IBM simulates cat brain
Mike Ash
http://arstechnica.com/science/news/2009/11/ibm-makes-supercomputer-signi
ficantly-smarter-than-cat.ars?utm_source=rss&utm_medium=rss&utm_campaign=
rss
Or, in short form:
In summary, IBM has built a simulator on a supercomputer that's capable
of simulating a cerebral cortex more complex than that of a cat.
This, obviously, raises some interesting questions.
First, how realistic is it? How closely do these simulated neurons mimic
the real thing? How close do the structures match that of the real
thing? I haven't read the paper and haven't seen this discussed in the
summaries. They seem to believe that it's a pretty useful thing, so that
seems to indicate that it's reasonably close to reality.
This of course naturally leads into a more SFnal question: is it
conscious? I'd certainly say that my cat is conscious, and this thing is
supposed to be about as smart. It's interesting that the summary article
I linked to says that an advantage of the simulated brain is that you
can experiment on it without the ethical questions that come with
experimenting on real brains, but that's not really true at all, if you
make the leap to assuming that simulated brains can also be conscious.
The big thing I took out of all of this relates to stronger forms of AI,
though. It's always been on the horizon, with vague future-y dates. And
of course we still have no idea how to build all the complicated stuff
that an AI would require. But the trump card in that area has always
been the idea that, in the worst case, you can just simulate a human
brain and you have AI.
Well, they claim that this simulates about 4.5% of a human cerebral
cortex, and at 0.1% to 1% of real time. And they claim that the numbers
scale linearly with memory and computing power.
4.5% is a factor of about 22, roughly 4.5 doublings. Assuming Moore's
Law keeps giving us 18-month doublings, that's only about another 7
years before a full human brain can be simulated. Then another 10-15
years after that before it can be simulated in real time.
And all of this, of course, assumes that the software side of brain
simulation never gets any faster.
--
Mike Ash
Radio Free Earth
Broadcasting from our climate-controlled studios deep inside the Moon
Juho Julkunen
Mike Ash (mi...@mikeash.com) says...
> I'm surprised this hasn't popped up in here yet:
>
> http://arstechnica.com/science/news/2009/11/ibm-makes-supercomputer-signi
> ficantly-smarter-than-cat.ars?utm_source=rss&utm_medium=rss&utm_campaign=
> rss
>
> Or, in short form:
>
> http://tinyurl.com/ycp5xzf
>
> In summary, IBM has built a simulator on a supercomputer that's capable
> of simulating a cerebral cortex more complex than that of a cat.
Oh, cool. I'd only heard about this:
http://en.wikipedia.org/wiki/Blue_Brain_Project
--
Juho Julkunen
Mike Williams
>Oh, cool. I'd only heard about this:
>
>http://en.wikipedia.org/wiki/Blue_Brain_Project
For which there's a nice presentation at TED:
http://www.ted.com/talks/lang/eng/henry_markram_supercomputing_the_brain_
s_secrets.html
--
Mike Williams
Gentleman of Leisure
Dorothy J Heydt
I read the subject line off to Hal, who said, "Doesn't cooperate
at all?"
My immediate reaction was, "But can it catch mice?"
--
Dorothy J. Heydt
Vallejo, California
djheydt at hotmail dot com
Should you wish to email me, you'd better use the hotmail edress.
Kithrup is getting too damn much spam, even with the sysop's filters.
James Burns
> I'm surprised this hasn't popped up in here yet:
>
> http://arstechnica.com/science/news/2009/11/ibm-makes-supercomputer-signi
> ficantly-smarter-than-cat.ars?utm_source=rss&utm_medium=rss&utm_campaign=
> rss
>
> Or, in short form:
>
> http://tinyurl.com/ycp5xzf
>
> In summary, IBM has built a simulator on a supercomputer
> that's capable of simulating a cerebral cortex more complex
> than that of a cat.
>
> This, obviously, raises some interesting questions.
>
> First, how realistic is it? How closely do these simulated
> neurons mimic the real thing? How close do the structures
> match that of the real thing? I haven't read the paper and
> haven't seen this discussed in the summaries. They seem to
> believe that it's a pretty useful thing, so that seems to
> indicate that it's reasonably close to reality.
>
> This of course naturally leads into a more SFnal question:
> is it conscious? I'd certainly say that my cat is conscious,
> and this thing is supposed to be about as smart.
