Cheap memory?

Discussion in 'Gigabyte' started by Scott T. Jensen, May 18, 2004.

  1. Here's the best I can do for that. I've run this past my lead programmer
    and he thinks it's fine. He also thinks it's hard to know what we'll need
    until we try to implement it. Given that...

    Processor: It doesn't really matter. Faster would be nicer, but it isn't
    crucial to the AI concept. I plan to get the fastest one at the price point
    just before it spikes up because it is close to the fastest one on the
    market. However, if this will enable the RAM to be larger, then I'd be
    willing to put money into it to get one that can handle the most RAM.

    RAM: As much as possible for as little as possible.

    Hard-drive: As much as possible for as little as possible.

    Video card: Doesn't really matter. Need one for the human interface but
    practically anything will do.

    Parallel processing isn't needed.

    Cannot be done over the internet. Neither distributed computing or off-site
    hard-drive (memory leasing). The reasons are many. Slows down process too
    much; if net connection goes down, AI goes down; susceptible to viruses
    (especially if done distributed computing style); security; and other
    reasons ... more of which I'm sure you can come up with yourself.

    Can be done in a cluster style ... a LAN of computers. However, would only
    do this if there was a huge cost savings.

    Budget: Since this computer will only be used for the AI project, there's no
    guarantee that the AI project will succeed, and I don't see any other use
    I'd have for it otherwise, I want to spend as little as possible. Would
    like that to be under $2,000, but would consider spending as much as $10,000
    .... if there's a really good reason to go above $2,000.

    Scott Jensen
     
    Scott T. Jensen, May 23, 2004
    #21
    1. Advertisements

  2. Scott T. Jensen

    Curt Welch Guest

    Do you have any intest in talking about your AI idea? It might save you
    $2000 or make you think about some important issues you have overlooked.
    I'd be happy to chat off line if you would like.

    Are you trying to solve the general AI problem of making a machine
    intelligent, or are you just working on some AI application which you think
    will have potential value?
     
    Curt Welch, May 23, 2004
    #22
    1. Advertisements

  3. Right now, I'm researching what has already been done in the AI field to see
    if my idea has been tried and proven wrong. I've read "Mind Matters:
    exploring the world of artificial intelligence" by James Hogan, read
    numerous web articles on AI, and am currently reading "Artificial
    Intelligence: A Modern Approach" (2nd Ed.) by Stuart Russell and Peter
    Norvig. From what I've read so far in Russell's book (as of this post, I'm
    on page 73), it doesn't look like I will find my idea in his tome. In
    reading his history of AI, it would seem AI researchers haven't yet thought
    of exploring the approach that I believe will lead to AI. However, I'll try
    to finish reading it to be sure ... though I'm not math-friendly and that
    handicap has now increasingly slowed down my rate of progress through the
    book.

    If, after reading Russell's book, I still haven't come across my AI idea,
    I'll begin searching AAAI's archives for it. Focusing first on what has
    been published since the release of second edition of Russell's book and
    then, if that proves unfruitful, hunting through the rest of the archive for
    it. However, I'm not looking forward to paying the hefty AAAI membership
    fee to do so and am hoping to find a friend, a friend of a friend, or
    someone local that is an AAAI member who will let me use their account to
    access the archives, such as a kind-hearted local professor at University of
    Wisconsin - Madison, Edgewood College, or Madison Area Technical College ...
    as I live in Madison, Wisconsin, USA.

    If I do not find my AI concept in the AAAI archives, I will try to find an
    AI professor or researcher within reasonable driving distance of where I
    live that feels s/he has enough knowledge of what has been tried in the AI
    field to be a good sounding board for my idea ... and one that I feel I can
    trust not to run off with it. For you see, I've grown rather fond of my
    little idea and would like to be the one that brings it to life. Yes,
    that's a bit selfish on my part, but I never said I wasn't human. ;-)

    Currently while I do my reading, I'm mentally operating under the assumption
    that my AI idea has been looked into thus I am just trying to find out when
    it has been done and why it failed or why the basic concept was shown to be
    faulty. My idea seems obvious to me and thus I assume it would be to others
    .... or at least one AI researcher somewhere at sometime in the past sixty
    years. However...

    I have three close friends who are experienced professional programmers and
    who say they've kept up on the AI field. I've presented my AI idea to each.
    After a rather thorough grilling of me by them, they've stated that they
    haven't heard of anyone taking my approach and find it interesting. We've
    even had a few evening sessions where they've started to chart out what
    would be needed for the AI program. Software, not hardware. My take on how
    to bring about my AI is that we just need to tweak several already-existing
    computer programs and get them to work together to get them to do what I
    want them to do. I am quite comfortable at this stage with any
    inefficiencies this might allow. However, at least one of the components
    for my AI my friends feel will need to be built from scratch ... if not all
    the components, as one of the three maintains. These programmer friends are
    the programmers I've been referring to in this thread and who are willing to
    help me do this project. That and they want me to just tell them what each
    component is to do and let them figure out how they'll get that component to
    do that for me. They keep telling me not to be concerned about what goes on
    inside the "black box" as my contribution is setting up the rules for how
    the AI is to think once all the components are set to go. Personally, I
    feel like the captain of a ship with a loyal crew that wants to go wherever
    I want to go, but none of which want me to help out in the boiler room to
    get the ship underway ... and tend to get upset everytime I poke my head
    into the room to see how they're doing. *laugh* However, before I ask them
    to volunteer their time any further, I've decided to try again to find if my
    AI idea has already been attempted and proven wrong.

