Summer 2011 Monday, 07/18 Blade Runner • Was Deckard a replicant? Blade Runner • Is there a characteristic feature that makes us human, as opposed to a machine or a replicant? Blade Runner • What do you think of the "Voight-Kampff" test? Is it a better measure of humanity (or of the human kind of cognition) than the Turing test? Is emotion a better mark of being human than reason? Are emotion and reason even separable? Blade Runner • Could machines that don’t initially have emotions develop emotions on their own, without being explicitly programmed to have/develop them? Physical Symbol Systems • A physical device that contains a set of interpretable and combinable items (symbols) and a set of processes that can operate on the items. • A kind of automatic formal system. • Symbol = stable entities that are capable of semantic interpretation, that can participate in processes of internal manipulation, e.g. copying, erasing, conjoining, and that can be organized so as to preserve semantic sense. • Must be located in a wider web of real-world items and events. • A symbolic expression designates an object if “given the expression, the system can either affect the object itself or behave in ways depending on the object.” Physical Symbol Systems • The Physical Symbol System Hypothesis: A physical symbol system has the necessary and sufficient means for general intelligent action. • Sufficient, since any such system “of sufficient size” can always be programmed so as to support intelligent behavior. • Necessary, since nothing can be intelligent unless it is an instance of a physical-symbol system (PSS). • This is an empirical hypothesis. All cases of intelligent action will, as a matter of scientific fact, turn out to be produced by a PSS. Symbol Systems and the Brain • The working of the individual neurons in the brain may not be important in understanding the mind as a physical symbol system. • What matters is the operation of the “virtual machine”, that can be realized in all sorts of ways. • The role of symbols may be occupied by higher- level brain processes/structures. It may even be that different types of brain processes/structures play the role of the same symbol at different times. • You should not expect to find things like EI, EII, EIILIIL (etc) in the brain! Just some physical thing that plays the role of these symbols. Semantically Transparent Systems Systems whose computational operations are defined over “familiar symbolic elements” (Clark). For example: • A chess-playing program that applies procedures to symbols for rook, king, checkmate. • A sentence parser that uses symbols for noun, verb, subject. • A program for reasoning about liquids that has symbols for liquid, flow, edge. Why Treat Thought as Symbol Manipulation? • Thinkers are physical devices whose behavior patterns are reason respecting. • A pedestrian witnesses a car crash, runs to a telephone, and punches out 911. • Common sense Psychology makes sense of all this at a stroke by depicting the agent as seeing a crash and wanting to get help. • The simplest scientific explanation is that the agent’s brain contains symbols that represent the event as a car crash and that the computational state-transitions occurring inside the system then lead to new sets of states (more symbols) whose proper interpretation is, e.g. “seek help”, “find a telephone”, and so on. Why Treat Thought as Symbol Manipulation? • The thought “it is raining” often leads to the thought “let’s go indoors”. • Many of our thoughts are related to our other thoughts in virtue of their meaning in this way. • One explanation of such rational thought- transitions appeals to general syntactic rules that manipulate semantic representations in a way that preserves semantic sense. (think of the PQ— system again…) Why Treat Thought as Symbol Manipulation? • We have the capacity to understand an infinite range of sentences and to produce an infinite range of thoughts. • “Billy left his tricycle on the moon”. • What could explain our capacity to understand infinite new sentences like this and to entertain infinite thoughts of this sort? Why Treat Thought as Symbol Manipulation? • The ability to entertain certain thoughts is intrinsically connected to the ability to entertain certain other thoughts. • We don't find speakers who know how to express in their native language the fact that John loves the girl but not the fact that the girl loves John. This is appears true for expressions of any n-place relation, e.g. Eli (x) gave his paper (y) to Jon (z). • One explanation appeals to general rules for conjoining (interpreted) symbols in the language. • We’ll talk a lot more about this next week… Examples: Story Understanding • A computer program that deploys scripts. • The scripts uses a symbolic event description language to encode background information about certain kinds of situations, e.g. restaurant visits. • Takes as input a short story: “Jack goes into the restaurant, orders a hamburger, sits down. Later, he leaves after tipping the waiters.” You can then ask: “Did Jack eat the hamburger?” and the computer answers “yes” by applying the script. Examples: SOAR • ongoing project to implement general intelligence by computational means. • Uses symbol processing architecture. • All long-term knowledge is stored in a format called production memory. • Knowledge is encoded in the form of condition-action structures: “If such-and-such is the case, then do so and so”. Examples: SOAR • When it encounters a problem, it transfers all potentially relevant knowledge into a working- memory buffer. • A decision procedure then selects an action to perform on the basis of relative desirability. • SOAR is able to work towards a distant goal by creating and attempting to resolve sub-goals that reduce distance between current state and overall solution. • Learns by “chunking”. Research Program (GOFAI) • Design a program that can solve problems and interact with the environment in human ways. • If such a program is found, a good case can be made that it’s actually implemented by human brains. • We can thus study minds directly (by studying the software) without worrying about the messy details of the brain.
Pages to are hidden for
"Minds and Machines - Eli Alshanetsky"Please download to view full document