Advertisement

a different approach to machine intelligence

Started by June 19, 2005 09:53 AM
13 comments, last by cwhite 19 years, 5 months ago
i think it would be possible to make a machine intelligent the same way we became intelligent. when we are born we are born with 'human nature'. you know, things like curiousity, the need to be loved, wanting to mimic those close to you (that's how babies learn to talk. they copy the sounds that they hear), etc. one day we should make a 'baby' robot. just a robot with all the senses and potential abilities that the basic person has, and give it any human nature senses that apply. the robot could learn the same way a young child does. it could see someone holding an apple and saying 'apple' and relatively quickly (since robots don't forget much) the robot would associate the sound of 'apple' to what an apple looks like. it's just too bad we are only at the early stages of robotics where robots still walk and move in straight, slow, angled, robot ways. and artificial eyes aren't very much developed either. but once these fields become more advanced i think this idea might work.
See http://mainline.brynmawr.edu/DevRob05/
Advertisement
Quote: Original post by chuck22
i think it would be possible to make a machine intelligent the same way we became intelligent.


Unfortunately we don't really know what makes us intelligent yet. The brain and the mind are still largely mysteries to us.
i know this but i'm making the assumption that our intelligence probably comes from learning by example and learning from experience. we build our intelligence by learning from others and we extend our intelligence further by curiousity and exploring new places or trying new ideas.

i think intelligence is much simpler than what people make it seem like.
Quote: Original post by chuck22
i know this but i'm making the assumption that our intelligence probably comes from learning by example and learning from experience. we build our intelligence by learning from others and we extend our intelligence further by curiousity and exploring new places or trying new ideas.

i think intelligence is much simpler than what people make it seem like.

Not quite right. There are many important human systems which we are born with, and other intelligent animals seem to lack. For example one couple (psychologists) raised a monkey as if it was part of the family, using sign language to communicate with it and just generally treating it like a human child. Yet once it had grown up, it had no understanding of grammar what so ever. It was able to develop a vocabulary, but human language requires grammar to combine basic concepts into things more interesting. Instead, if he wanted a banana he would say "Banana Want Banana Banana".

I doubt tht we have the capability to create a robot that is even up to the level of the monkey. Would anyone happen to remember its name?
Turring Machines are better than C++ any day ^_~
Well, MIT has Kismet, which does something similar to what you want. The project has been going on for quite some time now....many many years.

http://www.fortunecity.com/skyscraper/lol/1165/kismet/kismet.html

Quote:
it's just too bad we are only at the early stages of robotics where robots still walk and move in straight, slow, angled, robot ways. and artificial eyes aren't very much developed either.


Actually, take a look at the current specs of Honda's Asimo and you'll be quite amazed. It is capable of autonomous navigation, facial recognition, obstacle tracking and avoidance, etc. It is pretty much state of the art in a 5 foot tall package that has wireless internet access. Currently, Asimo is the probably the most advanced working robot there is.

Japan is actually planning on mass marketing house keeping robots for commercial consumer use in the next 5 years. The current price tag for one of those house keeping, dish washing, laundry cleaning robots is around $1000 USD or less. They will probably be the most advanced concumer grade mass produced robots when they hit the market. As to how serious the japanese are about this? Well, let's just say that new homes that are built in the past 2 years come with the option of having a small wall space where a robot can go sit and recharge itself.

So we really aren't at an early stage of robots, its just that many countries are.

As for having robots that really grow and develop, maybe Asimov was right about positronic brains. :p
Advertisement
i think that we can make a robot smarter than a monkey. there's a repetitive process you might have to go through but i bet it could be done. just give the robot all the natural abilities and senses you think it will need in order to learn and behave independantly. if there is something lacking (like saying banana want banana) all you have to do is rebirth a new robot that fixes the problem OR shutdown that program and add the needed code. it wouldn't take as long as raising a monkey and you are able to destroy the robot and create a better, newer version...but you can't kill the monkey ;)

eventually after several repetitions of this process you would get a somewhat truly intelligent machine.
One of the primary mental attributes that humans use to process information in a more "intelligent" way than other species is the ability to make a mental connection between seemingly unrelated abstract ideas. This is what allows us to be creative. We not only learn by example, we take the stored memories of things we have experienced and apply them in unique combinations to situations we have not encountered before. We use a combination of fuzzy logic, neural networks, and the still mysterious "imagination" to achieve this. In essence, it is "simple". But in practice, to produce a truly self contained system of this magnitude is no small task. Just like game programmers, scientist can produce AI that mimics many aspects of human behaviour such as learning the direct correlations between input experiences, parsing through stored action/reaction/result data networks, etc... but to produce a dynamically self directed and "imaginative" system capable of bridging the gap between cold data handling routines and truly abstract thinking is still beyond our understanding. No harm in trying to figure it out though. ;)

"Not a robot, a cyborg. A machine covered in living tissue."
(Kyle Reese)
whew that was a lot but you are right. i would still like to someday see/make a program/robot that comes very close to the level of our intelligence, or atleast see/make the most artificially intelligent robot ever made within the next decade. it is likely to happen with that time frame but the sooner the better.

i was thinking earlier today on another reason why we give the robot only the basics and have it learn on its own. this just popped into my head. let's say you go to another country and tell someone "hey, nice wheels" and you say it in a very sarcastic voice. they will probably have no idea you are making fun of the fact that they either have no car or have a crappy car because they are not familiar with foreign dialect or sarcasm. they will assume that you think that they have some wheels and they look nice.

now let's say your back to your own country and you say that to a 4 year old toddler. he will probably have no idea what you are talking about but he will pick up on your sarcasm and react to your statement negatively. he'll be able to recognize your sarcasm because he hears it often, but doesn't necessarily understand it.

now you say the same thing to a young teenager, say a 15 year old freshman in high school. he will understand it completely, pick up on your sarcasm, and come back with a smartass comment.

the point i'm getting at is that you can't just give a robot complex knowledge from the start. it must be learned over time. and if complex knowledge must be learned over time then it can be assumed that simple knowledge must also be learned. this is why i think a great approach would be to make a robot with a clean slate of memory, but give it the ability to learn. give it the ability to break down the masses of info into small simple knowledge inputs. then use previously acquired knowledge to learn more or use that knowledge.
Quote: Original post by chuck22
i think that we can make a robot smarter than a monkey. there's a repetitive process you might have to go through but i bet it could be done. just give the robot all the natural abilities and senses you think it will need in order to learn and behave independantly.


Yeah, but you're really ignoring the whole problem, which is that we don't know how these things work. We can't just add a new program to make it work when we don't know how it happens. As pointed out above, there have been a lot of interesting advances in robotics but we are nowhere near understanding how animals actually learn or think, or how natural abilities arise. It's easy to program a computer to learn and adapty to one very specific task, or a several individual tasks, but a generalised learning process that can cover everything and yet still be effective and efficient is not likely to happen until we answer some more basic questions about both biology and psychology.

This topic is closed to new replies.

Advertisement