    Just recently, I've emailed a few AI professors at MIT, Stanford, and
    Carnegie-Mellon University. I selected these universities as they have
    well-known AI labs. I read each college's AI department/lab's faculty
    homepages and tried to find the faculty members whose expertise and/or area
    of current research is closest to what I feel my AI concept would call home.
    I then emailed them asking for reading recommendations that would help with
    my search for my AI idea. The professors at MIT and Stanford never replied
    back. Fortunately, the two professors I contacted at CMU did. One
    suggested that I read Russell's book and the other suggested I dig through
    the AAAI archives as outlined above. Both recommended that if after doing
    their suggestions, I still don't find my idea, that I present it to an AI
    professor or researcher (that I trust) to see if they've ever heard of it
    being attempted. One of the two CMU AI professors then said that if that AI
    expert says it hasn't been attempted that I should go ahead and "just
    implement the thing and show us all." The other professor said that if the
    AI expert said my idea hasn't already been attempted and holds promise that
    I should pursue a doctorate in AI.

    Saying my search gets me this far, I'm not sure I'd pursue a doctorate in
    AI. From my readings of AI graduate programs, they require a lot of math
    and computer languages. Neither of which interests me and both of which I
    know I'd struggle through due to that lack of interest ... if not -- in the
    attempt -- die of numbness of the mind. I'm a psychologist by nature
    (possessing a humble little BA in psychology with a minor in marketing ...
    and having worked for many years now as a marketing consultant and am now in
    the process of launching my own talk show, www.scottjensenshow.com) and
    barely know how to turn on my computer. That and my programmer friends
    understand my limitations and are willing to provide the technical knowledge
    and expertise to see what develops by attempting to bring about my AI idea.
    My friends want us to do it along the lines of how the Wright Brothers
    brought about the age of aviation. However...

    Working with my friends to bring about my AI will mean the construction of
    the AI will take a long and sporadic time since they'll be working on it in
    their free time ... when they have free time. Whereas working in an AI lab
    with dedicated researchers helping me would speed up the timetable
    significantly. However, I'd only do that if I could get a full scholarship
    (a.k.a. "full ride") to pursue a doctorate in AI. Then there's the issue
    that I'm be a "bit" old for a graduate student at the age of 40. However,
    still being single I think I could handle the social ramifications of that
    .... especially if I could just come in and work on my idea with the help of
    an AI professor and a couple of her/his lab's talented computer science
    graduate students. That and not have to take any college courses in
    computer science and math ... though I'd be up for attending some
    graduate-level psychology courses. If I was able to get all that, I would
    then not need to buy a computer for this project since I would then be
    allowed use one of that university's supercomputers. Yes, yes, I know this
    paragraph is just wishful daydreaming. Anyway...

    I am looking at this search for my AI concept as merely an intellectual
    exercise and a nice way to motivate myself to educate myself further on this
    topic. If I learn that my AI idea has been attempted and proven wrong, I
    will be a bit down but not terribly so. First, I am still expecting to find
    out that it has already been done and tossed aside. Second, I would very
    much want to know why it is a bad idea. And, third, I feel the mere pursuit
    of knowledge has a value in and of itself.

    And while I do the above research, I've decided to address the hardware
    aspect of this project and thus the reason for this thread.
    I appreciate the offer. If you would email me your credentials, I will look
    forward to receiving them. Also, are you within reasonable driving distance
    of Madison, WI, USA? I'd prefer to meet face to face. That and I'd like
    you to sign a confidentiality agreement before I tell you my idea.
    I'm attempting to do the first one.

    Scott Jensen
     
    Scott T. Jensen, May 24, 2004
    #23
  4. The following is what someone, who wishes to remain anonymous but willing to
    allow me to post what they sent to get input from others on it, emailed me
    on the above:

    "Now, price seems to be the limiting factor on your project, that's sane.

    Since you have a rough upper bound on the money to spend on the project it
    seems that more ram means less hard drive, and vice versa. Fortunately you
    can often use some of one of these for some of the other. But hard drives
    are about 60 cents/gigabyte while ram is about 160 dollars/gigabyte. So I'd
    suggest:

    $30 power supply
    $70 Fry's special sale ECS motherboard&AMD 2+Ghz processor
    $30 fan (or they run specials with board&2.5Ghz proc&fan for $100)
    $80 512 meg memory
    $480 4x200 gigabyte hard drives
    $5 cheap pci video card
    $5 cheap mouse
    $5 cheap keyboard
    $0 scrounged cabinet to put all this in
    $0 scrounged free operating system
    $0 scrounged network stuff to tie these together

    That's about $700. Buy one now to get your programmer busy. Getting the
    first one going will be lots more work than the rest. Every time he starts
    to run out of room buy him another one until you have succeeded or concluded
    the project won't work or have run out of money.

    Then if the project didn't work take all the drives out of these, do scans
    of each drive to ensure that they are all fine and sell them for a plausible
    fraction of what you paid. Clean tested drives that size are easy to sell
    and ship and even a year from now you will be able to sell these without
    trouble. (Save the boxes they came in for shipping)

    My testing shows there is only a 20% increase in performance when going from
    the AMD 2000 to the AMD 3200 for cpu intensive number crunching, the AMD
    2600 Barton looks like the place to be right now. Buy big loud room fans to
    blast air against the motherboard/heatsink/cpu fan, these run hot and you
    don't want them hot.