Now, that is an interesting question. I had seen an article about
that, but it didn't occur to me to wonder about what we have
already accomplished.
I would probably end up disagreeing about your cat being
conscious (as I mean "conscious", at least) if I knew your cat,
though I completely believe that some adjective a little weaker
does apply. "Sentient", maybe?
If your cat -- or a computer -- was able to model its own
workings, however imperfectly, in interaction with other
minds, then I might agree that we're getting there. Cats
are smart, but I'm not sure that they're that smart.
And I have a suspicion that, for heavy-duty mind-modeling,
you need to have language. At least, the uncontroversial
examples of conscious beings all have language.
However, I think it is fair to say that a cat has a point
of view. Does this mean that we will have a computer soon
with a "point of view"? I think the ethical questions
you ask come into play anyway.
I don't know if it's a good thing or a bad thing (ethically),
but these whole-brain simulations (cat, ape, human, super-genius?)
will first run a couple orders of magnitude slower than
real time. That means the brain's environment has to be
a simulated one as well, which I suspect means it will
be very impoverished in terms of STimulation. Are we
torturing the cat (ape, human, super genius) just by
Simulating it?
Perhaps the ethical brain-simulater needs to create an
actual, /physical/
environment that the simulated brain can interact with, even at
1-to-100 speeds. A play pen where things will fall down
and go boom at 100 times the acceleration the being evolved
to cope with, but safe nonetheless -- and /real/.
(Maybe a robot in a swimming pool to slow things down?)
That probably
means that the simulated brain will grow into a different
sort of brain than an in-body brain, but, if I'm right
about the sensory deprivation, it will be /less/ different
than the sad, mad creature created without the play pen.
> It's interesting that the summary article
> I linked to says that an advantage of the simulated brain
> is that you can experiment on it without the ethical
> questions that come with experimenting on real brains,
> but that's not really true at all, if you make the leap
> to assuming that simulated brains can also be conscious.
This is an excellent point, one which I don't think
can be dodged by claiming that the simulation isn't
conscious, only "conscious". The question is not
what is it like to be a computer; it is what is it
like to be us.
There may be a need for an analogue to the various
Societies for the Prevention of Cruelty to Animals.
I can see the Vatican getting involved in a big way,
once they get the theology sorted out.
> The big thing I took out of all of this relates to
> stronger forms of AI, though. It's always been on the
> horizon, with vague future-y dates. And of course we still
> have no idea how to build all the complicated stuff
> that an AI would require. But the trump card in that area
> has always been the idea that, in the worst case, you can
> just simulate a human brain and you have AI.
>
> Well, they claim that this simulates about 4.5% of a human
> cerebral cortex, and at 0.1% to 1% of real time. And
> they claim that the numbers scale linearly with memory
> and computing power.
(Plus, processing time scales linearly also.)
Yes! This is the really amazing thing. If what they
claim is true, this design has basically smashed
the AI problem. All that is left is getting enough
resources to finish the job.
Jim Burns
> 4.5% is a factor of about 22, roughly 4.5 doublings.
> Assuming Moore's Law keeps giving us 18-month doublings,
> that's only about another 7 years before a full human
> brain can be simulated. Then another 10-15 years after
> that before it can be simulated in real time.
>
> And all of this, of course, assumes that the software
> side of brain simulation never gets any faster.
"Soothly we live in mighty years!"
-- Poul Anderson
David Mitchell
> Mike Ash wrote:
>>
>> In summary, IBM has built a simulator on a supercomputer that's capable
>> of simulating a cerebral cortex more complex than that of a cat.
Very interesting, thank you.
> I would probably end up disagreeing about your cat being conscious (as I
> mean "conscious", at least) if I knew your cat, though I completely
> believe that some adjective a little weaker does apply. "Sentient",
> maybe?
>
> If your cat -- or a computer -- was able to model its own workings,
> however imperfectly, in interaction with other minds, then I might agree
> that we're getting there. Cats are smart, but I'm not sure that they're
> that smart.
I don't know about internal models, but our cats were certainly capable
of using tools (and I _was_ that tool* :-).
They were also capable of linking stimulus-response pairs together to
achieve more complex goals.
For example, on one occasion, our smartest cat, was hungry.
The goal here is for me to feed him, the problem is that I'm in my
office, working.
To achieve his goal, he has to do two things: get my attention, and show
me his empty bowl, at which point I will dutifully fill it.
