Bingo!! The technological world is primarily driven by practical application. The only direction I see human emulation going is exemplified by the voice-recognition and voice-response call-routing systems large businesses are using to handle incoming calls. Just like with new supermarket checkout systems, companies are working hard to take the human out of the loop i.e., no carbon-based lifeforms for the customer to complain to any more.

This is certainly nowhere approaching AI, but they're getting the human emulation part of it down pretty well at a superficial basic communications level. I'm expecting to go into a doctor's office one day and have a machine say, "tell me where it hurts." I'll respond with the same thing I say in the supermarket, "Where's a human -- I wanna talk to a human."

Luckmeister

__________________
Why the avatar grin? I'm watching my favorite actress/model through my 4D glasses -- it sure beats looking at a tesseract.

The development of AI that are exactly like humans is quite unlikely. For a start the speed of electronic processors is different to biological neurons, and the format of electronic memory is very different. I think that many different forms of sentience might be developed, or perhaps emerge, in the next several decades; but they will have very different characteristics to human minds.
For instance these hypothetical AI might process information much faster, and have access to much more data and memory capacity, but be much less self-aware and conscious. In fact to fully emulate a human-like mind, it seems likely that the problem will be approached from the opposite side. A computer with much greater processing power than a human brain will be needed to emulate the many varied processes within a single brain in a laborious and inefficient manner, and also restrictions will need to be placed on its speed and memory capacity accordingly, or the model will not resemble a slow-thinking, intuitive human with a typically fallible and limited human memory capacity.

So in my opinion, the creation of a fully human-like AI, with all our sense of self-awareness and our characteristic limitations, will require a much more powerful machine to be built to support such an emulation.


This puts the projected development of such a realistic human emulation even further in the future. And such an emulation would probably be an inefficient use of processing power; I think that fully competent thinking machines will be possible without having to emulate human minds. I am thinking of machines which can respond intelligently and inventively to situations, and can adapt their behaviour according to contingencies.

One of Kurzweil’s main themes is the creation of friendly AI, that is to say AI which are not dangerous to humans. It seems unlikely that you can constrain thinking machines by writing simple instructions into their programs, as suggested by Asimov in his Laws of Robotics. Instead thinking machines need to have deep seated goals which coincide with the best interests of humans; and if a machine is capable of creating its own goals, then those goals should not conflict with those of the human population.
It seems to me that it is possible that a fully sentient machine which resembled humanity in some respects would be more likely to spontaneously develop goals which did not conflict with the well-being of humanity; but on the other hand certain aspects of human behaviour are perhaps best not modelled with too much fidelity.

__________________
Orion's Arm . The Starlark . Voices: Future Tense . OA Flickr set

Maybe, but maybe we're just approaching it the wrong way.

Human minds have both discrete and analog components. Perhaps the addition of chips with more gradual thresholds than transistors would help in emulating intelligence. Removing a crystal-based system clock might also add some necessary uncertainty.

It all comes down to emergent behaviour. You stick a bunch of simple components together, and they do something very complicated. The thing is, we don't know how to predict exactly how such a system will evolve. It's chaotic in the extreme. On the other hand, we can in many cases get a general idea. A simulation of bird flocking with three or four simple rules can actually make it look like a flock of birds. Sure, maybe it's not exactly how they do it, but the general behaviour exists for a range of parameters.

So, how do we best emulate human intelligence? How do we decide which path is the one to follow? I think that if that's the goal (and I'm not convinced it should be, but ignore that), then we ought to do the same as that flocking bird simulation. Don't worry about if it's exactly the same. Just make something with a set of adjustable parameters and fiddle with them until it's as close to what we want as possible. Then, if necessary, add more rules (possibly meaning different hardware), or take out ones that hinder the process, and try again. We might not get an exact match, but we'll get something that's pretty close, and from there it's just a process of refinement.

__________________
"It's turtles all the way down."

And one practical application which has both the most use for AI and the most dollars behind it is military. Yet how much intelligence does a fully autonomous combat machine (call it Ralph ) really need? Ralph has to react to external stimuli -- something almost any animal can do. It has to make tactical decisions -- something almost any predator can do. And it has to learn from experience -- something almost any mammal can do. Realistically, Ralph needs not be more intelligent than a cat.

Ralph should have language-recognition capabilities, because humans convey information best by talking, but any speech on its part will be superfluous. Given the level of technology required, all friendly humans are likely to have head-up displays or some other visual interface, and Ralph will convey tactical information that way. Humans can absorb visual data MUCH faster than audial, which is one of the reasons talking computers never really took off. And that's another nail in the coffin of Turing Test -- I do not want an AI which can discuss implications of overfishing in the Gulf of Maine -- I want one which can put Gulf of Maine onto a mutimedia graph, and give ecology time-projections based on various fishing policies. Something no human can do, at least not in real time.

__________________
Fiction has to be plausible. Reality is under no such constraint.

First thing to do is to pump 20 yeaars of experience to some computer and make it all asociative. AI is required for many jobs, for example language translation: no machine can translate text well, because it does not underatand it. good translator must be educated to understand text. So first possibility is to create extremely huge database where words are asociated with meaning. If computer sees word 'hamburger' it must expect sentence about eating. as I know such semantic databases exsists but they are extremely huge and none of ususal computers can check everything at once.
Our brains can do that esily, so here is starting point for AI. It must work in paralel, and that can be done using FPGA instead of CPU. If you need realy fast processing, FPGA neraly always wins in sound and fast image recognition area.

People seem to think that human level intelligence is something that will be incredibly difficult to replicate in a machine, but having been outsmarted by a parrot, I'm not so sure.

I'm going to quote that somewhere, I promise!

__________________
"It's turtles all the way down."