[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Subscribe]
DM: Data Mining's (Machine) Psychological ProblemFrom: Franklin Wayne Poley Date: Sun, 26 Mar 2000 13:59:47 -0800 I thank all those who contributed the "mountain of data" on the = definition of data mining. The first thing I did in mining the data was = to write out notes (five pages) on what I thought were the most = important points made in all of the postings. Then I studied those notes = several times as I attempted to write out a useful summary of "the = essence of data mining". That is WHAT I did. HOW did I do it? Honestly I = don't know much about the 'how' of it. The late BF Skinner said, "If it = can be verbalized it can be programmed" and that dictum is at least = mostly true. Given that dictum, if I could verbalize the step-by-step = way in which I did this data mining I could pass it on to the computer = code writers who could then turn it into a useful piece of software. = Exercises with definitions are important. Every statute here begins with = a set of definitions. Recently a man in Ontario won a court case by = saying that according to standard law dictionary definitions he was a = "natural person" and not a "legal person" and therefore the court had no = jurisdiction. Perhaps the judge was amused...and ruled in his favour. Of = course my effort at "natural data mining" may be more amusing than = useful. And you are the judges. I can verbalize some of this process without much effort at = "introspection". For example, if I found some words like "The definition = of data mining is...." I would make a note. Similar efforts to get right = to the essence of data mining were also noted. With a little more = introspection I could probably come up with a list of key words I used. = But I also made notes of particular or specific points about data mining = techniques which I thought might shed some light on the essential = features. I don't know what criteria I used to select those units. I = also don't know if I could discover it even with a lot of introspection. "Homo sapiens" indeed. "Robo sapiens" (couldn't resist that one) is = much more cognizant of its (artificial) intelligence in that a robot = could be programmed to spell out in fine detail every machine step as it = processes data. Right down to the movement of electrons. We humans = really aren't very self-aware. Some of us, however, are more self-aware = than others and the self-awareness may be general or specific. There are some very interesting "big problems" in robotics and AI and = Skinner's dictum applies to them. Two other big problems have to do with = developing a general learning program (glp) and a conversational program = (cp) which would meet the criterion of human equivalency. In both cases = I would say that it is reasonable to ask the experts to come up with the = necessary verbalizations to prepare the way for code writing. Not an = easy task, but "doable". In the case of a glp I would ask psychologists = and educators as well as machine learning experts. In the case of a cp I = would ask linguists, translators and clinicians who teach patients how = to converse and interact socially. A general data mining program (gdmp) would have enormous value. Isn't = that what all of this work ultimately is expected to achieve? But who = could verbalize how we discover that which is important in the = "mountains of data" on our computer screens? Again we ask, "Who are the = experts?" In the case of data mining it sounds to me like the experts = are "good students" in a huge variety of fields. Not necessarily good = teachers or people good at setting up data files and displays. But we do = know there are people who can speed through mountains of data in = commerce with understanding, or mountains of data in medicine, or = computing science, etc. Beyond that we know that among those "good = students" some will be more cognizant of HOW they do WHAT they do than = others. When this kind of good student studies all those statistics and = graphs and verbal reports in commerce, he or she quickly arrives at a = useful and understandable rendition for what is initially confusion. If = we can then find the good students who also have unusually high = INTROSPECTIVE ABILITY and enlist their co-operation as they do "natural = data mining" with a variety of subjects we may be able to arrive at a = gdmp in a verbal form first and then convert it to a computer program. Well, that's about the best I can do to this point. Of course the = gdmp would be expected to evolve, step-by-step as all these programs are = written, as is happening now. Any guestimates as to how long it would = take to arrive at a gdmp in that manner? Any other suggestions as to how = one could go about it with a single mega-project? Any fallacies in the = reasoning above re finding these "good students with high introspective = ability who are willing and able to report verbally on their = introspections?" FWP http://users.uniserve.com/~culturex/Machine-Psychology.htm PS-As an aside, any comments on Neugents software with its ubiquitous tv = ads? See http://www.cai.com/neugents The tv ads tell us "this is the first software that can think" and the = web site says "This on-going self-learning process...can acquire far = more knowledge than any expert in any field". These are astonishing = claims. If taken literally, Neugents is already a gdmp which can acquire = the knowledge we seek even better than human experts in any field. We = are told it "...has the capacity to learn, accumulate knowledge and = apply this knowledge to new situations". That being so, it is more than = a gdmp. It can learn more broadly and apply this knowledge to new = situations. Since "Neugents learn by observation" why could they not be used as the software for a robot cognitive = sub-system in conjunction with a recognition sub-system, ie "machine = vision/machine perception"
|
MHonArc
2.2.0