So, knowing that I can't resist it, he lies on his back, with tummy
exposed, which invariably leads to my tickling it.
However, I was working and didn't notice, so he mewed, pointedly (really
- he has a special mew for "pay attention to the cat". He also sighs,
when I get it wrong. Which is often: I am clearly a terrible
disappointment to him).
It works, I get up, at which point he springs up, walks to door, looks
back at me. I follow down the corridor, with him checking periodically
that I'm following.
At the end of the corridor is the landing - he stops, looks pointedly at
empty bowl, and back at me.
I feed him.
It may not sound that clever, but one of the ways that people disparage
"intelligence" in animals is to say that they are only capable of simple
stimulus-response types of behaviour.
In this case, he, it seems to me, deliberately planned ahead, linking s-r
pairs together to achieve a more complex goal.
I would say that this is definitely intelligent behaviour.
The other cat would do similar things - one of his favourite spots to
sleep in was my office chair (when it was pre-warmed, of course), to
achieve the goal of an empty, warm, chair he would also lie with his
tummy out by the office door, until I would get up, to tickle him, at
which point he would spring to his feet, and make a bee-line for the
chair.
To me, that's a two-step s-r pairing.
> And I have a suspicion that, for heavy-duty mind-modeling, you need to
> have language. At least, the uncontroversial examples of conscious
> beings all have language.
What about deaf-mutes? Helen Keller believes she was conscious even
before she learned to sign.
> I don't know if it's a good thing or a bad thing (ethically), but these
> whole-brain simulations (cat, ape, human, super-genius?) will first run
> a couple orders of magnitude slower than real time. That means the
> brain's environment has to be a simulated one as well, which I suspect
> means it will be very impoverished in terms of STimulation. Are we
> torturing the cat (ape, human, super genius) just by Simulating it?
I would say yes, most animals show stress behaviour if you sensorially
deprive them - I think we'd need a plausible environment for our
simulations to live in, or they'd either not function properly, or at all.
> Perhaps the ethical brain-simulater needs to create an actual,
> /physical/
> environment that the simulated brain can interact with, even at 1-to-100
> speeds. A play pen where things will fall down and go boom at 100 times
> the acceleration the being evolved to cope with, but safe nonetheless --
> and /real/. (Maybe a robot in a swimming pool to slow things down?)
I don't think it need be real - just realistic.
>> It's interesting that the summary article I linked to says that an
>> advantage of the simulated brain is that you can experiment on it
>> without the ethical questions that come with experimenting on real
>> brains, but that's not really true at all, if you make the leap to
>> assuming that simulated brains can also be conscious.
Absolutely not. I'm perhaps more willing to see consciousness where
others might not, but if it's conscious enough to be useful, it's
conscious enough to have rights, IMO.
There's a philosophy called "painism" (See http://en.wikipedia.org/wiki/
Richard_D._Ryder) which I think has a great deal of validity: it combines
the utilitarian view that moral status comes from the ability to feel
pain with the rights prohibition on using others as a means to our ends.
If the simulation is conscious enough to react to, and attempt to avoid,
pain, I think we should treat that as sufficient justification to avoid
inflicting it.
After all, we could, perhaps, be "nothing but" simulations of our own
brains ourselves, one day. The idea that others might be free to perform
experiments on us ad-lib horrifies me.
--
=======================================================================
= David --- [*]: Insert obvious joke here.
= Mitchell ---
=======================================================================
Justin Fang
Dorothy J Heydt <djh...@kithrup.com> wrote:
>In article <mike-602F47.0...@news.eternal-september.org>,
>Mike Ash <mi...@mikeash.com> wrote:
>>In summary, IBM has built a simulator on a supercomputer that's capable
>>of simulating a cerebral cortex more complex than that of a cat.
>>
>>This, obviously, raises some interesting questions.
>>
>>First, how realistic is it? How closely do these simulated neurons mimic
>>the real thing? How close do the structures match that of the real
>>thing? I haven't read the paper and haven't seen this discussed in the
>>summaries. They seem to believe that it's a pretty useful thing, so that
>>seems to indicate that it's reasonably close to reality.
>I read the subject line off to Hal, who said, "Doesn't cooperate
>at all?"
>
>My immediate reaction was, "But can it catch mice?"
From the FAQs here:
http://p9.hostingprod.com/@modha.org/blog/2009/11/post_3.html
"How close is the model to producing high level cognitive function?
