|
| |
|
|  uknugget | Jan 2, 2006 1:49pm | I just joined the forum today.......Over 200 members and absolutely nothing worth mentioning in the field of AI? I come across exciting stuff every single day, and the potential for AI/MI in manufacturing, labour intensive applications, software, social science.....basically every single study of every single subject (If it's an -ology it's science :-)) ever considered by man will be affected by machine intelligence. And nobody has anything to write on the subject! :-)
I've got some suggestions for discussion threads:
How about starting with some basics in the field of neural networks?
1. Why neural networks? Why not algorithmic programming?
2. What are neural networks, and how closely related is it to the biological model?
3. How do machines self-learn? (Types of learning in NN)
4. What are the hardware/software requirements?
5. What is Machine Intelligence?
What are the implications for the future?
1. Can machines ever be self-aware?
2. What kinds of social change can be envisaged?
3. Will they be good/evil?
4. How will we treat them?
Reader recommendations on websites/books/university courses
You know, that sort of thing. I think I can help and give my 2 cents worth to any of the questions above, so ask away if you're curious or post away if you have something to tell us. :-)
All the best, Yong
(AI & NN was an optional unit I took at college) |
|
| 
Sponsor |  Nexus76 | Jan 2, 2006 6:05pm | Lol - it IS a pretty quiet forum :) (I s'pose I should take the 1/200+ bit thats my fault though :/) - hopefully your post will spark a few replies - since you've posted a whole load of possible threads I'm not sure where to start (but I'll give my 2c worth anyway)
1. Why neural networks? Why not algorithmic programming?....It depends on what you want to achieve - the fields cover much of the same ground but I think historically NNs have been researched to a greater depth and have been demonstrated to perform well in a number of areas - I'd like to see a bit more work done on algorithmic programming myself - the main difference between the two is probably in the ability of each to provide optimal solutions - which in many cases is more preferable than ANY solution.
2. What are neural networks, and how closely related is it to the biological model?...They're similar mechanisms - but NNs are not biologically accurate (and weren't meant to be).
3. How do machines self-learn? (Types of learning in NN)....there are SO many types of machine learning algorithms...if its NNs only theres a partial list available at ftp://ftp.sas.com/pub/neural/FAQ.html#A_kinds
4. What are the hardware/software requirements?...To me its all about what you want it to achieve...just as with any computationally expensive program the more hardware you can afford to throw at the problem the better.
5. What is Machine Intelligence?....LOL - thats an evil question :) - Its used by AI students everywhere - mainly to bring a seminar to a fast conclusion :) (file alongside "what is intelligence?" and "is the brain just a complex computer?") - In other words DON'T EVEN GO THERE...hehehe
One thing I'd be interested in knowing is the number of people on here that have studied AI/ALife at college/University / how many want to know more about it / how many just have a passing interest ....and for what reasons (social/IT/programming/games etc) |
|
|
| | uknugget | Jan 7, 2006 7:03pm | Okey dokey. Many thanks to Nexus76 for his efforts on our behalf. I don't believe we've had any other takers, so I'll just finish off this thread and pose some questions which everyone can join in.
The answers given by Nexus76 are all valid. I'll add a little for spice:
1. I'll add that neural nets are much more robust than algorithmic programs, easier to program, faster to run, and have the ability to be trained to an optimal solution. The reason computers are so stupid is that it does exactly what it is told, and nothing else. Hence you have to tell it exactly what to do, step by step - the definition of an algorithm. However, some neural nets have a self learning, self categorising ability which means it sorts out the inputs by itself, and can learn to achieve the desired effects. It sounds like magic but it really isn't, and it is closely related to 2. Text book reasons for using NN are listed here: doc.ic.ac.uk/~nd/surprise_96/journal/vol4/cs11/report.html [doc.ic.ac.uk/~nd/surprise_96/journal/vol4/cs11/report.html]%
20neural%20networks
2. If you want to understand neural networks, you have to understand the concept of a perceptron - the model of a neuron (MacCulloch and Pitt engineered a model of a biological neuron, and the perceptron is a modified version). Wiki has a good explanation of it. According to research at Ecole Polytechnique Fédérale de Lausanne (EPFL), they are actually trying to map the circuitry of the neocortex, modelling the brain at the cellular level to shed light on internal processes such as thought, perception and memory.
Obviously, NN's are not biologically accurate, because we have to use electronic hardware, but we are copying biological models of the brain circuitry to learn more about the basis for intelligence. There is no other way to make machine intelligence - the only examples of intelligence we have is ourselves, so the template for creating AI/MI must reside in our own heads.
