Making Expert Systems Smarter

I wish you luck.

Seems to me the way humans solve problems is via concepts, not via rules. And concepts are fuzzy things that don't have firm boundaries. Personally I don't think that anything that isn't loose and sloppy can make progress in the AI field. Besides, humans make decisions based on emotion rather than on rules.

And it seems to me that expert systems have had the most success when they have had very restricted limits to their inquiry. But to make progress you need to remove those limits. You can look at the irrigation problems on your farm all you want, but if you don't know that you and all your neighbours are drawing down the aquifer and you're collectively running out of water, then you won't arrive at a good solution.

DrClapa at 2007-7-13 14:56:05 >

"You mean like 'Hal' in 2001?"

eventually -- there's more chat capability being worked on in robotics today than anywhere else as far as I can see .... but on the research end, there's the "upper ontology" project that aims to create standard vocabularies where the terms are understood ... not just included in a list (expressed in xml)

RogerFGay1a at 2007-7-13 14:56:05 >

OK ... I get it now. Sarcasm doesn't always come across clearly when you write ... unless you're explicit.

I've been chatting with people who are really serious about this stuff and aren't just chatting, but working on real deliverable technology. Believe it or not, they're actually quite concerned about robot ethics and Azimov's robot laws.

My views are tempered. You can see in my presentation that I jump from long-term vision in one slide to short term expectations in the next. I don't think the first generation applications of HLL are going to drive machine communities that will take over the world. I just want applications that can "think" (in a highly structured fashion not so wildly removed from programs we're used to) their way through more complex situations ... take things in context and so on.

Azimov's robot community tried to take over the world because the mamma computer thought outside the box (the three laws). Protecting humans being fundamental to the nature and purpose of robots, she reasoned she could protect humans better if they were all under her control. Her new view was more statistical than individual ... i.e. not respecting individual integrity but looking for a better statistical result. Azimov spent his early years in Russia before moving to America ... this tension between the statistical (commie) view and individualism (American liberalism) was part of the charm of his stories.

RogerFGay1a at 2007-7-13 14:56:05 >

One more thing ... I guess it depends on which sci-fi you think about and what you choose to emphasize. Within the length of my lifetime, as least as far as I invision my efforts having an effect, I don't see computers that are just like humans at all. Somebody does, I'm sure, but my basic orientation is engineer.

I want machines that do useful things, that do those things efficiently, and that are reasonably easy (at least "relatively" in consideration of cost / benefit) to construct and maintain. I want technology that is re-usable for a range or purposes.

I have come to the point of agreeing with those who think the human model is one that is worth looking at. Evolution provided humans with some of the "right answers" for building efficient, effective machines that are useful for a variety of tasks.

This isn't a philosophy that I've jumped on immediately after having heard it, just because it sounds interesting. I've been through years of paying dues, encountering and working to solve technology problems .. and now I think there's something to it.

RogerFGay1a at 2007-7-13 14:56:05 >

The main problem here, is it not about computational power or lack

there of. Last time I read about this, a reasearch team managed

to simulate a neurol net comparable of a rat in a slowmotion rate of

1:1000. The models used are crusial, yes, but if we lack the tools

to try them, how do we know they will work in large scale? Some

people might say, lets create a scalable model and see how it works

with the computational power of today. I think we are not even

close in computational power to be able to find a model that will

scale well. Its like trying to build a scalable model of Airbus A310

in the size of fly and expect it to have the same flying charactistics

as the real one.

Then some people turn to expert system. The idea is solid: When

the computational power does not allow to create general solutions,

specialize to extreme. I would say that this strategy is working

much like a toolbox. Every tool (expert system) is higly adapted to

its perpose but if the request of the client does not apply to its

specifications the tool can not help the client. I write tool because

without the computational power needed to see more complex patterns,

it will never be much more then any other tool we have today.

I guess my main point here is that, why spend huge amount of money

on AI research that is hugly dependent on areas that has not reach

far enought yet. Do not get me wrong, I think that human like AI will

be the most important creation that we will every invent.

parza at 2007-7-20 12:23:37 >

> > > what is your goal?

> >

> > To seek the Holy Grail.

>

> I have one that is engraved. Not only is it engraved

> but it is also unwashed. I'll start the bidding at,

> oh let's say $2 million.

2 million Duke $? That's a bargain!

prometheuzza at 2007-7-20 12:23:37 >