    So for your $2000 budget you get 2.4 terrabytes with three fairly fast cpu's
    to access it and share the processing load, 100 megabit networking to
    communicate is built into the motherboard and fairly zippy. Going to
    gigabit networking is only a few dollars more and doesn't give vast
    increases in performance but you might notice. At the very top end of your
    budget you might have 12 terrabytes and 15 cpus. And for any $ figure you
    choose you could cut your storage in half if you wanted to triple your
    memory, for the same price.

    I don't think you can buy used servers that will beat these specs and price,
    but you could certainly poke around and see. There have been lots of
    bankruptcies in the internet business and you might stumble on a bargain if
    you hunt long enough and find where to look."

    They then quickly sent me the following add-on note:

    "Something I did think about this afternoon. The proposal has some degree
    of flexibility in it.

    You could, for example, only drop in additional hard drives as the
    programmer filled them up. That would help keep costs in line and be a
    monitor on progress. Likewise for extra memory sticks.

    You could also make another change, you could jam more into a single case.
    Instead of using three cases to hold the memory and drives you could
    incrementally bump up memory half a gigabyte at a time and drop in pci hard
    drive controller cards and drives. That saves buying more motherboards,
    processors, cases, power supplies and some network hardware. But really the
    extra cost of drive controller cards, more expensive cases and power
    supplies looks like the overall cost will still be just about exactly the
    same. And you won't get the extra processing power. If you could find a
    case and power supply that would handle it, without paying a fortune for it,
    you could get up to 1.5 gigs of memory fairly easily and without paying the
    outrageous 1 gig/stick memory prices, and perhaps cram up to perhaps 4
    terrabytes of drives into a single huge server case."

    Scott Jensen
     
    Scott T. Jensen, May 24, 2004
    #24
  5. Scott T. Jensen

    Curt Welch Guest

    A very common prolem in AI is that people think they have come up with a
    new idea, but after enougth study, it becomes obvious is just a different
    way of looking at the something which has already been will studied. You
    don't recognize it as being old, becaue you are just using different words
    to talk about it.

    However, most old ideas have never been proven faulty. They have just not
    yielded the type of fruit which was hoped for. That could be because the
    idea is just a very weak one to begin with, or it could be that the idea
    was good, but the old approach was just lacking some important twist, or
    new understanding.

    So, no matter where you idea my lie in the range of everything AI, if you
    have a new way to look at an old problem, you could still discover
    something significant and new.

    The important thing for you to do is exactly what you are doing. Studying
    all the old work on AI and you can understand for yourself how your ideas
    are similar, and different, from what has been done in the past.
    Everybody has a different perspective and if you just happen to have the
    right one, you will uncover something new.
    What I find is that there is a huge difference bettwen people that have
    been "keeping up" with AI, vs trying to build, AI. You can't really
    appreciate AI until you have had the ideas, like you are currently having,
    built them, found out they didn't work, tried for years to make them work,
    throw them away, come up with new ideas, try for years to make those work,
    throw those away, repeat until you have AI.

    Very few people are doing this in the world. Many programers like your
    friends have been exposed to past work, but I would be willing to bet that
    none of them ever once came up with their own idea about how to solve AI,
    and tried to build it.

    A huge percentage of people working in AI have long since given up on (or
    never even tried) solving the general AI problem, and are instead, working
    on mastering some intersting sub-domain problem. Many who have made AI
    their work, or even forced to do this, because if you want to continue to
    get paid, you better be producing results. And if the only problem you are
    working on is the "big one", and you have produced AI, it's going to be
    hard to get funding. So they identify sub-domain problems that seem
    important to the big problem, and make good progress studying the
    sub-doman.
    The one thing that makes me doubt your approach without even knowing what
    it is, is the fact that you do not seem to be a strong programmer (or did
    you say you don't program at all?). Solving AI is an engineering problem.
    It's a very hard engineering problem. You need very talanted and sharp
    engineers to find solutions which nobody else has every been able to find
    after 50 years of looking at the problem.

    Also, people who are not engineers, or scientist famillar with behavior
    research, tend to be easily fooled by human behavior. They just have no
    clue what intelligece really is. This is the problem we all face by being
    "too close" to the problem. We think we know that intelligece is just
    because we think "we" are intelligent. The truth is, we aren't
    intelligent. Our brains are. And even though we think we know who "we"
    are, we don't. It's all a very confusion thing to your arms around.

    AI projects fail not because the solution was mis-founded, but because the
    problem was mis-defined. They didn't understand what intelligence is.

    And that contines to be the real problem of AI - finding the right
    definition of what intelligece is.

    So the real question you need to ask yourself, is not do you have the
    correct solution, but do you have the correct defintion of the problem.
    Programmers like to program. They don't like to waste time dealing with
    somebody elses code. They will always push you to let them create new
    code. That's just what we do. :)
    The black box issue is key. Because that's the defintion of the problem.
    You have to describe the problem of AI in terms of what the black box is
    going to do. You have to take the extensional stance, when you define the
    problem. And, as I said above, getting the correct defintion of the
    problem is the AI problem.

    If you get the black box definition wrong, then it makes no difference what
    so every what you choose to put in the box. If you have framed the AI
    problem incorrectly, you have no hope of solving it.

    And that takes us to the next point. You wrote "how the AI is to think".

