On 11/28/2014 9:55 AM, Bruno Marchal wrote:
On 28 Nov 2014, at 04:27, John Clark wrote:
On Wed, Nov 26, 2014 at 11:30 PM, meekerdb <[email protected]
<mailto:[email protected]>> wrote:
> More likely we will make an AI that is intelligent, is not conscious like
a human
with an inner narrative but is conscious in some other way which will be
very
difficult for us to recognize.
With the obvious exception of our own, consciousness is not very difficult to
recognize, it is IMPOSSIBLE to recognize, and it doesn't matter if it's another human
or a computer. All we can recognize is intelligent behavior and then try to make a
conclusion from that observation using one of the enumerable theories about
consciousness that are available. And all the many consciousness theories are
different from each other and all of them work about equally well (or badly).
Some have testable consequences. You don't need to understand the step 3, to understand
the math parts, and the testability of the theory extracted. In the "Brussel's thesis" I
present the UDP (the universal dovetailer *paradox*- only to provide the motivation for
the definition in the math part). (I was warned against philosophers allergic to thought
experiments!). In Lille, that was not a problem, as there were no literary philosopher
in the jury.
And having no facts that must be fitted to theory is why the profession of
consciousness theoretician is so incredibly easy and why they are so common on the
internet. However it's hard as hell to find a good intelligence theory because it must
be compatible with a astronomical number of very diverse facts, so it's not surprising
that intelligence theoreticians are very rare on the internet. Consciousness is easy
but intelligence is hard.
I think the contrary. Intelligence is basically very simple. It requires no more than a
universal machine. Well, at least two, to be exact. A couple <child, mother>, but in an
abstract sense, in which the mother can be the arithmetical truth/reality (the structure
(N,0,+,*)), and the child any universal programs defined and executed, relatively to
oracles or to other universal programs.
I don't think that's enough. Intelligence must be more than the mere possibility of
learning, it must include the motivation to learn and to act. Without a value system
there is no reason to learn and all knowledge is equally valuable - like the "knowledge"
obtained by video camera.
Brent
Consciousness is perhaps not that difficult. Once you accept it is invariant for some
recursive permutation at some description level, then the incompleteness theorem
explains it, I think. We got an explanation of why self-consistency *appears* obvious to
us, but is incommunicable/non-justifiable, unless we invoke the notion of truth, which
is explained to be non-expressible, and makes consciousness non definable. The math
provides a general theory of qualia, and it is testable because the quanta appears to be
the first person plural sharable of it.
The math shows in particular that a reflexive and symmetrical structure appears at the
place it should appear in case QM is correct.
What *is* really difficult is the competence things, and its development, as they grow
on many incomparable lattices of incomparable abilities. There are many results in
theoretical computer science, initiated with the work of Putnam, Blum, Gold, Case &
Smith, Royer, Oherson, ... and many others. There are many beautiful theorems, like the
theorem of Blum and Blum which answers the question about what can be immeasurably more
competent than a machine? The answer is: a couple of machines! There also equivalent
speed-up theorem for the inference inductive abilities, and in general the more errors
machine can do, the more competent they can become. There are no universal learning
machine (universal competence) unless you accept machine doing unbounded number of
errors, and with the ability to change their mind infinitely often, even for equivalent
theories (that is they change their mind without experimental reason).
Those are theories of intelligence, in a sense, as they are metatheories on learning
strategies (competence development). It studies the effect of intelligence of the many
possible grows of competence. It is necessarily a non constructive theory of
intelligence, but the contrary would have been astonishing, and ring false with the
absolute non definability of the first person by the first person. Intelligence is not
programmable, except trivially by a meta-program like "Add and multiply and help
yourself". And hope, for enough memories notably.
You can find references by searching in the biblio of the Lille thesis and the
bibliography of the french long text.
Bruno
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]
<mailto:[email protected]>.
To post to this group, send email to [email protected]
<mailto:[email protected]>.
Visit this group at http://groups.google.com/group/everything-list.
For more options, visit https://groups.google.com/d/optout.
http://iridia.ulb.ac.be/~marchal/ <http://iridia.ulb.ac.be/%7Emarchal/>
--
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]
<mailto:[email protected]>.
To post to this group, send email to [email protected]
<mailto:[email protected]>.
Visit this group at http://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 http://groups.google.com/group/everything-list.
For more options, visit https://groups.google.com/d/optout.
