds of thousands of championship
>>>> level games made by human players, but AlphaGo Zero started with nothing
>>>> but
>>>> the simple rules of GO and instructions to learn to get better. At first
>>>> the
>>>> program was terrible but day by day it got better and after 40 days of
>>>> thinking about the problem became the best at it in the world. But of
>>>> course
>>>> after 40 days of constant self modification no human being can say how
>>>> AlphaGo Zero works.
>>>>
>>>> https://www.nature.com/nature/journal/v550/n7676/full/nature24270.html
>>>>
>>>> It seems to me the next logical step would be to switch the program's
>>>> interest from getting better at the game of GO to improving computer
>>>> code,
>>>> including its own. I wonder where that could lead.
>>>>
>>>> John K Clark
>>>>
>>>> --
>>>> 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 https://groups.google.com/group/everything-list.
>>>> For more options, visit https://groups.google.com/d/optout.
>>>
>>>
>>>
>>>
>>>
>>> --
>>> Alberto.
>>>
>>> --
>>> 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 https://groups.google.com/group/everything-list.
>>> For more options, visit https://groups.google.com/d/optout.
>>
>>
>> --
>> 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 https://groups.google.com/group/everything-list.
>> For more options, visit https://groups.google.com/d/optout.
>
>
> 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 https://groups.google.com/group/everything-list.
> For more options, visit https://groups.google.com/d/optout.
--
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 https://groups.google.com/group/everything-list.
For more options, visit https://groups.google.com/d/optout.