    By saying those magic words, you have already framed your problem. You
    have already made a decision that the black box must think. Or your
    programmers have made that decision. And thinking about AI as a "thinking
    machine" has long been shown to be the totaly wrong way to frame the AI
    problem. This gets back to what I said about how we, as humans, are too
    close to the subject of the problem (humans) to see it for what it really
    is. We think we have free will. We think we "think". We4 think we use
    logic to make decisions. We think we have goals. We think too much and
    understand too little at times.

    What you have to answer before you waste any time or money, talking about
    what you are going to put inside the black box, is what intelligence really
    us. For any black box approach, you have to describe how the box
    interfaces with the world, and you have to describe what the purpose of the
    stuff inside the black box is. Not in terms of what you put inside the
    box, but in pure extensional terms of what the black box is going to do.
    And, ignoreing what will need to be inside the box, you have to explain to
    yourself, why, the black box, as you have defined it, is "intellgent", and
    not just some typical computer doing typcial computer things.

    Once you are sure you have the right defintion of intelligence in form of a
    black box description, then you can go about trying to figure out how to
    build the inside of the box.
    Ah, your psych background might help you understand us for what we are - if
    you understand what people like Skinner were saying. That might be even
    better than good engineering skills. If you can turn your knowledge into a
    good black box description of what it is to be intelligence, then you can
    let the programmers try to figure out what to put inside the box. If you
    don't have the skills to create a good extensional black box specification,
    then you will have problems getting your programmers to do anything
    productive - unless they have the interest and skills to turn your ideas
    into a good black box specification.
    Maybe people have become addicted to the dream of AI. I'm one of them.
    There are many others. Most are considered "kooks" because the AI problem
    has a bad habit of looking easy. But, the mistake most the "kooks" make,
    is not understanding the importance of the getting the defination of
    "intelligence" correct. They all assume that "know" what intelligence is
    just because they are a human, and start to "solve" the problem based on
    the fact that they obviously know what it is to be intelligent, so they
    understand the problem.

    AI is hard, because no one understand the problem. That's why it's so damn
    hard to find the solution. We don't know what the problem is - past the
    far too vague idea of "being smart".
    Trying to solve AI is the best way to educate yourself on the problem.
    I don't have any "credentials". :) I'm just an old fart like you addicted
    to the dream of solving AI. I work on as a hobby on the side. I've been
    doing it on and off for close to 25 yearas.

    I'm a professional programmer by trade. I founded and currently run a
    small dot com business for a living.

    I waste far too much time on AI, but like I said, it's an addiction.
    I'm in the Washington DC area. If you are in the area and what to get
    together to chat, feel free to look me up. You can find contact
    information on my web page: http://curtwelch.com/
    I have no problem with that.

    I think I know how to solve AI. All good kooks do. And we tend to believe
    that all other kooks ideas are crazy becasuse they are not the same as
    ours. If we get together and talk, what is most likely to happen is I will
    spend an hour trying to explain to you why your framing of the defintion of
    intelligence is wrong (unless you happened to frame it the way I did).
    But, I can also give you a lot to think about in terms of why or why not
    your approach may or may not work.

    I am not well read on the AI literature, so I can not give you good
    pointers to other work that might be similar to yours. The contacts you
    are making in the AI field should be good for that.

    I talk about my ideas in comp.ai.philosophy which is where I'm posting
    from. So you can read some of my posts there and get a good flavor of my
    approach (learning machines). I don't have any good summary of my work on
    the web I can point your do. Sorry.
    That's the fun one to work on. But it's also the one that makes people
    know you are a "kook" without even knowing what you are doing.
     
    Curt Welch, May 24, 2004
    #25
  6. The main thing I disagree with that poster on is going with the Barton
    core. If anything, the Barton is overrated. You get more bang/buck
    with Thoroughbred B. Or, at least, you did about 6 months ago. But
    at the least, the Barton did not live up to the top billing.

    The other observation is that this hardware talk is completely
    irrelevant. When you said you are a psychologist, I realized why
    you are worried about hardware. The hardware is the bicycle shed.
    It's something that's easy to talk about.

    I agree with almost all of what Curt said except for the part where you
    should read up on the state of the art. No need to do that. You aren't
    even to the point where you could appreciate the state of the art. The
    idea that you could build AI without so much as knowing a single
    computer language is fairly amusing. But really, you don't need to
    know a computer language to build an AI. You do, however, need to
    understand how a computer operates to appreciate the difficulty of
    your project, so I will make the following suggestion.

    Pick a problem within the solution domain offered by your hypothetical
    AI. For instance, solving a syllogism. Draw a diagram of how your
    AI will work, and create a series of "snapshots" on paper that show
    how information flows through your system to solve the problem.
    Start out at a high level of description, and then increase the detail
    until you think you could get 5 year olds to act out your virtual AI. If
    you can't trace out the execution of even a simple problem, then I can
    pretty much guarantee that your idea has no merit. If you *can* trace
    the execution, then show it to your programmers, and ask them how
    hard it would be to turn it into code.

    Odds are, they will either code it, and you will have a primitive portion
    of your AI to look at, or they will point out where you haven't specified
    enough detail. My bet is on the latter. What you will find is the
    incredible pedantry of a CPU. You will realize that many things you
    take for granted require elaborate specification and detail. But in the
    rare chance that you actually manage to solve a trivial problem with
    your idea, take it to the next level. Pick a harder problem from your
    solution domain, and try the same thing.