Please note that the rat (-scale simulation) does not sniff cheese, and
the cat (-scale simulation) does not chase the rat. :) Up to this
point, our efforts have primarily focused on developing the simulator as a
tool of scientific discovery that incorporates many neuroscientific
details to produce large-scale thalamocortical simulations as a means of
studying behavior and dynamics within the brain. While diligent
researchers have made tremendous strides in improving our understanding of
the brain over the past 100 years, *neuroscience has not yet reached the
point where it can provide us with a recipe of how to wire up a cognitive
system.* Our hope is that by incorporating many of the ingredients that
neuroscientists think may be important to cognition in the brain, such as
a general statistical connectivity pattern and plastic synapses, we may be
able to use the model as a tool to help understand how the brain produces
cognition."
Emphasis added. So, while they may have the raw computing power, they
don't have enough information about brains to actually model a cat (much
less a human). I'm guessing that this will continute to be the hard part
of the problem even if the raw computing power part is solved.
--
Justin Fang (jus...@panix.com)
Wayne Throop
: [... cat overcomes adversity to achieve dinnertime ...]
: It may not sound that clever, but one of the ways that people disparage
: stimulus-response types of behaviour.
Some time ago, one of our cats became facinated with strings. Sometimes
we wouldn't dangle strings to his satsifaction, so he eventually took up
playing with them himself; a take a string in mouth, and run like mad
among obstacles, make patterns. So far, very stimulus/response like.
But he liked *chasing* strings. Hard to do that if you're the one
pulling it. So, he discovered that if he found a string of the
appropriate length, he could run around an object trying to bat at
the far end of the string as it fled before him in terror.
Still not horribly clever, perhaps. But he also noted that some strings
are too short, but more interestingly, some are too long. So he'd guage
the distance around the object-du-jour, measure off a length of string,
and take *that* *point* of the too-long string in his mouth, so that
the string tip remained at the correct tantalizing distance when chasing
'round said object du said jour.
Just a complicated way of tailchasing, I suppose, but I was impressed.
Of course, he was also adept at people-wrangling, but I hand't
really encountered a cat quite so adept at string-wrangling before.
Wayne Throop thr...@sheol.org http://sheol.org/throopw
Robert Martinu
> This of course naturally leads into a more SFnal question: is it
> conscious? I'd certainly say that my cat is conscious, and this thing is
> supposed to be about as smart. It's interesting that the summary article
> I linked to says that an advantage of the simulated brain is that you
> can experiment on it without the ethical questions that come with
> experimenting on real brains, but that's not really true at all, if you
> make the leap to assuming that simulated brains can also be conscious.
Speaking of ethics, with a simulated brain you also gain the option of
doing backups and restoring the consciousness to any point in its existence.
Is it bad to lets say torture it until it snaps just because you can
push the undo-button?
And then there is the question of indentity, have a mind put in various
bodies simultaniously and the question "who am I" gets a whole new meaning.
OTOH how would one define death in such cases, would being tortured/
murdered be actually less severe then a bad dream for us?
We'd have to view such a mind not as a single entity, but more in a
multiverse way of thinking.
Robert
Ryan McCoskrie
In Reynold's /The Prefect/ a back of a dead man's mind denied being anything
more than a clever mimic on the basis that that was what the living version
thought.
--
Quote of the login:
A computer scientist is someone who fixes things that aren't broken.
Ryan McCoskrie
Personally I suspect that a perfect, digital simulation of a cats mind
would immediately turn into crude simulation of a dogs mind just to spite
its creators.
David Mitchell
> : David Mitchell <david.robo...@googlemail.com> : [... cat
> overcomes adversity to achieve dinnertime ...] : It may not sound that
> clever, but one of the ways that people disparage : "intelligence" in
> animals is to say that they are only capable of simple :
> stimulus-response types of behaviour.
>
>
> Of course, he was also adept at people-wrangling, but I hand't really
> encountered a cat quite so adept at string-wrangling before.
Nice story.
There's a quote somewhere to the effect that anyone who says that cats
aren't intelligent has never have lived with one - I would certainly
never have attributed cats the personalities and intelligence that I do
now, before I had two.
I _do_ find it telling that whenever we find animals displaying behaviour
which we previously would have categorised as intelligent, we immediately
disparage that behaviour, and move the goalposts.
