View unanswered posts | View active topics It is currently Fri Sep 19, 2014 9:51 am



Reply to topic  [ 24 posts ]  Go to page 1, 2, 3  Next
Mycroft Holmes currently being developed? 
Author Message
Centennial Attendee
User avatar

Joined: Thu Apr 10, 2008 8:21 am
Posts: 735
Location: Sunnyvale, CA
Post Mycroft Holmes currently being developed?
A scientist working in Switzerland says he is creating the world's first conscious computer and will complete it by 2018.

_________________
“Don’t believe everything you see on the Internet.” –Abraham Lincoln


Thu Jan 07, 2010 9:21 am
Profile YIM WWW
PITA Bred
User avatar

Joined: Wed Apr 09, 2008 12:17 pm
Posts: 2400
Location: The Quiet Earth
Post Re: Mycroft Holmes currently being developed?
We've seen variations of this claim before. I recall some back in the 70s, when the first LISP, Prolog and SmallTalk computers were being developed. To quote a favorite Brian Daley line, "More likely someone's stock is in trouble. The gullibility quotient never goes down..."

I will admit we seem to be closer to the processing densities and speeds needed to approximate higher brain function. I doubt that horsepower alone is enough, though.

_________________
"Hier stehe ich. Ich kann nicht anders." - Luther
In the end, I found Heinlein is finite. Thus, finite analysis is needed.


Thu Jan 07, 2010 10:02 am
Profile
User avatar

Joined: Tue Dec 02, 2008 8:22 am
Posts: 562
Location: Reno, NV
Post Re: Mycroft Holmes currently being developed?
My brother would agree with Jim on that, and artificial intelligence and neural/human interfaces are his field. Consciousness is apparently something other than (and much more complicated than) the narrow thing we call "intelligence". It's doubtful that we can create it artificially, and if we can, it will require (just for a start) understanding and operating comfortably at the quantum level in physics. *Then* we can start looking at emulating biology using non-biological methods.

I'm not holding my breath.

_________________
Catherine Jefferson <tw86034@ergosphere.net>
Home Page: http://www.ergosphere.net


Thu Jan 07, 2010 7:29 pm
Profile WWW
Heinlein Nexus
User avatar

Joined: Thu Apr 10, 2008 8:10 am
Posts: 2105
Location: Pacific NorthWest
Post Re: Mycroft Holmes currently being developed?
I gotta love an article that says:

Quote:
The professor is not mad, but he is unsettling.


One of the few sessions I was on a panel at the Centennial for was on precisely this topic.
On the one hand, the argument that the claim of artificial intelligence is bogus because every time in the past the claim has been made it has been bogus is fallacious: at some point it will happen. The question is, is this that point?

On the other hand, a timeframe of 8 years in cybernetics is equivalent to a 20-year timeframe in any other branch of technology. It is beyond the event horizon: it is out in the space where everything short of time travel and invisibility cloaks is possible. To me, the only interpretation possible is, "We have no idea what we are doing."

I've seen assertions in years past that we would have hardware approximating human brain power readily available around 2012. I don't find that hard to believe; the problem is one of software. How do you write a human brain? Since time immemorial people have had the most half-baked models for human cognition. The notion, for instance, that electrical shocks of specific brain locations triggering various responses tell us anything useful about how the brain works whatsoever is both childish and quaint.

Now, this dude is at least going about it the right way: model the primitive functionality of the brain and let it teach itself. But no one has a clue how to do that. We've known the dendritic structure of the brain for decades and that's only been good enough to build neural networks. Useful, yes, but they do not think. So just modeling the currently known physical structure of neurons is no good. There's been a project underway for some years to teach an expert system as many facts as possible; I think they're up to a million. They seem to believe that if they put in enough, it will become alive. This is like shoveling a hundred pounds of offal into a sack in the expectation that it will turn into a pony.

Still, it keeps 'em off the streets...


Sat Jan 09, 2010 12:15 am
Profile WWW
Heinlein Biographer

Joined: Thu Apr 10, 2008 1:33 pm
Posts: 1024
Post Re: Mycroft Holmes currently being developed?
PeterScott wrote:
I gotta love an article that says:

Quote:
The professor is not mad, but he is unsettling.


One of the few sessions I was on a panel at the Centennial for was on precisely this topic.
On the one hand, the argument that the claim of artificial intelligence is bogus because every time in the past the claim has been made it has been bogus is fallacious: at some point it will happen. The question is, is this that point?