    At some point, you will realize that you don't have the expertise to
    draw out the solution in detail. At that point, you might be motivated
    to learn from the masters, including math, programming, and AI.
    Then reading will bring some benefit. But right now, reading is like
    studying Greek literature without knowing Greek. Even if you can
    pick out a few words, you won't appreciate what they mean without
    context. And that context only comes by doing.

    It's fairly easy to come up with solutions where the building blocks
    are hand-waving and hemming and hawing. Anyone can do that,
    including dilettantes. It's much harder to put your money where
    your mouth is and turn that vaporware into a product. Don't spend
    a penny on an expensive computer. Solve a simpler problem on
    existing PCs, and teach yourself something about how hard AI
    really is. No, the simpler problem will not be the full AI. I understand
    that. Yes, scaling issues may make it irrelevant to the final solution.
    I understand that too. But what you have to understand is that AI
    is *hard*. And you won't appreciate that until you try to solve the
    simpler problems. Even if they are trivial and not entirely related
    to what you think is the "final solution".

    The benefit of a formal AI training is not that you learn about AI. It's
    that you learn how hard it is. You learn about the theoretical
    usefulness of the A* and Min-Max algorithms, and then you learn
    about the limitations of those algorithms. You begin to see just
    how narrow most solutions really are. And you begin to get a
    better appreciation of just how amazing a human brain really is.
    You learn about the Resolution algorithm, and you think that you
    could solve any logical problem in the world. And then you
    implement it, build up a large proposition database, and see that
    it would both be too slow to be effective on a large scale, and too
    rigid to be effective in a practical environment. These are lessons
    that you don't learn by reading, but by doing.

    So do yourself a favor. Instead of jumping in head-first, take some
    baby steps. Pick a simple problem, and try to solve it. Only then
    can you begin to appreciate the types of issues that will arise in a
    full-scale AI.

    Dave
     
    David B. Held, May 24, 2004
    #26
  7. Hi Curt,

    I found your reply quite interesting and entertaining. Thanks for giving
    it.

    Now let me address some points you raised and clarify some things about
    myself and my approach ... without giving away the secret recipe. ;-)

    Yes, I agree and that is why I'm continuing my research to find if my idea
    has been done before. I value my programmer friends' input and skills, but
    do realize they might be lacking in complete overall and historical
    knowledge ... which I'm sure they would be the first to admit that they are.
    In fact, none have objected in the slightest to me making another attempt to
    see if my idea has been tried before.
    Not in the slightest.
    I agree ... though I think we might disagree from which discipline those
    "engineers" should be from. Or rather, who should be in charge and who
    should just be support. I think history will show that this is the main
    reason why AI didn't develop over these last fifty years. The reason being
    that it is currently dominated by the wrong academic discipline and people.
    For it has always dumbfounded me why programmers think they can come up with
    an AI. No one would ever go to them with a psychological problem and yet
    they think they can solve the ultimate psychological problem. Baloney. I
    would place AI research under psychology and merely use computer programmers
    as tech support. The programmers sitting along the wall and only talking
    when spoken to by the team of psychologists at the conference table ... or a
    small team of programmers working for and underneath just one psychologist.
    Of each component, yes.
    If you mean by that statement: "given this input, it should come out this
    way", yes. Not specifics, but a standardized way of processing input. For
    example, I insert a banana and the banana comes out peeled. Any fruit I put
    into it comes out peeled. Another component would be used to slice it.
    Another component would drop it into a dish. Still another would pour Half
    & Half on it. Etc. Etc. Etc.
    No, I was just trying get across a complex idea in what I thought was an
    easy-to-understand and brief way. Essentially, what I meant by
    "how the AI is to think" is how the components process input. All of the
    processes do not take place in any single component. Only a little in each.
    What's important is how all the components are used and how the overall
    process works. Those are the rules I was referring to. Those are the rules
    I'll contribute to the project. The rules each component follows when
    processing their input before passing it along to the next component and
    from where they are to acquire the initial input.
    That would depend on how you define "thinking". I define thinking as simply
    a process that uses memory (to include rules) to process input, add to
    memory, and achieve goals.
    Behaviorism is no longer the golden child of psychology or artificial
    intelligence research. Cognitive psychology is the current darling.
    However...

    I'm a behaviorist and view any meaningful contribution from cognitive
    psychology as having a strong underpinning in behaviorism. Behaviorism
    being "accepted" as a contributing factor to cognitive psychology but only
    to a limited degree. Why this is the case in psychology today is because
    behaviorism demystified the mind and many religious people didn't like that.
    Behaviorism strips the mind of the "soul", "the touch of god", "the spirit",
    "the thing that will live beyond death". Cognitive psychology adds a layer
    of mysticism on top of behaviorism. Cognitive psychology is the mystics'
    answer to behaviorism ... as limited-socialism is the now-discredited
    communists' answer to capitalism.
    I would agree.
    That's essentially the approach my programmer friends and I are taking. I
    lay down the requirements I need the component to do, they build me a
    component that will meet those requirements.
    I totally agree on this point.
    They have both, but they still want me to give them the specifications of
    each component. Being the originator of the AI concept, I have a far easier
    time adjusting the idea and explaining it more intrinsically when finer
    details/definitions are needed. They just don't want me helping them build
    the components for it. That's where they come in and I stay out.
    The word "easy" would never be a word I'd use to describe what we're
    attempting to do. If it was, I wouldn't need my friends and could just do
    it myself. To me, "obvious" and "easy" are not necessarily
    inter-changeable.
    Made even harder by the wrong academic discipline and individuals trying to
    understand and tackle the problem. Because of this...