For instance, use of language was always the canonical sign of
intelligence until chimps were found using it (after appropriate
training, of course), then it became irrelevant, and "tool use" was the
only valid indicator, until we discovered that certain animals (notably
crows) seem to use primitive tools, at which point we decided, _that_ was
irrelevant too.
We're just not willing to lose our special status, and open the huge can
of worms that would result.
--
=======================================================================
= David --- If you use Microsoft products, you will, inevitably, get
= Mitchell --- viruses, so please don't add me to your address book.
=======================================================================
Erik Max Francis
> There's a quote somewhere to the effect that anyone who says that cats
> aren't intelligent has never have lived with one - I would certainly
> never have attributed cats the personalities and intelligence that I do
> now, before I had two.
I've had a few friends over who were surprised to see my cats show
emotion (in this case, contentment as I was holding them petting one who
had a dumb smile on his face and was clearly happy). I've heard it from
others, too (who have never had pets or been around animals for extended
periods, at a farm for example). It's a rather funny conclusion,
because if you think about it, emotions are the _primitive_ things that
our cerebral cortex is built onto and around and to interact with. So
of _course_ higher animals have emotions.
> I _do_ find it telling that whenever we find animals displaying behaviour
> which we previously would have categorised as intelligent, we immediately
> disparage that behaviour, and move the goalposts.
>
> For instance, use of language was always the canonical sign of
> intelligence until chimps were found using it (after appropriate
> training, of course), then it became irrelevant, and "tool use" was the
> only valid indicator, until we discovered that certain animals (notably
> crows) seem to use primitive tools, at which point we decided, _that_ was
> irrelevant too.
>
> We're just not willing to lose our special status, and open the huge can
> of worms that would result.
You make it sound defensive, but I just see it as the desire to find out
what makes us different from the rest of the animals. There clearly
_is_ a qualitative difference between human thinking and the thinking of
other higher mammals. We after all, can do calculus, construct
buildings, and build civilizations. They can't, and so there's
something different between them and us. The trick is to find out what
it is.
When we think it's _this_ set of things, then we find examples of some
of those behaviors in other animals, we knock those out. Now it's _this
other_ set of things, and then we iterate like that.
What you see as defensiveness can simply be an attempt at understanding.
--
Erik Max Francis && m...@alcyone.com && http://www.alcyone.com/max/
San Jose, CA, USA && 37 18 N 121 57 W && AIM/Y!M/Skype erikmaxfrancis
Nobody's interested in sweetness and light.
-- Hedda Hopper
David Mitchell
> You make it sound defensive, but I just see it as the desire to find out
> what makes us different from the rest of the animals. There clearly
> _is_ a qualitative difference between human thinking and the thinking of
> other higher mammals. We after all, can do calculus, construct
> buildings, and build civilizations. They can't, and so there's
> something different between them and us. The trick is to find out what
> it is.
>
> When we think it's _this_ set of things, then we find examples of some
> of those behaviors in other animals, we knock those out. Now it's _this
> other_ set of things, and then we iterate like that.
>
> What you see as defensiveness can simply be an attempt at understanding.
All true, and a good point. Perhaps I'm just too cynical.
Tim Little
> There clearly _is_ a qualitative difference between human thinking
> and the thinking of other higher mammals. We after all, can do
> calculus, construct buildings, and build civilizations. They can't,
> and so there's something different between them and us.
I don't think there is much difference, on an individual level. I
suspect that we just happened to cross an individually insignificant
threshold that led to global changes in capability and behaviour. I
suspect that groups of isolated human babies, subtly supported in
their physical survival but without any cultural contact from other
human beings, would develop societies no more advanced than that of
many other species.
Perhaps the rise of our technological society is better viewed as
intelligence acting as a generational "gain factor". With a "native"
0.9 gain factor, under ideal circumstances a society would be capable
of passing on 90% of their own acquired knowledge and techniques to
the next generation. Innovations (e.g. hunting methods) would last up
to a few dozen generations and are lost again, or displace other
useful things.
Perhaps we have a "native" 1.01, and under near-ideal circumstances
during our prehistory our ancestors were able to impart enough to the
next generation that knowledge could start to actually accumulate.
Most early societies no doubt encountered less-than-ideal
circumstances and their actual gain dropped below 1.
At some point those accumulated techniques included more advanced
concepts in speech and ways to teach those concepts. This widened the
range of knowledge that could be passed on, and raised the gain factor
above the "native" level. Meanwhile, the other techniques allowed the
societies employing them to adapt their environment and increase the
chance of conditions being right for knowledge to be retained rather
than lost and having to be rediscovered.
