Since duxx wanted to move our discussion to another thread, i thought i should post it. The basis for this can be found in the fuzzy hypertrophie thread. Essentually the topic is the use of expect systems, fuzzy logic, NN etc. for coaching/simulating human intelligence.
As to M’s response on the thread, i think it is possible to “mimic” a coach with a large NN. Since a coach makes decisions based on observed data, there is a relationship between input and output that a NN could be trained to. A limitation to this would be the amount of computational time to train a NN to this data and the ability to extrapolate outside of the coach’s experience would be heavily limited. (In addition to the difficulties with obtaining such a large dataset). In other words, it probably could never replace a coach at the elite level. This is not to say that it couldn’t be done. Economic firms have be using very large NN for years…
This is interesting. But the problem that now exists with AI is the new research (actually it is getting quite old now) from the realms of neuro-physiology and affective cognition (Damasio etc) which shows that a lot of what we originally called “decision” come from a combination of ‘feelings’. The problem for AI is that feelings are individual (e.g. they arn’t the same for everyone) and change from day to day as we learn new things and change our position on topics, what kind of music we like etc so creating a deterministic model (predictive model) is epistemologically problomatic.
In plain english this (epistemologically problomatic) simply means that there is a conflict between saying that people are all different but then going to say that despite this we can model thier behaviour using mathematics (something that is much more useful for modelling things that don’t change very quickly such as stuff from the physical sciences - physics questions). Another way of thinking about it is that it is kind of like being hypocritical but with respect to your theoretical position on something.
I’m not going to bore you much more right now because this is a very difficult area (which took me about a year to read about and fully understand during my doctoral study). But suffice to say I would not hold out much hope for our jobs as coaches being overtaken my even the most sophistocated NNs any time soon.
If you want to read more I suggest looking at Terrence Love’s work on Design Research or Yoram Riech’s work in Engineering and AI.
Guys,
I am a student of sport faculty, so I cannot deal with you math/cyber/techno freaks all my knowledge from this areas comes from “reading math just for fun” and some simulations (personal experience) in MATLAB etc. You would not believe if I say to you that, withoud advanced math knowledge, I discovered on my own differential equations and solve them using numerical computations in step-like manner (Euler and Runge-Kutta). Year after, when reading some math and sim book I discovered what I have did, without any knowledge… man, I wish I was living before Maybe the Runge-Kutta would be called Duxx methods
Anyway, I will translate my article in couple of days, so I will let you know when it is finished. It is called “Triangle of truth” and in it I put my point of view at AI, consciousness, God, Universe etc… It is more philosophical in it nature without math at all. So if you are interested, please let me know before I start the translation process
Again, I will repeat, I find myself as Descartes Dualism proponent, thus I believe that “life” is not just an cosequence of “physiological activities” in brain and CANNOT be explained nor simulated nor artifically created because it is NOT algorithim in nature. Free will is what I call a exclusive right of living things, an ability to break through determinism of universe. STOP me if I start to BS, please
Anyway, this doesnt mean that we cannot create expert system that will help us, or even replace us as a coaches, but this system is unable to feel, to be creative, to have free will… it is an algorhitm… it is not alive… Maybe it will have more succes than we do, but the first thing we should do is to find a “coach algorithm” and this is very hard (read allmos imposible to do), because, in one hand I belive that gross human behaviour can be explained by algorhitms (determinism), but the thing that make a difference is a “free will” which cannot be simulated… So forget about virtual Charlie
I will expand more on this in my article…
BTW, my major reference is Roger Penrose “Emperor’s new mind”, a book that I found extremly hard to read because a lot of math in it, but some chapters are readable…
Keep the discus going…
I am thinking, after finishin Faculty of sports to either start motor control graduate, cybernetics graduate/ungraudeta or rehab graduate course somewhere out-of-my-country…
My current research (as a grad student) is in the realms of perception and system identification, leaning mostly toward auditory perception (hence why i am in sonic boom land.) Studying perception as an engineer has been an interesting experience. (Hence why i try to apply math to everything )
TC, i still think that it can be modelled (in at least a rudimentary way). If we assume that Locke’s idea of tabula rosa is true and take into account that there is only a finite amount of information the human sensory system can process. Then we have a starting point for an adaptive algorithm which changes based on the sensory information presented to itself. The algorithm could not be purely deterministic however. It would have to be stoicastic (a pretencious mathmatical term for have random elements based on probabilities) and would have to adapt based on previous experience. In a lot of ways when a person makes a decision, they build a cost function based upon their experience and the information present (regardless of the validity of these information sources) and makes a decision based upon their exploration of this function (in the time it they took to explore it).
I think that NN can fit this bill as a really power non-linear regression technique that can adapt and re-organize to new data, given enough computing power, data and complexity. (I agree it is no time soon).
Although, in this light, i agree that most AI research (especially involving language processing) is following a red herring…
I should probably read the penrose book too. Thanks for the reference.
I do understand that i am being somewhat fanatical about applying math to things that do not necessarily suit its application well (unlike physics), but that is what happens when people grow up around physicists and become engineers.
Thats ok, this is normal. The problem is that these days science has become divorced from philosophy, so scientists are just not aware of philisophical ideas relating to ontology (our own personal set of beliefs about the world - or world view) and epistemology (relationship between our world view and how we view thoery) other than those used in the natural sciences. The problem is that it seems like it must be possible to model humans using mathematics and I am sure it is with respect to sensory organs and other biological entities but the problem is when we try and model human thought everything becomes so much more difficult.
I wish it wasn’t it would make my research a lot easier!
If as a scientist you can learn about philosophy and the limits of analysis (there’s another book http://www.staugustine.net/limitsofanalysis.html) you will do everyone in both our fields a lot of good because you will help prevent us making the same mistakes over and over that throws research off in the wrong direction and makes it almost impossible for me to cite your work without making the same criticism (epistemological contractions)! While it might be bloody annoying to make these realisations (it totally messed up my planned research schedule) in the long run it will help everyone because no matter how far you travel if you start off going in the wrong direction it is very hard to find you way to your destination…
That makes sense. I don’t know if science in general has this problem (all of the physicists i know are more philosophical than me), but i am sure engineering does. Wasn’t there a quote somewhere that said that “the worst philosophers are retired scientists”. Have you read/heard anything by Sir Macfarlane? He has some interesting views on this and AI, and control theory research which discusses the trap that you mentionned.
I feel for your problems with your research schedule. I have had to dodge a lot of bullets to avoid delays in mine.
Godel’s theorem seems to screw up pratically anyone’s research in AI etc.
I don’t have the formal theorem with me, so this might not be 100% accurate. It essentually states “that no formal mathmatical system can hope to represent all statements about natural numbers”.
I have also heard this discribed in more sarcastic ways (mostly combined with some other jokes about modern math). The theorem has some interesting philosophical interpretations as well (especially if you in regards to considering the brain as a formal system or not). If you want more info, i suggest finding a book on it, as my knowledge of number theory and modern math is regretably incomplete.
Agreed. The worst part is that i have friends who like to joke about it. The best one is:
Q:“What is the greatest achievement of modern mathmatics?”
A:“1+1 does in fact = 2”
There is a bit more to Godel’s than that, but it is an amusingly flippant way to state it.
all my knowledge from this areas comes from “reading math just for fun” and some simulations (personal experience) in MATLAB etc. You would not believe if I say to you that, withoud advanced math knowledge, I discovered on my own differential equations and solve them using numerical computations in step-like manner (Euler and Runge-Kutta). Year after, when reading some math and sim book I discovered what I have did, without any knowledge… man, I wish I was living before 
