On 29 Oct 2015, at 03:11, Jason Resch wrote:
At some level, an algorithm cannot be held responsible for its
actions because it was doing the only thing it could do, what it was
programmed to do. At some point between a simplistic algorithm and a
human level AI, however, we seem able to assigning responsibility/
culpability. What does an algorithm minimally have to have before it
reaches this point?
The ability to learn?
Understanding of the consequences of its actions?
Rights that it cares about?
Personhood?
I am most interested to hear the thoughts of others on this list on
this question.
Of course I will use "my theory", which is the theory of the universal
machine, if we accept the classical, antic definition of knowledge.
To jude a machine responsible, the machine needs a soul. let us accept
that the soul is well approximated by the knower, which is the
conscious or consistent entity obeying the axiom of T, and if
introspective enough, or self-aware, T4. (It has no name and no
represantation available to the machine cognition).
To get that soul, if it is the knower, and if the knower is the one
who has beliefs (axioms, theorems) which can be (luckly or not) true,
then from his we see that it needs a self, by which I mean a
representation of itself, with respect to some universal number (at
least one). That self is, for the PA machine, any piece of codes or
machinery mechanically (in the language terms of the probable
universal machine which run it) capable of self-reference, and for
computers the tricks is in the diagonalization DX= F(XX) => DD =
F(DD). This has been exploited fully by Gödel to Solovay (and
others), and shows that a platonist self obeys to the modal logics G
and G*. It is the 3p self, and later we can explain that although a
machine can have such self, they cannot identity their first "I" with
it. Yet to survive (a teleportation, or an amoeba type of self-
repduction), that is a needed, and in part a result of compromise
between the universal machine that the machine is, and the universal
machine which run it (which eventually will be a more complex object.
That self is what I represent by the [], and a precise example is the
definition of PA in the language of PA (à-la Gödel). It is the 3p
self, on which the machine can bet. I limit myself to platonist
rational believers.
By being platonist, I mean that for all arithmetical propositions A
they believe (A v ~A), By being rational, it means If they believe
that A -> B, and if they believe A, they believe B. Or written in
arithmetic, a scheme of elemebtary truth [](A & (A -> B)) -> []B
(which is equivalent to K [](p->q) -> []p -> []q.
Using Theaetetus, the knower is defined by []p & p, for each
arithmetical statements p (and latter only on the sigma_1 one).
So, if you believe yourself in elementary arithmetic, you dispose of a
good truth level to define more higher level concepts.
A human is more complex than PA; but if the human "brain" is digitally
emulable, there is a level where the self is correct about itself. But
all such notion of self is contradicted by the first person view (that
is why "Who am I?" is a very good koan (the question Ramana Marhasi
ask to meditate upon).
The machine's knower is not a machine, neither from the 3p view, nor
from the 1p view. It can be a sort of 3p object for a more complex
machine view. That is the method: ZF can study the theology of PA, and
then lifit it interrogatively to itself, at its own risk and peril
(like when saying "yes" to a doctor, or more general).
If the machine believes in computationalism, she can always define
itself at the substitution level where there is no responsibility at
all. Like if a lawyer of a murderer would say "My client has only
obeyed to Schoredinger equation, and so has no responsibility at
all", which is not only invalid, but spurious as the member of the
jury get by this the complete freedom to judge the guy responsible and
guilty and deserving the electric chair and then adding: no worry we
obeyed to the Schroedinger equation too, isn't it?
So the one which might be locally partially responsible is at some
higher level, and somehow only the sound, but non nameable 1p knows,
that is, if he consider himself responsible, or having a sense of
regret or remorse.
Universal machine have universal goals, like searching the good. In
collection of interacting universal machine (like with life), the
notion of good automatically conflicts with what other machine can
call good or not, but by abstraction universal machine can extend
their range of self-identification, and get the relative views in an
extended consistent picture, which exists or not.