I would not be surprised if some other animals are so close behind us
that their gain factor would be 0.9. Perhaps some are even innately
ahead of us in raw intellectual potential but have had no need to
develop it - dolphins, maybe? With the right culture it might be
possible for even animals slightly below 1 in native gain to maintain
a civilization, even if one wouldn't naturally develop. The important
thing is being able to train at least some of the next generation to
be able to use the previous generation's knowledge, and perhaps
painstakingly acquired techniques could make up for lack of native
ability if the shortfall is small enough.
- Tim
Erik Max Francis
> On 2009-11-20, Erik Max Francis <m...@alcyone.com> wrote:
>> There clearly _is_ a qualitative difference between human thinking
>> and the thinking of other higher mammals. We after all, can do
>> calculus, construct buildings, and build civilizations. They can't,
>> and so there's something different between them and us.
>
> I don't think there is much difference, on an individual level. I
> suspect that we just happened to cross an individually insignificant
> threshold that led to global changes in capability and behaviour. I
> suspect that groups of isolated human babies, subtly supported in
> their physical survival but without any cultural contact from other
> human beings, would develop societies no more advanced than that of
> many other species.
> I would not be surprised if some other animals are so close behind us
> that their gain factor would be 0.9. Perhaps some are even innately
> ahead of us in raw intellectual potential but have had no need to
> develop it - dolphins, maybe? With the right culture it might be
> possible for even animals slightly below 1 in native gain to maintain
> a civilization, even if one wouldn't naturally develop. The important
> thing is being able to train at least some of the next generation to
> be able to use the previous generation's knowledge, and perhaps
> painstakingly acquired techniques could make up for lack of native
> ability if the shortfall is small enough.
I think it's clear that this "gain factor" is important for the ability
to develop deep knowledge and a civilization, but it still leaves out
one obvious point. The problem with this theory is that you should
still be able to teach individuals of these other species the knowledge
_we've_ acquired. I recall the horses that could supposedly add, but I
believe those were all shown to be carnival tricks.
There's some limited and meaningful communication with apes, apparently,
and it even can be somewhat symbolic at times. Anyone taught any of
them arithmetic, and then moved onto algebra? Will it ever even be
possible? It sure doesn't seem like it, given all the attempts that
have been tried, and that certainly is something objectively different
that's not merely a difference in generational gain factor.
In other words, you can't teach an ape calculus. We have the gain
factor sufficient for having acquired that knowledge, so it doesn't
matter if your attempt is to _teach_ it to the animal. If it's
impossible to do, then there's something going on more than this
generational factor, which I think is pretty clearly the case. If the
brightest individuals of their species can't learn that, then it doesn't
matter what their species' "gain factor" is. We still don't have a good
idea _what's_ different, but something is.
--
Erik Max Francis && m...@alcyone.com && http://www.alcyone.com/max/
San Jose, CA, USA && 37 18 N 121 57 W && AIM/Y!M/Skype erikmaxfrancis
The mind is not a vessel to be filled but a fire to be kindled.
-- Plutarch
Tim Little
> I think it's clear that this "gain factor" is important for the ability
> to develop deep knowledge and a civilization, but it still leaves out
> one obvious point. The problem with this theory is that you should
> still be able to teach individuals of these other species the knowledge
> _we've_ acquired.
In theory, I suppose so. In practice, our hard-won repertoire of
teaching methods are all adapted specifically to us. For example, it
has been shown that "baby talk" tones are fairly cross-cultural, and
they appear better suited for teaching (human) babies their native
language(s) than normal talk.
Is there a different form of "baby talk" that would suit chimpanzees?
We don't know; we wouldn't have found it yet if there were. What's
more, all our nonverbal signals are adapted for other humans. On top
of that, human children are completely immersed within a culture that
expects them to be fluent in its language, with practice and
correction almost every waking minute from a wide variety of peers and
adults.
I think the fact that any nonhumans can be taught anything at all by
us despite these barriers is evidence that they have a lot more
potential that we cannot yet actualized.
> In other words, you can't teach an ape calculus.
We don't know that yet. All we know is that we cannot employ
human-centric teaching methods to apes, or the variations on it that
we have yet tried. I also point out historically recent events in
which certain subgroups of *humans* have been believed incapable of
learning a great many things much less complex than calculus.
