2 ms·
The Lack of Intelligence About Artificial Intelligence
- daly 10y agoWhy Machine Won't Think... (ref: [1] Jammer, Max "The Conceptual Development of Quantum Mechanics" McGraw-Hill (1966) pp 178-179 [2] Persig, Robert "Lila: An Inquiry in Morals" Bantam (1991) Chapter 11 ... In [1], Jammer quotes from Principles of Psychology ... The Stream of Thought, like a bird's life, seems to be mde of an alternation of flights and perchings. The rhythm of language expresses this, where every thought is expressed in a sentence, and every sentence closed by a period. The resting-places are usually occupied by sensorial imaginations of some sort, whose peculiarity is that they can be held before the mind for an indefinite time, and contemplated without chainging; the places of flight are filled with thoughts of relations, static or dynamic, that for the most part obtain between the matters contemplated in the periods of comparative rest... The rush of the thought is so headlong that it almost always brings us up at the conclusion before we can arrest it. Or if our purpose is nimble enough and we do arrest it, it ceases forthwith to be itself... The attempt at introspective analysis in these cases is in fact like seizing a spinning top to catch its motion, or trying to turn up the gas quickly enough to see how the darkness looks. .... in [2] Persig writes about "Dynamic Quality"... But Dynamic Quality is not structured and yet it is not chaotic. It is value that cannot be contained by static patterns. ... Science values static patterns. Its business is to search for them. When nonconformity appears it is considered an interruption of the normal rather than the presence of the normal. A deviation from a normal static pattern is something to be explained and if possible controlled. The reality science explains is that `reality' which follows mechanisms and programs. That other worthless stuff which doesn't follow mechanisms and programs we don't pay any attention to. ... A thing doesn't exist because we have never observed it. The reason we have never observed it is because we have never looked for it. And the reason we have never looked for it is that it is unimportant, it has no value and we have other better things to do. ... but 'spur of the moment' decisions... that directed the progress of evolution are, in fact, Dynamic Quality itself. Dynamic Quality, the source of all things, the pre-intelectual cutting edge of reality, always appears as 'spur of the moment'. Where else could it appear? ... Mechansims are the enemy of life. The more static and unyielding the mechanisms are, the more life works to evade them and overcome them What current AI and machine learning is doing is capturing the "things that don't change", recognizing the resting places, finding the common elements of the flow of information. As such, it will not see the flight between the static points. Neural nets compute the function that extracts the 'static reality' from a flow of patterns. As such they are only able to represent the resting places, not the flight between the resting places. They only attack half the problem of thought. Given two neural nets, what is the dynamic relationship between them? What it the intuition that connects them? Is there an imaginative story to connect one to another? Is there a "rationalization machine" that can make up a connection between them? Persig says "One can imagine how an infant in the womb acquires awareness of simple distinctions such as pressure and sound, and then at birth acquires more complex ones light and warmth and hunger. ... We could call them stimuli but the baby doesn't identify them as that. From the baby's point of view, something, he knows not what, compels attention. This generalized 'something' Whitehead's 'dim apprehension' is Dynamic Quality. When he is a few months old the baby studies his hand or a rattle, not knowing it is a hand or a rattle... If the baby ignores this force of Dynamic Quality it can be speculated that he will become mentally retarded, but if he is normally attentitive to Dynamic Quality he will soon begin to notice differences and then correlations between the differences and then repetitive patterns of the correlations. But it is not until the baby is several months old that he will begin to really understand enough about that enormously complex correlation of senatations and boundaries and desires called an object to be able to reach for one. This object will not be a primary experience. It will be a complex pattern of static values derived from primary experience. Is thinking possible without this primary experience, this 'Dynamic Quality', this flight between resting places? Neural nets can recognize a resting place, capturing what Persig calls 'Static Quality' but how do we model the flight? Does this require senses? The closest I can come to this question would be to think about a NN that focuses on changes. It would be comprised of many sub-nets that would recognize static parts of a scene. For example, a collection of NNs, each of which recognize static things in a room scene such as tables and books. But the 'main NN' would look for changes, e.g. the bee that just entered the room. Not knowing what a bee is, not knowing what it means to fly, but knowing the 'primary experience' of change of position, how do we dynamically construct a new NN that recognizes this 'bee thing' when it sees another one, having only seen one? How is this new 'bee NN' added to the collection of 'static NNs' behind the 'main NN' so that it 'knows about bees'? Besides change of position, there are other primary values such as size, color, sound, etc. Can we construct a new 'static NN' from 'primary NNs' that recognize these primary 'sensations'? How do you build a Neural Network to recognize 'size'? 'Pressure' is a primary sensation, requiring a sense organ. But 'size' is not primary, though it is primitive (aka non-specific) and derivable from something like images. We recognize cats from images but how to we use images to recognize 'size'?