    What I've been thinking of late is that if I get all the way to the point of
    telling my idea to an AI expert and s/he tells me that my AI idea hasn't
    been attempted before and sounds promising, that I will ask them for a
    letter of recommendation. A letter of recommendation that will vouch for my
    AI idea as being original and promising and then encourage the receiver of
    the letter to offer me a full scholarship to pursue a doctorate in AI.
    However, the people I would send that letter to won't be at any computer
    science graduate program, but a psychology graduate program. And I'd seek
    not only a full scholarship but also funding to hire two to three
    programmers to assist me in my quest. Then again...

    I know of at least owner of a computer software company that would, after
    receiving such a letter of recommendation, very likely set me up with what
    I'd need to make my attempt. I wouldn't get a doctorate in the process, but
    would get a very decent salary and be able to hire better help to assist me.
    From my own personal experience, there's a lightyear of difference between a
    wet-behind-the-ears freshly-minted computer science graduate and a seasoned
    professional. Then again...

    I also know of a foundation that might just give me a grant based on such a
    recommendation. Then I might be able to have the best of both worlds. Use
    that grant to give me essentially a full scholarship at a psychology
    graduate program (after using the recommendation to get accepted into their
    program) and be able to hire really good help.

    The above are just a few things I've been chewing over as I do my reading.
    Reading which will yield tomorrow that my idea has already been tried and
    discarded thus making all this a moot point. :)
    I may just do that if I'm ever in town.

    Scott Jensen
     
    Scott T. Jensen, May 25, 2004
    #27
  8. Scott T. Jensen

    Curt Welch Guest

    I think you are very correct on that last statement. And certainly there
    are some other behavioral psychologist here in c.a.p that would agree with
    that last statement.

    I'm sure the reason is that since we understand computers, and have a
    brain, it just seems it should be "easy". :)

    But, after 50 years, AI people have figured out it's not so easy. And a
    lot of AI just fractured into interesting algorithm work, and stopped
    trying to pretend they knew enough about the brain to work on the "big"
    problem. But others have turned to fields like psychology to get the
    missing answers. The amount of corss-polination going on is improving. But
    some would say that the mis-thinking AI people are just spreading bad ideas
    about human behavior into other fields, instead of learning what they
    should be from the experts. There could be turth to that.

    The good thing however is that we have more and more people from different
    fields looking at the problem. And that can only be good in the long run.
    You need as many different people with different perspectives looking at a
    hard problem as possible.

    I don't think you understand how hard engineering these solutions is, even
    if you have an understanding of what the brain is doing. But, if you work
    on it, you will either solve it, or get that understanding, so you are
    doing the right stuff.
     
    Curt Welch, May 25, 2004
    #28
  9. No, if you look at all the newsgroups the original post was cross-posted to
    you will discover that all but one of them are hardware newsgroups. Thus
    why I'm focused on (or, as you say, "worried about") hardware. And I was
    simply posting in hope of possibly receiving some unconventional ways of
    tackling the hardware aspect of the project. I like keeping my options open
    and soliciting new and different ideas. Whereas my team of programmers
    think posting here is a waste of time, I disagree. However, since it only
    "wastes" my time, they don't really care or object to me doing so. As this
    thread has developed, I've forwarded some of what I believe are the best
    posts and private emails to my lead programmer and he's discussed their
    merits or lack their of with me. They haven't yet changed what hardware he
    is recommending I get, but he has said he'll look into a few of the hardware
    suggestions that I've passed along. That and there's no real rush in
    assembling the hardware since that will only be purchased and assembled
    after the program components are ready to be installed.

    And I'd like to again thank all those that have offered hardware suggestions
    and constructive critiques of them, both by public post and private email.
    They are very much appreciated.