> We have the gain factor sufficient for having acquired that
> knowledge, so it doesn't matter if your attempt is to _teach_ it to
> the animal.
Developing techniques to pass on knowledge in a manner that can be
learned by the recipients is itself a part of that knowledge, indeed a
prequisite for passing much else. And all those hard-won techniques
were designed around the central assumption that the student is human.
- Tim
Ilmari Karonen
>
> Perhaps the ethical brain-simulater needs to create an
> actual, /physical/
> environment that the simulated brain can interact with, even at
> 1-to-100 speeds. A play pen where things will fall down
> and go boom at 100 times the acceleration the being evolved
> to cope with, but safe nonetheless -- and /real/.
Fortunately, we're already quite good at creating "virtual play pens",
complete with rather good (though inevitably coarse-grained) physics
simulation. And these play pens don't need supercomputers to work --
they run in real time on commodity hardware. You've probably used
some yourself, if you've ever played a 3D computer game or used
"virtual reality" programs such as Second Life.
Researchers being the kind of practical fellows they are, they'll most
likely just grab some existing VR code off the shelf and adapt it to
their needs. So perhaps the first simulated intelligence will grow up
in the world of Garry's Mod.
--
Ilmari Karonen
To reply by e-mail, please replace ".invalid" with ".net" in address.
Eivind
> Perhaps the rise of our technological society is better viewed as
> intelligence acting as a generational "gain factor". With a "native"
> 0.9 gain factor, under ideal circumstances a society would be capable
> of passing on 90% of their own acquired knowledge and techniques to
> the next generation. Innovations (e.g. hunting methods) would last up
> to a few dozen generations and are lost again, or displace other
> useful things.
I think it goes deeper than that, actually. To progress at all, we can
agree that in some sense or other, the next generation must, on the
average, be better off than the former. For this, some sort of transfer
is nessecary.
But you can extend the baseline. I would argue that DNA and, to a lesser
degree, sexual reproduction, where major steps in carrying information
forward effectively.
But "learning" by natural selection and inheritence trough DNA is a very
slow process; even something as simple as "doing X has a 10% chance of
killing you on the spot" will tend to take at the very least dozens of
generations to sink in.
Language, is radically quicker for transfering information, and many
animals have it, I'm defining language widely here, the bee-dance that
tells which direction to fly, is language. But there's a problem with it
too: most forms of language aren't stored, thus there's a risk that much
of the information learnt will be forgotten.
Writing, as in any form of reliable *storage* of a message, fixes that
flaw. Not as in "perfectly fixed" but as in improves the situation
considerably.
Each of these methods is orders of magnitude faster than the previous.
Human beings haven't -stopped- developing at the DNA-level by natural
selection, but the speed is ignorable next to the changes caused by our
systematic gathering of knowledge in writing.
Some argue that making a human-level AI is a logical next step, and that
doing so would allow us to accelerate our learning even more.
On the flipside, the unsolved problems become harder and harder, much of
the low-hanging fruit will be picked already. Being 10 times more
capable of learning, doesn't help much if the unsolved problems are 100
times more difficult.
In short, I think all species are learning, just to different degrees.
As to why you can't teach a chimpanzee algebra, I personally find it
likely that there's strong interactions between intelligence and
language, and that there's tresholds; being smarter is, other things
being equal, probably always an advantage, but it's enormously much MORE
so if your language is advanced enough that the conclusions you draw,
using your smart brain, can be imparted to your kids and tribe.
So I find it likely that chimpanzees aren't -quite- smart enough that
what they'd have to say would give a huge advantage to the better
language-users, and also not -quite- good enough language-users that
they can effectively communicate the advantages of being smarter.
Other things aren't equal offcourse, a brain is an expensive organ to
supply with nutrients, I would think it depends on the environment if
it's a net advantage or a net disadvantage to be somewhat smarter, but
also to consume somewhat more food.
Eivind Kj�rstad
Eivind
> In theory, I suppose so. In practice, our hard-won repertoire of
> teaching methods are all adapted specifically to us.
True. But if there where another way of teaching, that would enable a
chimpanzee-kid to learn at a rate comparable to that of a human kid, or
even just a lot better than the average chimpanzee-kid, there'd be
strong evolutionary pressure for chimpanzes to discover the trick --
even if it has to happen trough accident and natural selection.
I.e. *we* found a way to teach our kids calculus, the chimpanzees did not.