On the other hand, a timeframe of 8 years in cybernetics is equivalent to a 20-year timeframe in any other branch of technology. It is beyond the event horizon: it is out in the space where everything short of time travel and invisibility cloaks is possible. To me, the only interpretation possible is, "We have no idea what we are doing."

I've seen assertions in years past that we would have hardware approximating human brain power readily available around 2012. I don't find that hard to believe; the problem is one of software. How do you write a human brain? Since time immemorial people have had the most half-baked models for human cognition. The notion, for instance, that electrical shocks of specific brain locations triggering various responses tell us anything useful about how the brain works whatsoever is both childish and quaint.

Now, this dude is at least going about it the right way: model the primitive functionality of the brain and let it teach itself. But no one has a clue how to do that. We've known the dendritic structure of the brain for decades and that's only been good enough to build neural networks. Useful, yes, but they do not think. So just modeling the currently known physical structure of neurons is no good. There's been a project underway for some years to teach an expert system as many facts as possible; I think they're up to a million. They seem to believe that if they put in enough, it will become alive. This is like shoveling a hundred pounds of offal into a sack in the expectation that it will turn into a pony.

Still, it keeps 'em off the streets...

The owverwhelming problem in AI has always been conceptual. The idea that hardware speed and capacity has anything at all to do with AI is bizarre and dates back to Turing's ideas in the 1950's, which were based on rather primitive neuroscience of the time. They had an extremely fuzzy notion of the anatomy of the brain, none of the studies of fine-grain structure having yet been done, and thought intelligence emerged out of the total number of neurons, rather than the organization.

Some numbnuts casually mentioned to my brother-in-law last week that Rice's Theorem made AI forever impossible -- something she thought she had learned in a college course. Well, of course, Rice's theorem has nothing to do with AI: if it's as true of human languages as it is of the kind of self-referential mathematical propositions it's concerned with, which nobody knows, then it's manifestly obvious that human languages manage to cope with it somehow. And if it doesn't apply to human languages, then it's not a valid criticism of natural language computability.

But a side comment of hers caused me to realize that popular knowledge of AI disciplines is also plagued by the fact that as individual problems are solved they have been moved out of AI, so that the category is left only with the intractable problems. If you take the definition of AI as it existed in 1951, AI is here and has been here for fifty years. I think the computer chess problem was solved about the same time as the first human-language mimicing program was developed -- imitiating a non-directive psychological therapist.

So, depends on what you mean by AI.


Sat Jan 09, 2010 9:29 am
Profile
Centennial Attendee
User avatar

Joined: Thu Apr 10, 2008 8:21 am
Posts: 735
Location: Sunnyvale, CA
Post Re: Mycroft Holmes currently being developed?
BillPatterson wrote:
But a side comment of hers caused me to realize that popular knowledge of AI disciplines is also plagued by the fact that as individual problems are solved they have been moved out of AI, so that the category is left only with the intractable problems.

This reminds me of the field of psychiatry. Behavioral disorders are assigned to psychiatry until medical science figures out which discipline they really belong in, or they're redefined as not being disorders (e.g. homosexuality). Syphilis and epilepsy were originally psychiatric disorders, for example.

_________________
“Don’t believe everything you see on the Internet.” –Abraham Lincoln


Sat Jan 09, 2010 10:17 am
Profile YIM WWW
PITA Bred
User avatar

Joined: Wed Apr 09, 2008 12:17 pm
Posts: 2400
Location: The Quiet Earth
Post Re: Mycroft Holmes currently being developed?
PeterScott wrote:
...the argument that the claim of artificial intelligence is bogus because every time in the past the claim has been made it has been bogus is fallacious: at some point it will happen.

Not sure to whom or what you're referring, but I said nothing of the sort. All prior claims of a breakthrough have proven false; that's where we stand.

I remain agnostic on the notion that "classic AI" - a machine that can for all intents and purposes "think like a human" - is possible. I am caught between the idea that at some point (as Heinlein put it) a certain level of processing complexity will result in sentience, and the idea that true sentience, or even a very good simulacrum of it, involves something more subtle than anything machine computing can do. I claim no expertise whatsoever in the matter.

Something of a sidelight, but I do find it endlessly amusing that Golden Age authors (and beyond) assumed that speech recognition was trivial and would be solved long before speech synthesis. There are hundreds of examples in the lit.

_________________
"Hier stehe ich. Ich kann nicht anders." - Luther
In the end, I found Heinlein is finite. Thus, finite analysis is needed.


