Artificial Intelligence

tuXthEhxR 12 years ago

As you stated you were kind of looking for a debate upon the subject, I thought I would post a comment.

I think I would have to fully disagree with your first bridge. I believe the first bridge isn't necessarily the gap between Artificial Intelligence and human cognition, but the gap between any form of intelligence and human cognition. These hormones are emotions, at the basic level. Emotional decisions are not known to be intelligent decisions; so I would argue in moving towards "Smart Computers", one would move towards preventing this rather then encouraging it.

The middle bridge I fully agree with, and believe will be the biggest hurdle in forming actual AI.

What you consider to be the last bridge kind of baffles me.. I don't know if it's just I don't understand where you are going with it, or if its just the fact that I disagree with it because I am not a religious person. Either way I believe your last bridge deserves a little more clarification.

Your article really got me thinking, not what I expected from the title. Very interesting article.

suid 12 years ago

I have been recently doing a very large amount of AI research constructing neural networks and genetic algorithms; therefore, I suppose I will give my opinions. I think I agree with the previous comment on the first bridge; although, I feel like this bridge is an appendage of the second. The ability to decide what to do when faced with a certain situation is a part of learning and reasoning. The actions taken are simply determined by the output of whatever algorithm used in decision making, whether biological or mechanical. Maybe I just don't understand the point you're trying to explain. I believe the second "bridge" is the most vital to AI development. There has been a lot of progress in this aspect of AI and it is fascinating to research. Learning how to create a learning algorithm is very enlightening. Almost ironic. There has been many surprising results, which cause speculation on the future. Experiments done with GA's astound me with what success comes from them, providing intelligent and elegant solutions even the programmer cannot sometimes comprehend after much investigation. The difficulty is that there isn't an algorithm that works for any given problem, some are better than others in different situations. There isn't a program we can load into a computer which determines all decisions intelligently, not even close. Humans on the other hand seem to have the ability to learn anything using the same "laws" governing the mind. We know how to make a computer reason, we know how to make it learn, but that is only taking our own intelligence and representing the mechanism in code; emulation if you will. There is still much R&D to be done in this field before we get close to any science fiction "singularity" or an agent which is self-aware and conscious with an identity all its own. (good luck with that :) )

In your third point, I will also agree with the previous poster, what the heck are you saying exactly? I get the impression you are trying to abbreviate the topic I briefly mentioned above of consciousness, or making a computer "wake up." Hitting 'Execute' and having your program become self aware, or even the eventual capability of such. This seems like science fiction still to me, some sort of humanoid robot which yearns to continue, is terrified of annihilation or ceasing to be. I guess this could be because AI could still be considered in its infancy, depending on how you view the progress made thus far. Well those are some of my own comments. They may seem relatively negative, but that is probably because I see AI from more of a realistic CS standpoint rather than more of a philosophical, speculative one. While it is healthy to speculate on this topic, questioning leads to relevant experimentation. Rated 'Good'.

Osculum Umbra 12 years ago

For the first bridge I was shooting for the uncontrollable programming in our bodies. When you body pushes out adrenalin you will react a certain way. It is not really AI per say but more hard coded programming which is required for the human body to adapt to certain situations. Such as a computer turning off when it over heats. I would also have to disagree with one point from tuXthEhxR, hormones are not emotions but rather the key to them depending on which hormone is released. If you claim that hormones ARE emotions then you are saying you can bottle them and sell them. True, you could bottle hormones and sell it with its corresponding emotion as its name but the emotion its self is the brain reacting to it and returning with physical changes not the liquid in the bottle.

The Third bridge is the part of us that asks why for the most part. You can program a computer to eat when it is hungry or sleep when it is tired but what is it that makes us wish to know about the unknown. This is different from the second bridge because nothing set it in motion, it is just its own will to understand something. It is the part of us that cares at all about making AI or reconstructing the big bang or finding if there is a god out there. Yes @ suid it could also be the consciousness, or the soul as some may say. You could argue that this is a form of self preservation if you go as far to believe that we understand at a subconscious level that understanding the unknown removes fear of it. By the "Mind of God" I was referring to the unknowns of the universe, whether you believe it to be a supreme conscious or a cataclysmic anomaly that spiraled into what we have now.

tuXthEhxR 12 years ago

The more I think about this subject the more interested I get. Do one of you guys know a good book on the subject, perhaps more directed towards the more novice aspects? Or perhaps a website that has a steady stream of information about the subject? I would like to learn more about the subject, very intriguing..