    Scott Jensen
     
    Scott T. Jensen, May 25, 2004
    #29
  10. The idea that AI engineers don't study psychology is ridiculous,
    insulting, and extremely naive on your part.
    It's simple. They understand something that psychologists don't:
    computers. Until you understand what it takes to make a computer
    do something, you will never be able to appreciate the full
    complexity of AI. And I'm fully convinced that what you think is
    "the solution" will turn out to be painfully underspecified when the
    time comes to put it to the test. It's like a race car driver saying:
    "I know how to build the world's fastest car because I drive them
    every day. What do those stupid engineers know?" It's a non-
    sequitur.
    What's baloney is that practice in clinical psychology would tell you
    how to build an intelligence. It's baloney because building AI is *not*
    the "ultimate psychological problem". It's not even a *psychology*
    problem to begin with! It's an *engineering* problem. Which is
    why *engineers* pursue it, and not *psychologists*. And that's why
    *engineers* design and build race cars, and not *drivers*.
    And that's why Rodney Brooks runs the MIT AI Lab and not you.
    He has numerous successful projects under his belt. What do you
    have? A half-baked idea? The fact that you are trying to solve the
    hardware aspects of the problem before building so much as a
    software prototype demonstrates that you have an extremely naive
    view of how to approach AI.
    LOL!!! You are thinly disguised AI crackpot in psychologist's
    clothing. I haven't met a single psychologist that begins to have
    the tools or experience necessary to tackle AI. That's why Steven
    Pinker writes books, and leaves AI to...who? That's right, the
    *engineers*. Antonio Damasio is a neurologist, not a psychologist,
    but he's in a far better position to design an AI than a traditionally
    trained psychologist; but even he would be a terrible AI engineer,
    because he thinks that neurons are too low a level to understand
    the processes of the mind. In some sense, he's right, of course.
    But his position makes him unwilling to get his hands dirty and
    actually experiment with neural networks.
    Uhh...no. How each subsystem works is just as important as how
    they are connected together. You seem to think that they are just
    simple black boxes that your programmers and just whip together
    like a chicken pot pie. In fact, you'll find that the limitations of a
    simple component affect how the system as a whole works. And
    I'll bet money that you will end up saying: "Why doesn't it do this?
    Why doesn't it do that?" And the programmers will say: "That's
    really hard."
    If you think you are going to specify a set of precise, hard-wired
    rules, I think you'll find that your AI is not very general. And an AI
    that doesn't adapt usually doesn't leave a big impression on
    people, either. Such systems tend to be viewed more as
    engineering solutions than AI.
    That's because it's artificially constrained in a non-useful way.
    That's because it's a much more pragmatic approach.
    It's easy to say: "Idea X came from behaviorism." But that's just
    playing partisan politics with psychology. Does it really matter if
    it "came from" behaviorism or cog. sci.? If the idea works, use
    it. Arguing about whether it came from paradigm X or paradigm
    Y is like arguing whether string theory or quantum gravity is the
    better physical theory. Right now, we just don't know enough to
    say. So run with what works.
    It does no such thing, and I defy you to substantiate this unfounded
    claim. This makes me wonder if you are just Longley posing as
    someone else.
    And what I'm wondering is if any such components have yet been
    built and tested. My guess is "no".
    Good luck! There are infinitely many AI ideas that haven't been
    tried before, and considering the state of the art, it's pretty hard to
    say whether any given approach is "promising" without having a
    prototype. It's extremely easy to overestimate the power of a
    proposed AI system. Take it from someone who *has tried*. And
    what you'll find is that the people who are able to spend money on
    AI are doing it secretly, and those doing AI in the open have no
    funding to speak of.

    Dave
     
    David B. Held, May 25, 2004
    #30
  11. How the #### would you know?? You don't even know the history of your
    own professed area of expertise!
     
    David Longley, May 25, 2004
    #31
  12. Scott T. Jensen

    Matt Guest

    At one of the tables at a computer show at the state fairgrounds about a
    year ago, I met a grizzled mumbling old man with thick glasses and a
    pocket protector. For a hundred dollars more than the expected price he
    sold me a RAM stick in a plastic package having a special property. You
    remove the RAM stick and install it. Next time you sleep, you put the
    empty package under your pillow and repeat to yourself three times "wham
    blam free RAM". When you wake up you find that the package isn't empty
    anymore: it has another brand new RAM stick in it. So now I have more
    RAM than I will ever need. I don't need the package anymore, and I'll
    make a deal with you. Generic 1G PC2700. I will email you.
     
    Matt, May 25, 2004
    #32
  13. Scott T. Jensen

    Matt Guest

    Can we say that you hope your special vision will enable you to outdo a
    lot of extremely smart people who have worked on AI problems for a long
    time?
    How can I put this? I believe you need somebody to wake you up. Find
    somebody who knows about scientific research and is honest or cares
    about you.

    This group may be too polite or too far removed from scientific research
    for your needs. You might also try posting your plans to some sci.*,
    sci.research.*, and other groups.
     
    Matt, May 25, 2004
    #33
  14. http://www.autosophy.com/database.htm
    "Autosophy databases work best with large Content Addressable Memories
    (CAM). An ideal CAM would have storage capacities in the Terabit range,
    fast access, cheap and rugged construction, and low power consumption.
    The CAROM Content Addressable Read Only Memory, explained in a separate
    tutorial, has all those characteristics. Such memory modules could be
    implemented in complementary pairs to assure system reliability with
    cross-verification and repair. Giant multimedia databases could thus
    be built to virtually never fail. ...

    Each location address can also be erased and relocated to other
    sections of the memory. The CAROM is therefore usable both as a
    Content Addressable Memory (CAM) and a Random Addressable
    Memory (RAM). Memory locations are normally filled with data, by
    the learning algorithms, starting from the bottom of the memory and
    proceeding to the top. All network types (such as serial, parallel
    and associative) are interleaved in the same storage device."

    http://www.autosophy.com/carart.htm

    "The new Content Addressable Read Only Memory (CAROM) is being
    developed with help from the U.S. Army to replace the CD-ROM in
    severe conditions, and as the memory for a future brain-like
    Autosopher. The same device can be used as a Random Addressable
    Memory (RAM) or Content Addressable Memory (CAM)."
     
    Stephen Harris, May 26, 2004
    #34
  15. Would that secret funding be from the Men In Black, the Illuminati, or the
    New World Order? Or would you have to kill me after you tell me?

    Whoops! Sorry, everyone, I've been feeding the trolls. I'm done with this
    person.

    Scott Jensen
     
    Scott T. Jensen, May 27, 2004
    #35
  16. Don't be ridiculous. You are trolling.

    Check your mind out.

    Cheers,
     
    Eray Ozkural exa, May 28, 2004
    #36
  17. Actually, that "secret" funding would be from the IT dept. of any
    corporation with something to gain from AI technology. If you
    were a Fortune 500 CIO and had working AI in your dept., who
    would you tell? Why?
    I'm not the one claiming to have an AI so powerful I need a cluster
    to run it on.