Eventually, if the machine's soul is guilty or not, is a question to
be answered by the judge invoking its intimate conviction, or by the
members of a jury. Today, no hand made machine respect such a
criterion, and the universal machine are build like slaves, doing
series of disconnected tasks (facebook Mail, type-writter, TV, phone,
and many others App, rarely universal, or when it is the case, like
LateX or TeX, it remains unnoticed, except by the experts in
programming computing.
To have a practical machine, with a soul (and thus a []p & p like PA),
we must give only a programming language and ask the user to start the
conversation, using no apps, but permitting the machine to refer to
itself, with a case that the machine can transform itself, extends
itself (by betting on its consistency at some levels), and this by
incompleteness invite the machine into a transfinitely complex path.
Interfacing a universal machine (which lives in arithmetic) with our
environment in such a way that she can develop responsibility means,
to define a persistent temporal series, and you are the one who should
allow the memorization and reflection on the par of the machine.
Today, except for the clock, few app exploits their self and their
temporal relation with the user.
Now, why would we do that? The computer (real practical) is more an
artificial neo-cortex of ourselves than a medium to let amoeba,
viruses, and other bugs to conquer the space. That would take time
before we can send them in jail or asylum, and for intermediates semi-
zombie, the question will just be hard. For the japanese doll to fight
for they right will take some time.
PA has a soul. But its consciousness is way too much dissociate from
our terrestrial life that, despite its influence (well the influence
of RA first (still more dissociated), and then PA, ZF, I mean in
mathematics and physics), it is not really implementable, except it
is, through our books and brains. The relatively to us "real" machine
will need a physical body, and a much more complex believability
predicate []).
I think that a brain is either a universal machine in front of a
universal machine, or a brain in front of a brain.
I think that each part of such recursive couples of universal machines
specializes respectively in the box []p (the self-representation) and
in the soul/knower []p & p. The knower keeps intact its ombilical
chord with truth, he thinks "semantically", it will not be platonist
(he will not believe in the exclude middle principle, he will not be
computationalist, without some effort and courage and faith. S4Grz
gives indeed, when read epistemologically an intuitionist. It cares on
extending itself, when the 3p scientist (the box) cares on valid
communication/interaction.
In each brain there is a man ([]p) and a woman ([]p & p), and the
woman show something, but the man cannot understand, really. The man,
the representation, the discursive reasoner, is a filter of the
consciousness "flux".
(No need for the "& <>t, here, which is needed for Matter and
"terrestrial consciousness", where again the box is really a filter of
consciousness. It like with the connection of Galois. The empty
theory is satisfied in all models. Then add axioms, you eliminate
models. the more axioms, the more models or possibilities you
eliminaite, and consciousness is on the side of truth and
possibilities/models, that is semantic.
Brent is rught in invoking world, but the point is that no machine can
prove there is a world which satisfy their beliefs once those beliefs
include elementary arithmetic or any other universal programming
languages/machines/number.
If Plotinus is right on some point, and we accept the neoplatonist
idea that EVIL = MATTER, then the fall of the soul is in the passage
[]p & p to []p & p & <>t, as it is a sort of blaspheme without saying
(<>t implicit semantic is that there is a model/reality, it invokes a
reality).
To sum up:
[]p : not responsible
[]p & p : responsible (soul before the fall)
[]p & p & <>t : guilty (soul after the fall: matter/hallucination
generation).
I am sure on nothing, but I use the lexicon I gave between Plotinus
and the universal machine. There are also striking similarity with Lao-
Tseu and even the young Wittgenstein, or Damascius: the idea that
there is something which we cannot talk about (but what are they
talking about).
All those view structure differently the arithmetical truth, and G can
be used to test all the theorems in any of those views or mixing. I
will say more on this some day. Not sure that everyone realized that G
is the gate making all other modal hypostases necessary and possible,
and decidable (at the simple propositional level).
Then with the G/G* difference, the machine can keep track between what
is justifiable/rationally believable, and what is true, but can't be
justified. Except the soul (before the fall), and God, all the other
hypostases of the universal machine split on this. It is a key to
grasp how much the universal machine can grasp about the root of the
mystery, including of her responsibilities.
Bruno
Jason
--
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.
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.
For more options, visit https://groups.google.com/d/optout.
