Neural Nets

Post Reply New Topic

Sorry, no real world personal data to be disclosed here.



Such concepts already exist. What you talk about reminds of me of protein classification using decision tree learning. Sample data are primary structure represenations of proteins, becaue those string data are easy to handle. Regular patterns are used on the nodes of the decision tree. The learning process selects the most succesful pattern, which is the pattern that comes with the best information gain value (check information theory for information gain and entropy).



Current computer vision projects use multilayer neuronal networks and variations of gradient descent learning. This is an iterative algorithm that is based on a multidimensional function that calculates an error value. The idea is to follow the direction to the lowest error value. You can imagine a landscape of mountains, you put a ball somewhere and the ball starts moving towards lower altitude. To avoid staying with a local minimum there is a variation called momentum gradient descent learning.



Iterative machine learning schemes have got a working memory: The values computed during the previous iteration.



You are right: People already have tried that stuff and current projects involve the cooperation of many scientists.

May be you already have studied a lot of literature but I would like to recommend two books to you:

"Machine Learning" by Tom Mitchell
"Artificial Intelligence - A modern approach" by Russel and Norvig

The second one is a good reference, it quite up-to-date and complete.

Thanks for the tips, I'll take a look at those books.

I don't professes to be the greatest expert on AI (can you tell?). However AI is in it infancy so any so called 'experts' need to be humbled, I think just like anyone. I do know about programming and embedded/mechatronic systems.

It seems to me in your concept just going by the interfaces seems way too complicated. How long have you got to do this project and what would be considered 'complete'?

I think the principle of KISS applies (Keep It Simple Stupid). Simplify your system as much as possible to reach your objective. It is not the amount of gear, theories, animatronic realism that is important. It doesn't have to be sexy. It is the conclusions that are important. If you waste time on pointless complexity you might not draw any or it will be unreliable. You can always build on something simple. Also if you explain your dedication to this principle some people might be more likely to give you funding.

You get over yourself

Machines are mortal just like us. I suppose you will make the machine that destroys us all. Well it will be your fault. You could say there is some ambiguity there in this hypothetical but really you always have the option to limit scope of the application.

It is bad publicity-seeking science fiction that powers people's fear of AI. There is that ridiculous "Frankenstein" stuff and there are movies like "I, Robot" or "Matrix" telling people there would be gigantic risks involved. Drawing conclusions from comics or movies is not really a smart move.

The truth is entirely different. Of course there never was a Frankenstein and anybody like that person would be expelled from any research project before he could say "Hello". Scientists are no maniacs, most of them are hard working, talented and gifted people who have chosen a career in science because they want to contribute to society and the well-being of all people.

What people fear and have always feared is the unknown. The usual stuff around artificial intelligence is incomprehensible to most people and of course people fear what they do not understand. Finally they turn their back to science and seriously begin to believe in "God created the world in seven days" again. Ups, beginnig to ramble going off-topic.

There are some risks if we talk about employment opportunities. Many AI projects are finanzed or sponsored by the industry and the most obvious reason is that having computers doing things people did in the past is cheaper, faster and more reliable. The downside is obvious as well: Jobs get killed that way. If there is anything to worry about than it is the social aspect and how we can deal with that.

Machines becoming the dominant species on this planet is the last thing I worry about. Machines are no species and never will be.

I'm hot stuff though.



One step is the control logic (chips) would have to be made external. You can be sure some where in our Universe biological organism spread over an entire galaxy are fighting a restlessness battle of endless self-replicating machines attacks. Like a virus. Some think our first detection of other intelligent life will be observing the mega scale engineering projects of such self-replicating machines.

http://en.wikipedia.org/wiki/Dyson_sphere



I submit they have already in a number of locations in the Universe and are trying to explore the entire Universe in the same way a bacteria tries to explore an entire jug of milk its in.

I see, there is more science fiction coming this way. I really do like science fiction, but please, we should not mistake it for reality. Many things are possible, winning the lottery is one and earth getting hit by a huge meteor is another one.

The odds of unlikely events approach 1 in 1 in an infinite time interval.

There is no infinite time interval. There are no infinities in nature at all. The concept of infinity is a result of necessary simplification in our mathematicsl models of what we call reality.

The problem is it seems like a likely event to me. Your take on this reminds me of Bill Gates' book in the mid 90's in which he almost completely forgot to mention the internet and had to rush a second addition out to cover it. Presumably he was very well informed on short term technology trends. I think your being short sighted to and either can not or refuse to see the possibles. Because AI is lame today, because a chip production plant is the size of a football field and because most machines need some human assistant in there production your assuming that all those things will remain constant. They want.