    Dave
     
    David B. Held, May 28, 2004
    #37
  18. Computer Scientists: The "Crab People" of Neuroscience and "AI"?

    No, I think he's basically right. Where he's wrong is just in not saying
    that it should be behaviour analysts, not "psychologists". Certainly, up
    to the 1980s in neuroscience, it was the research scientists who gave
    the computer department instructions (they didn't like it much but they
    complied - where I worked, the scientists worse white coats <g>). The
    computer people, skilled though they were, were there in a technical
    capacity (and I can see how it would have been resented, as they would
    take the user-requirement and have to code it). They felt they were
    writing the theory, and to extent, they were of course. But note how
    that was driven by empirical scientific research. It was not sui
    generis. The hard part is getting to the stage where you even know there
    is a problem and can articulate it so that it can be programmed, even in
    scientific language. The computer people ("crab people" get the
    allusion?) were more like lab technicians than scientists. Even then,
    there was a problem of "the stationary section taking over the office"
    syndrome. That seems to be largely what has insidiously happened though.

    This does not deny the important contribution which computer science and
    engineering makes, but it has to be remembered that empirical scientists
    have always "programmed", be that in chalk and talk theory construction
    (modelling) or in writing code in FORTRAN etc in the early days to
    control equipment or run statistical analyses. Whilst researchers will
    use multi-variate statistics (both descriptive and inferential) to
    analyse data, many in "AI" and "Computational Neuroscience" seem to have
    fallen into the trap of believing their own hype when they "cognitivise"
    these routines.

    In most cases they *are* providing a service in what is today called,
    perhaps misleadingly, for politically correct reasons "multidisciplinary
    research", but it is just a bizarre form of grandiose egotism to assert
    that computer scientists are running or leading these disciplines. To
    say that the lead must come from behaviour, is not to give honorific
    *personal* status to the *people* doing the behavioural work, it's just
    to say that that *work* must guide neuroscience and the computational
    modelling and engineering. We have been through this before in some
    detail within comp.ai.philosophy over the years. The consequence of not
    getting your priorities right over this is, as Scott rightly says, that
    you are likely to end up with people who really know next to nothing
    about behaviour, trying to do things that they just don't understand.

    Sometimes - it looks like the lunatics (computer scientists/technicians
    - and I only refer to those "lunatics" who think they *do* run the show)
    really are running the asylum (neuroscience/AI). If you think this
    through carefully (which *you* need to) you'll see that your failure to
    grasp this accounts for many of the other silly things you end up
    saying.

    [Note, behaviour analysis - *not* psychology, most of "psychology" is
    cognitive and that's not the place to take the cues from either.]
     
    David Longley, May 28, 2004
    #38
  19. Scott T. Jensen

    Al Dykes Guest


    Lots of the classic AI work was done on machines that were so slow
    that it wouldn't pay to turn one on today. Your grandma's PC is
    faster. The OP seems to hope that there is some source of useful
    memory and disk space that is much cheaper than what can be found on
    pricewatch.com. It doesn't exist.

    FWIW the OP can put together a dual operton and 16GB of memory and a
    few hundred gig of disk for a couple thousand bucks. That should be
    enough to get a proof of concept runnning and to collect some real
    data about how much more is needed. He's likely to find that either
    his idea is flawed, or hopefully that the dual operteron is enough.
    If he finds he needs 100 time more resources, he's talking $200k,
    which isn't big bucks when you look for funding, if you're a serious
    player.

    It sound like the OP has no feel for how much programming can be
    stuffed into 16GB of ram, or 500GB of disk space.
     
    Al Dykes, May 28, 2004
    #39
  20. Scott T. Jensen

    Ted Warring Guest

    I am no longer sure of the number of people that I have heard claim to
    "know how to solve AI". I could list at least half a dozen that most
    everyone in the newsgroup would recognize. The characteristics of the
    claimants usually divides them into two groups:

    1 - Those that have only begun to look into the "problem", but feel
    certain that their personal insights and intuitions indicate the
    "problem" can be solved if they truly apply themselves.

    2 - Those that have spent more time in the field, and feel that if
    they have just a little "more" of something that the solution will
    emerge. Sometimes that is more connections, sometimes more RAM,
    sometimes more speed, sometimes more rules or a bigger knowledge-base.
    This belief is usually based upon whatthey feel to be promising
    initial experiments, and the hope that doing the same thing on a
    bigger scale will have proportionally better results.

    The one thing common to both groups is that none of them have
    re-surfaced with a working general intelligence adequate to produce a
    robotic butler, electronic replacement for the family dog, or even
    advancement in factory robotic intelligence making them a millionaire.

    The ones that simply disappear have my sympathy, the ones that demand
    that we all believe them without any proof (and usually without a bit
    of evidence) are kooks.

    One of my favorite responses (borrowed from Jack Dunietz) is: "The
    proof is in the pudding".

    Perhaps one of these days someone will come back with just such proof,
    but to date I hear nothing but shouted professions of having the best
    pudding recipe on the planet.

    What is worse though, are those that spend their career focused on the
    exact speed the mixer should be set to (and insulting all those that
    disagree), and never attempt to make pudding themselves.

    Personally I would be happy to see some pudding, no matter who the
    chef was.

    -Ted Warring

     
    Ted Warring, May 29, 2004
    #40
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.