Boris,
the biological one should be emulated
https://www.biorxiv.org/content/biorxiv/early/2018/11/27/478644.full.pdf
On 18.03.2019 02:56, Boris Kazachenko wrote:
We must define a process in which language can emerge from
incrementally complex encoding of analog sensory input. Anything short
of that is a cargo-cult AI.
On Sun, Mar 17, 2019 at 2:53 PM Jim Bromer <jimbro...@gmail.com
<mailto:jimbro...@gmail.com>> wrote:
I guess I should add that I think strong AI may start out with
simple 'principles/ or methods but it might start out with very
complicated principles and methods as well. I do not think an
attitude that animal physiology -must- be simple is very
realistic. However, I do not see any good evidence to assume that
simple methods cannot suffice as a starting point for stronger AI.
On the other hand, I think there is lots of evidence that
complicatedness is a major problem for stronger AI
So when I argue that the study of natural language processing is a
major move toward strong AI I am talking about AI that can adapt
to special languages that are used frequently amongst a group,
just as we have our own language to talk about what we are talking
about. The average person would have no idea what I am talking
about, but most of you can make some sense out of what I am saying
(whether you agree with it or not.) If a natural language
processing programming can adapt to novel usages of terms and
sentences, then it can learn, and I would say that it would also
need to have overcome the present day hurdles of complicatedness
in some way.
I think there are undiscovered mathematical methods that will one
day take a giant step over the present-day hurdle of complicatedness.
Jim Bromer
On Sun, Mar 17, 2019 at 10:01 AM Jim Bromer <jimbro...@gmail.com
<mailto:jimbro...@gmail.com>> wrote:
This argument from Robert Levy is not quite right, in my
opinion. While most animals do not have a sophisticated
language, it can be seen that animals are capable of learning
about routine events and attach meaning to linguistic cues (or
other kinds of sensory events like bells) to those routine
generalizations. That would constitute a language, and it
exemplifies the contention that to collect insight about (the
generalizations of a kind) of event constitutes a
symbolization of a precursor of the event. The knowledge that
a precursor might represent an event thereby demonstrates that
the animal has a basic 'linguistic' ability. And the idea that
an animal can associate a learned signal with a possible event
(like dinner) shows that the animal has the power of a
'linguistic' imagination.
Could designing a robot that has to learn to walk be the
breakthrough in strong AI according to Robert's thesis?
Because there are some animals that can learn to walk within a
few hours of being born A foal is an example. Foals have
spindly legs that splay a little with the first steps but they
are not mechanically designed for stability like a stationary
landing pod on a spacecraft. The idea that designing an
artificial process that is simple for some animals might
represent a breakthrough in AI does not make sense for one
reason. It does not take complexity into account. (I am
speaking of complicatedness of course.) It is very easy to
design AI programs that can operate within extremely simple
domain data-spaces The problem is dealing with extremely
complicated domain environments where complexity is a major
hurdle.
It is a mistake to think that language research in AI is not a
pathway towards AGI. However it is a mistake to think that
linguistic abilities are themselves strong AI just as it is a
mistake to think that designing a robot that can learn to walk
is strong AI. Both of these challenges can be met by
simplifying the environmental domain sufficiently. The
challenge is finding a way that true learning can take place
when confronted with thousands of complications.
Jim Bromer
On Thu, Mar 7, 2019 at 7:24 PM Robert Levy <r.p.l...@gmail.com
<mailto:r.p.l...@gmail.com>> wrote:
It's very easy to show that "AGI should not be designed
for NL". Just ask yourself the following questions:
1. How many species demonstrate impressive leverage of
intentional behaviors? (My answer would be: all of them,
though some more than others)
2. How many species have language (My answer: only one)
3. How biologically different do you think humans are from
apes? (My answer: not much different, the whole human
niche is probably a consequence one adaptive difference:
cooperative communication by scaffolding of joint attention)
I'm with Rodney Brooks on this, the hard part of AGI has
nothing to do with language, it has to do with agents
being highly optimized to control an environment in terms
of ecological information supporting perception/action.
Just as uplifting apes will likely require only minor
changes, uplifting animaloid AGI will likely require only
minor changes. Even then we still haven't explicitly
cared about language, we've cared about cooperation by
means of joint attention, which can be made use of
culturally develop language.
On Thu, Mar 7, 2019 at 12:05 PM Boris Kazachenko
<cogno...@gmail.com <mailto:cogno...@gmail.com>> wrote:
I would be more than happy to pay:
https://github.com/boris-kz/CogAlg/blob/master/CONTRIBUTING.md
, but I don't think you are working on AGI.
No one here does, this is a NLP chatbot crowd. Anyone
who thinks that AGI should be designed for NL data as
a primary input is profoundly confused.
On Thu, Mar 7, 2019 at 7:04 AM Stefan Reich via AGI
<agi@agi.topicbox.com <mailto:agi@agi.topicbox.com>>
wrote:
Not from you guys necessarily... :o) But I thought
I'd let you know.
Pitch:
https://www.meetup.com/Artificial-Intelligence-Meetup/messages/boards/thread/52050719
Let's see if it can be done... funny how some
hurdles always seem to appear when you're about to
finish something good. Something about the duality
of the universe I guess.
--
Stefan Reich
BotCompany.de // Java-based operating systems
*Artificial General Intelligence List
<https://agi.topicbox.com/latest>* / AGI / see discussions
<https://agi.topicbox.com/groups/agi> + participants
<https://agi.topicbox.com/groups/agi/members> + delivery options
<https://agi.topicbox.com/groups/agi/subscription> Permalink
<https://agi.topicbox.com/groups/agi/T191003acdcbf5ef8-M8d5654b1e3c7a33c6033c805>
------------------------------------------
Artificial General Intelligence List: AGI
Permalink:
https://agi.topicbox.com/groups/agi/T191003acdcbf5ef8-M2a757bec7541e9adcd58a513
Delivery options: https://agi.topicbox.com/groups/agi/subscription