Yes, this is true, while it may seem like more fun to tackle everything at once for now I'm going to start with recognition of simple shapes, I've decided.

Given the fact you are accusing me of being short-sighted I want to tell you that your ideas are taken from the science fiction works of others. The authors of these works most probably just wanted to entertain people by demonstrating those ideas, they did not claim those things would be a likely part of future reality. Your willingness for accepting those things for real events to happen in the future is your personal business, I do not mind it. Your way of judging my way to see those things speaks for itself.

I did not see the capabilities of AI as limited in any way but there is not much more to self replicating machines or dyson-spheres than the notion they are appealing or fascinating ideas. There is no indication whatsoever that those things are linked to any trend in technology today. The simple notion that those things are possible is not enough to talk them seriously.

Are these things possible? Yes, they are. Are these things likely? No, they are not. From what we see in current technology today we can predict a probability of 0.5 what means nothing more that we do not have a damn clue what to think about that. Believing in these things is just exactly that: Believing. Do it and be happy.

Your analogy is wrong. We had network technology when Bill Gates wrote his book, a book that is not containing much more than Microsoft's and Bill's personal marketing gossip, btw. But we neither have the means nor the needs to build self replicating machines.

http://www.islandone.org/MMSG/aasm/

The need will be to solve the shortages for man.

Good.

Ok, you score. That is a good one. Perhaps there IS more to that. But it is just a demonstration of possiblity and it is a what-if-scenario. This is not MUCH more than science fiction, actually just a little bit. Scientists can have dreams and ideas, that is part of what they are scientists for. I am sure you can find more research papers of those things. This just demonstrates other people have thoughts too. Institutions sometimes put up especially interesting programs for many reasons, including to attract attention by media, sponsors and young people they want to get interested in what they do. There are many think-tank facilities investigating what is possible or might be possible in the future.

Please, there is no need to continue that discussion as if it would be a fight. I am much more closely linked to current real world research, I need to care for what is possible today and I have a natural tendency to see a million little problems where others see fascinating ideas. Perhaps you can afford to dream a 250 years into the future, I do not share that luxury.

Sure that is what I've always professed and is why I said 'hypothetical'. I was playing devil’s advocate.

About the social side I see your point but on the other hand it could make people more productive, less stressed, by making work easier. Giving people access to more and more knowledge/fields goes hand in hand with this otherwise it would be counter productive because they would not be challenged. You do need fluidity; you don't want the equivalent of Mao's Great Famine from putting small industrial units in every village so there was nobody to grow any food. So any kinds of diseconomies need to be carefully monitored. Sure it is easy to think that some machine might make your life easier in isolation. But if there are not enough machines or access this might cause an over reliance causing under production based on hype.

My hope in the future that there would no 'careers'. I’ve always thought that is a bad idea. It is anti motivation, it doesn't make the most of people’s skills and that is unproductive. Think about it how many people go to university do what they are trained to for a couple years tops then get promoted to ‘management’. People have no incentive to do what they are trained for, besides they have many more skills that can be useful else ware, but they are largely locked in. Very few fields such as research actually encourage people to uses their skills and promote it almost to the highest level, then again you still get people faking work, meh. Then politicians have the audacity to talk of skills shortages. It is not skills shortages it is lack of people willing to do what they are trained/interested in because there is no incentive. I don’t think every student is interested/dedicated in their field than say and aspie is to their obsessions, they are doing for career reasons not the interest. I’m a self taught programmer you can see the crap that comes out of some graduates. This is why we need to move away from misnomers like ‘career prospects’. Less permanent employment and more doing what you are interested in and good at, will make people happier and more productive. We can start to move in that direction ant a feasible rate.

On a purely selfish level I would like task based AI to help me with my executive dysfunction. If I could do that and it was a genuine aid unlike my PDA, for example, that still relies on some degree of organisation and awareness that I don’t have. You may think a PDA is to organise, but you still need some basic executive/planning/sense of time, etc for it to be of use to you. It might be something others might take for granted. Even if it did work that doesn’t mean I can get enough done to live on. Anyway if I could have that I would consider it the same as a cure. I don’t need to feel I’m smart, just the capability to explore my curiosities. But AI probably would make people more capable so long as it is in line with people needs and not work against them.