Sat Jan 09, 2010 11:49 am
Profile
Heinlein Nexus
User avatar

Joined: Thu Apr 10, 2008 8:10 am
Posts: 2105
Location: Pacific NorthWest
Post Re: Mycroft Holmes currently being developed?
JamesGifford wrote:
PeterScott wrote:
...the argument that the claim of artificial intelligence is bogus because every time in the past the claim has been made it has been bogus is fallacious: at some point it will happen.

Not sure to whom or what you're referring, but I said nothing of the sort. All prior claims of a breakthrough have proven false; that's where we stand.

Ah, my apologies.
Quote:
I remain agnostic on the notion that "classic AI" - a machine that can for all intents and purposes "think like a human" - is possible. I am caught between the idea that at some point (as Heinlein put it) a certain level of processing complexity will result in sentience, and the idea that true sentience, or even a very good simulacrum of it, involves something more subtle than anything machine computing can do. I claim no expertise whatsoever in the matter.

I'm certain it will be done at some point, almost certainly within 10 years, but it may be like neural networks - it'll have trained itself and we won't know how it did it. We'll have to ask it how it learned, and hope it knows, and hope we can understand the answer.
Quote:
Something of a sidelight, but I do find it endlessly amusing that Golden Age authors (and beyond) assumed that speech recognition was trivial and would be solved long before speech synthesis. There are hundreds of examples in the lit.

Not to mention Star Trek, of course. The Japanese government technology arm, MITI, had near godlike status by the mid '80s and when they announced that they were launching projects to solve four tough problems, well, everyone figured it was curtains for those problems, one of which was continuous speech recognition. Well, of course they didn't make much of a dent.

But here's the interesting thing. In the last couple of years, speech recognition has hit the prime time. I use the Google voice search app on the iPhone. It's practically psychic, and it required no training. I also got the Dragon NaturallySpeaking app for the iPhone, and it does a better job than the PC version of that program did when I first got it in late 1999 - and yet the PC program required an hour of training on my voice where this required none, the PC program required a high end headset where this works with the builtin hardware, and the PC program cost $400 in 1999 dollars whereas this is free... and its accuracy rate is considerably better.

I'd say that speech recognition on the iPhone alone is now good enough to be useful.

BillPatterson wrote:
So, depends on what you mean by AI.

I'll define it as something that passes every version of the Turing Test you can throw at it.


Sat Jan 09, 2010 5:51 pm
Profile WWW
PITA Bred
User avatar

Joined: Wed Apr 09, 2008 12:17 pm
Posts: 2400
Location: The Quiet Earth
Post Re: Mycroft Holmes currently being developed?
PeterScott wrote:
We'll have to ask it how it learned, and hope it knows, and hope we can understand the answer.

"There is... now."

:D

_________________
"Hier stehe ich. Ich kann nicht anders." - Luther
In the end, I found Heinlein is finite. Thus, finite analysis is needed.


Sat Jan 09, 2010 6:34 pm
Profile
Heinlein Biographer

Joined: Thu Apr 10, 2008 1:33 pm
Posts: 1024
Post Re: Mycroft Holmes currently being developed?
PeterScott wrote:
BillPatterson wrote:
So, depends on what you mean by AI.

I'll define it as something that passes every version of the Turing Test you can throw at it.

The classical Turing test as phrased is meaningless. A computer intelligence would have a vastly different experience base than its human questioner, and we would expect that to be reflected in its speech. The classical Turing test only requires that the human questioner recognize the difference between a human and a computer -- which means that a successful machine intelligence would likely not be able to pass the test.

However, if the test simply recognizes an independent intelligence, then it might be useful.

There is also the problem that we don't really know what intelligence is. There are some things that, if you do a successful mimicry, then you're it -- such a teaching. If you mimic teaching succcessfull, you're ipso facto a teacher, however you go about it. Is a machine program that can successfully mimic a human intelligence ipso facto intelligent? Pass -- ask me in a hundred years, maybe. That ought to be a long enough frame to -- if not "solve" the problem, at least get enough insight on the issues to make a stab at an answer.

Fortunately, I don't have to deal with these thorny issues.


Sun Jan 10, 2010 8:49 am
Profile
Display posts from previous:  Sort by  
Reply to topic   [ 24 posts ]  Go to page 1, 2, 3  Next

Who is online

Users browsing this forum: Google [Bot] and 3 guests


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group.
Designed by STSoftware for PTF