I personally find it a lot more plausible that chimpanzes just aren't
very good at calculus, as opposed to: they -are- good at it, but they're
lacking the right way of learning it, adapted to a chimpanzee. (what
would be the evolutionary advantage of being good at something you'll
never have the chance to learn?)
> Is there a different form of "baby talk" that would suit chimpanzees?
If there was, wouldn't the chimpanzes have discovered it ? For the same
reason human parents discovered human baby-talk ? I'm skeptical of "a
lot of unused potential" claims, most potential is expensive, and if
it's generally unused, I'd tend to think evolution would get rid of it
over time, trough randomness. i.e. a trait needs to be selected for
positively to some degree to not disappear over time.
> And all those hard-won techniques
> were designed around the central assumption that the student is human.
No. Not true at all. Humans have spent millenia consciously and
systematically trying out and perfecting methods for teaching several
different types of animals. We've spent a -lot- of time figuring out how
to efficiently train horses and dogs, for example, and we've done so
using modern methods like language, writing and systematic science.
Yeah, it's in -principle- possible there's some much superior method,
and we just ain't stumbled across it yet, perhaps because we ourselves
are not horses and dogs, but I don't find it particularily likely.
Eivind Kj�rstad
Tim Little
> True. But if there where another way of teaching, that would enable a
> chimpanzee-kid to learn at a rate comparable to that of a human kid, or
> even just a lot better than the average chimpanzee-kid, there'd be
> strong evolutionary pressure for chimpanzes to discover the trick --
> even if it has to happen trough accident and natural selection.
As you pointed out in a previous post, accident and natural selection
is *slow*. We got there first, and no other species has had time yet
to catch up, regardless of whether they are capable of doing so or
not. Note also that there isn't much evolutionary pressure to learn
rapidly until there is so much worthwhile to be taught that the time
it takes is a constraint on survival. It's chicken-and-egg.
> I personally find it a lot more plausible that chimpanzes just
> aren't very good at calculus, as opposed to: they -are- good at it,
> but they're lacking the right way of learning it, adapted to a
> chimpanzee.
I'm not arguing that they're good at it, just that they (or some other
comparatively smart animal) might be barely capable of learning it, if
only we had an optimum way of educating them. As far as that actual
example goes, I do suspect that they are not capable of it, as it
seems to be beyond the capability of a significant portion of humans.
>> Is there a different form of "baby talk" that would suit chimpanzees?
>
> If there was, wouldn't the chimpanzes have discovered it?
Sure, and they may well have. We aren't chimpanzees though, and
probably wouldn't recognize it, or be able to use it if we knew it, or
adapt it to teaching more advanced concepts than the chimpazees have
yet reached natively, without a vast amount more observation and
experimentation than we have conducted to date. It may be
nonexistent, true - the point is that we just don't know. Despite (or
because of) our massively accelerated development, we haven't taken
anything like the time to find out for them as we have for ourselves.
We do know that at least some chimpanzees *can* be taught much more
advanced forms of language than we've ever observed them using when
raised by other chimpanzees. I think that proves that there exists
unused potential. The only question is how much.
>> And all those hard-won techniques were designed around the central
>> assumption that the student is human.
>
> No. Not true at all. Humans have spent millenia consciously and
> systematically trying out and perfecting methods for teaching several
> different types of animals.
That's nothing compared to the tens to hundreds of millennia spent
intensively raising human children, in far greater numbers and with
much greater intensity of effort. What's more, most of our animal
efforts have been directed toward species that are probably not
natively the most intelligent.
- Tim
Spiros Bousbouras
Mike Williams <nos...@econym.demon.co.uk> wrote:
> Wasn't it Juho Julkunen who wrote:
> >Oh, cool. I'd only heard about this:
> >
> >http://en.wikipedia.org/wiki/Blue_Brain_Project
>
> For which there's a nice presentation at TED:
>
> < http://www.ted.com/talks/lang/eng/henry_markram_supercomputing_the_brain_s_secrets.html >
>
> = http://tinyurl.com/yezchdy
[ Fixed link ]
There used to be a time when you could download a TED video as an mp4.
No more ?
Howard Brazee
wrote:
>
>There used to be a time when you could download a TED video as an mp4.
>No more ?
I subscribe to the TED podcasts - they are MP4 for me.
--
"In no part of the constitution is more wisdom to be found,
than in the clause which confides the question of war or peace
to the legislature, and not to the executive department."
- James Madison