четверг, 1 ноября 2012 г.

Consciousness Hurts

Being interested in the field of Artificial Intelligence I often stumble upon the question of how to make some artificial system capable of telling good things from bad ones. Because I am much more of a dreamer than of an engineer, multiple supervised learning techniques, which are widely applied and have proven quite useful, offer little satisfaction to me. The problem with them is that supervised learning algorithm merely provides the system being learnt with a set of questions associated with presumably right - or, to be more precise, desired - answers. Afterwards the algorithm does it's best to adjust the system's guts in such a way that it will give the right answers to the right questions - at least, in most cases. As for unsupervised learning techniques, these differ in some way - particularly in terms of applications, although they remain external algorithms trying to adjust the trainee's internals in a manner allowing to minimize the proportion of wrong answers, which are now defined by the algorithm implicitly. While dealing with different machine learning stuff I saw both these approaches producing amazing results, so I am more likely to exaggerate their power, than to underestimate it. Still I never ceased thinking that these fitting activities are too distant from the notion to which the term "learning" originally refers. The information fed to our machine learning tools obviously becomes a subject to some sophisticated memorization and even generalization, nevertheless I can hardly say that it is learnt.
This issue was made apparent to me by artificial neural networks. Inspired by nature - the nature of our own brain - many of these models clearly demonstrate how helpless they are until someone arrives with a proper learning algorithm and how much closer they stand to an adjustable screwdriver than to an ant. All the learning is applied to these things from the outside - it is the overseeing mechanism, learning algorithm, that makes adjustments to the system if it performs not well enough during the process of learning. Moreover, this same algorithm is the only thing, that knows what does it mean to "perform well". I know I am not of this kind: I can get beaten by a teacher, alive or not, if I don't learn a lesson properly, but it's not the teacher, who makes me learn it and imprints knowledge into my brain - much greater role is played here by my emotions and feelings: pride or shame, satisfaction or disappointment, relief or pain.
Pain is an interesting and off-putting thing. Few if any things in life concentrate our attention so completely and terribly, and few things are more boring to listen to or read about. 
Dan Simmons, The Rise of Endymion 
Pain must be the most well known teacher - I mean there is hardly anyone conscious in our world who hasn't ever learnt something through pain - almost everyone has felt it and everyone tries to avoid it. Thus pain is possibly the most comprehensible indicator of doing something wrong and the memories of pain experienced as a result of some actions do their best to prevent one from behaving the same way in future. The high-level mechanism here looks quite simple: pain is not a pleasant sensation at all, so the things which cause it are assessed unpleasant as well. Actually, I believe we can go a little bit further to note that the thing is tangled - although pain is harsh itself, it is also strongly associated with probable injury and is therefore - mainly subconsciously - deemed dangerous and unwelcome. I also feel that it is quite safe to assume that our mind associates the feeling of pain not only with injury, but with the possibility of death as well. Taking this into consideration, there is nothing surprising about the fact that a severe agony may totally absorb one's mind - when death is possible it doesn't seem affordable to let the mind's owner think of something except the ways to escape the least desirable event. An argument supporting the idea of this strong association between pain and death may occur from the fact that an agony itself may be fatal, although I'm afraid that the underlying mechanism of neurogenic shock is too weakly related to the notion of consciousness and so this example is not that applicable here.
So pain is a powerful mechanism allowing us to avoid some negative events or, put another way, telling us that certain things are not good, may even appear harmful and therefore should be avoided. It is important enough, but what makes it more interesting is that this mechanism is embedded in the system - our mind and body - in such a way that it is impossible to distinguish it from the entire system itself. We could suggest that multiple sensors spread through our body act as sources of pain - it is true to some extent, but it is not the only case, unless we claim phantom limb syndrome a sort of magic. Furthermore, I believe there is no special 'device' which is fully responsible for sensations of pain in our body - the entire nervous system takes part in producing pain as well as most other feelings, which we experience during our lives. Resulting from the overall behavior of myriad neurons linked together, instead of being produced and managed by some localized distinct components of the nervous system, these strongly related sensations are also linked closely to such activities as motion, vision, thinking, learning. This sole fact allows pain to be quite a successful teacher and makes it so different from all the mechanisms used in machine learning techniques, which I have seen so far. Since it is so overwhelming - in terms of both it's impact on our life and of the degree in which it is interwoven with all activities of our mind - it is impossible to ignore the concept of pain when one wants to ignite a spark of consciousness in something that didn't show it so far.

Maybe, speaking of pain I have made the colors too dark, but the picture is not that grim. All kinds of our feelings and emotions play a vital role when it comes to thinking and learning - to being conscious. I could have chosen anything more appealing - say, pride, satisfaction, love - whatever. It was the relentlessness and severity, with which pain invades our mind, that made me write of it.