4. Software/Hardware requirements for a truly intelligent machine that can pass the Turing test? That's a question everyone asks. I think the seeds of an idea for creating MI lies in the research being undertaken at EPFL. The template for creating intelligence lies within our own brains. Obviously, there are philosophical as well as technical questions to be answered about the research, such as how one would develop a stable intelligent machine with its own concepts of the id, ego, and superego. A mind that can learn as well as invent. A mind flexible enough to reprogram itself without damaging essential parts. A mind strong enough to resist what humans often suffer from: such as psychoses, and mental disorders. :-)
I've recently thought about the hardware requirements when I stumbled upon this site: longbets.org/1 [longbets.org/1] , which is dedicated to taking bets as to whether a computer will be able to pass the Turing test by 2029. (The Turing test is the famous imitation test devised by Alan Turing in the 1950's to determine the answer to the question "Can Machines Think?". An interrogator is connected to a terminal which is connected to a machine and a person. Her task is to determine which is human by asking a series of questions.) BTW, I don't think Alan Turing is the sole inventor of modern computing. He was a damned good mathematician, but Collosus, the first electronic computer, was designed and built by Thomas Flowers, who has not been given the recognition he deserves.
As far as software is concerned, I think EPFL will give us some good answers. As far as hardware is concerned, I have no doubt that it will happen well before that date. According to most journals, based on the number of neurons we have in our heads (around a trillion), a rough estimate is 100 trillion connections capable of 10 quadrillion instructions per second. In 2004, IBM's BlueGene/L recorded 71 trillion instructions per second.
Just for fun:
The brain/computing ratio
10x10^15 / 71x10^12 = 141.
So, as at 2004, in terms of raw computing power, the brain is only 141 times more powerful than the fastest machine!
Using Moore's law that computing power doubles every 18-24 months, the computer will be able to process more instructions per second than the human brain in less than 16 years.
5. What is machine intelligence? Unlike Nexus76, I love questions like these. Not because they are unanswerable, but because it raises more interesting questions. I think machine intelligence will simply be human intelligence in machine form, and by and large it will not be much more different than the criteria with which we define our own intelligence. In creating, we can not help but instill a part of us in the machine. Even during the process of creating a template for intelligence, we will have much to discover about ourselves before we can use that knowledge to create intelligent life. I can't imagine what the world will be like living with intelligent machines, but I am very hopeful that it will lead to better and more wonderful things.
I started the thread with the title Dead as a dodo, and I think the world is in desperate need of intelligent life before we too end up like the creatures in the title. One of the defining moments of sentient life must have been when the first ape projected himself into the eyes of the other ape and tried to understand him. Empathy, communication, cooperation, and co-existence have all been hallmarks of civilisations and intelligence. Therefore, intelligent beings can not be violent, evil, or aggresive. When we do have our first MI in a few years time, I hope it will mark the end of conflicts, wars, and mismanagement which mark the times of today.
What do you think? |
|
| 
| | cramudgen | Jan 9, 2006 2:43am | "Therefore, intelligent beings can not be violent, evil, or aggressive"...soooooo, you just called humans stupid (lacking intelligence).
programmers who are working on AI/NN, will not be able to "create" AI of any consequence for many, many years..too many egos and not enough intelligence. |
|
|
| | uknugget | Jan 9, 2006 2:29pm | Erm.....I'll take that as a joke. :-) Certainly, one wouldn't call violent, evil, or aggressive behaviour smart things for an individual to exhibit?
"programmers who are working on AI/NN, will not be able to "create" AI of any consequence for many, many years..too many egos and not enough intelligence."..............Soooooo you just called programmers egotistical and lacking intelligence? :-)
Well, that's what this forum is all about. I emphasized the importance of neural networks in creating AI, precisely because painstaking programming in the classic algorithm sense, does not yield a result as good as self learning neural nets. For example, if you want a computer to recognise a cup every time, you have to put in the parameters for all the cups in the world. With a NN, you can train it with a few examples and it will recognise a cup every time. Even ones it has never seen before.
It is amazing how even the simplest NNs mimic human child learning behaviour. The ability to self categorise inputs, the ability to learn from negative and positive reinforcement, and the ability to follow examples. (Unsupervised, Reinforcement, and Supervised learning). Extrapolate that learning behaviour to more complex problems, and you have a fast, smart expert system. Extrapolate the expert system to include circuitry for thought and perception, and you have AI.
I anticipate you might ask, "What's the point of extrapolating anything? Nobody knows for sure how or when AI will emerge." True. But Moore's law has been pretty consistent. Even when people worried about the nano barrier (aka nanometer bridge) to making semi conductors more compact, new technologies emerged to keep the industry on track. And extrapolating data is a proven scientific concept. We would not have the Kelvin temperature scale starting at a theoretical absolute zero if extrapolating data was useless. In terms of the rate of advances in NNs and the anticipated rate of advances in hardware, it seems to me to be pretty certain that I'll be seeing a thinking machine, with a mind better or much like my own within my lifetime. |
|
| 
|  Mikecimerian | Jun 26, 2006 10:48pm | Did Man decide to create culture? Can an isolated human individual bootstrap himself to personality and positive thinking?
AI is already emerging. It just can't say hello yet any more than a microbe can.
Who decides when we cross the treshold of non-determinism toward autonomy?
We have already seen one century of false antagonism between psychology and biology trying to define "human ". Imho, information consistency in human-machine conversation is more important than imposing a locus on the phenomenon at this stage of evolution. |
| |
|
You need to Sign-up for StumbleUpon to post to this forum
| |
|
Artificial Intelligence → Discussion