Osculum Umbra 12 years ago

First I am not sure how novice you are but from your post you seem to have a good grasp on the Idea. If you have taken any entry level college psychology classes then you should continue with that . If you wish to recreate something then its best to know how the original works. If you have not taken any classes like that then I suggest you start there and get your self a good foundation into understanding the actual human mind. Now to try and point you in a real direction I can not give you my input on any good books since I have not read any. I just have some background into psychology which is why I said start there lol. If you wish to keep up with current advances in AI you should look into the “Turing Test” competitions.

suid 12 years ago

I know that I am more experienced in actually trying to create AI by modeling algorithms and toying with programming exercises to see how it really works. The psychology is also interesting, which seems to be more of what Osculum Umbra has looked into. For more of a CS perspective on Cognitive Computing (or AI), I could suggest the text we used for one AI course I took, the title: Artificial Intelligence: A Modern Approach (2nd Edition) by Russell and Norvig. It is a very good read and contains explanations of several AI techniques, but does not really delve into philosophy, psychology or what AI really means. It's more about applying it to useful situations to make computers better. Examples include neural networks, genetic algorithms, non-heuristic searches such as alpha-beta pruning of min/max, support vector machines, natural language processing, voice recognition, etc.

Also, just googling around the web can lead you to places where you can learn a great deal on aspects of AI and approaches others have made in understanding what it is and if it is possible.

tuXthEhxR 12 years ago

I've taken a PSY101 course, and know the fundamentals that were taught, but it was a course that I considered more mute then others; hence I didn't exactly apply myself 100%. At the time I wasn't looking to major in Computer Science, so my I wasn't exactly thinking down that road, wish I had been. Just got done downloading the book you reccommended; thank god for thepiratebay, saved me almost a $100. Sounds like exactly what I was looking for, and a thousand pages should be able to eat up most of my spare time throughout this coming semester. The code that you uploaded into the code bank looks very intriguing, suid. What would a practical use for the code be? What exactly does it do? I couldn't really tell from the description that you gave about it inside the source code, and I have only been coding for about 6-8 months now, so it kind of surpasses my abilities. Thanks to both of you for the pointing me in a better direction to learn about this subject.

suid 12 years ago

I wouldn't hold to high of expectations on my Python code that I uploaded, but it was what I did during an AI course I took. We were on the topic of artificial neural networks at the time, which I already had an interest in. The Python code took forever to train, so I started coding the nets in C (was much faster and I prefer C anyway). If you haven't looked into neural networks before and you are interested in AI, then I suggest you do some research on them. They are a primary technique of machine learning. The basic unit in the network is called a neuron, and the concept is inspired by the biological neural network that comprises the architecture of the human brain. Neural networks are composed of layers of neurons which are usually connected to other layers. The first layer is called the input layer. This is where input to the network is introduced and the input must be encoded into numeric data, most often binary digits. Each input digit is assigned to an input node. Now, just like in the brain, there are synapses connecting neurons together in an artificial neural net as well. These are each assigned a weight value to represent how much the connecting neuron influences the overall result of the network. Typically, neural nets use a fully connected feed forward architecture, which means that layers propagate their output only to layers after them, or "feed" the next layers, and also that each neuron in a layer connects to every neuron in the next layer. The actual propagation of input from one layer of neurons to a neuron in the next layer is calculated by taking the weighted sum of the first layer and applying what is called an activation function to the result, which determines whether or not the current neuron will fire (output a one), if the activation function says the neuron does not fire, the neuron will output a zero and this becomes the input into the next layer, and so on. This is generally how the human brain works, but for most learning algorithms it is more useful to not use this type of threshold activation function which outputs one or zero. One more thing to understand is how this is actually useful. In order for a neural network to learn anything, a process called training is done. This is where the programmer prepares a set of examples called the training set to be a sequence of inputs for the network to process. Within each training set example is normally a value which corresponds to the true value or 'answer' which the network is supposed to output. After propagating the input of one example, the network computes how wrong it is and uses this value to 'update' the weights connecting each neuron together and it does this in a way that it will minimize how wrong it is until the margin of error is within an acceptable degree. There are several algorithms for doing this but the most common one is called back propagation. Very trivial algorithms are used for a single layer neural network which is known as a Perceptron (cool sounding I know) but it gets more complicated when you start adding more layers. The great benefit of neural networks is that, when they are properly trained on a good set of examples, the properties that they possess allow them to 'generalize' on the data they have previously seen. Therefore, if you give the network input it has never seen before, it will make a very educated guess as if it 'knows' what it sees. So for instance, if you train a neural network to learn what the letter 'A' looks like by using patterns of varying pixels which all resemble the letter 'A', it will be able to generalize very well what is or is not an 'A'. The code I uploaded in the C section of the code bank is a neural network which learns the simple function of eXclusive-OR on 2 bits. It was just a toy project I used in order to understand how back propagation and neural nets in general work. You can read more about the in depth details online, but I hope my explanation can at least lead you to informed Google searches.

ellipsis 12 years ago

This was an excellent read. We are primitive gods attempting to understand ourselves better by building better AI, but eventually…someday…we will want to use this knowledge to create biological AI, AI which can reproduce and which is built in our image. And when we achieve this feat, we will have become like the gods we believe created us, thus repeating an infinite loop of creation. I love to get all philosophical, but this was seriously excellent, so you get an "Awesome!" rating from me. Good luck with your endeavors, fellow godmind.