INTRO TO COGNITIVE SCIENCE
INTRO TO COGNITIVE SCIENCE COGS 2120
Popular in Course
Popular in Cognitive Science
This 272 page Class Notes was uploaded by Arne Kulas on Monday October 19, 2015. The Class Notes belongs to COGS 2120 at Rensselaer Polytechnic Institute taught by Bram Heuveln in Fall. Since its upload, it has received 43 views. For similar materials see /class/224757/cogs-2120-rensselaer-polytechnic-institute in Cognitive Science at Rensselaer Polytechnic Institute.
Reviews for INTRO TO COGNITIVE SCIENCE
Report this Material
What is Karma?
Karma is the currency of StudySoup.
You can buy or earn more Karma at anytime and redeem it for class notes, study guides, flashcards, and more!
Date Created: 10/19/15
Perception and the Brain Introduction to Cegmtive Science Vision The Eye Pupil works as a pinhole so that light coming from a certain direction will hit retina at a specific spot camera works the same way Without pupil there would 39ust be a lightsensitive patch we couldn39t tell what direction light came from Rods m more lightsensitive not sensitive to color in periphery mm mm Cones less lightsensitive pick up green39 blue and red in center of vision gt l vz n m maul wwu39mr hm lm yildrgm l Vis39on The Retina G n H ceausg on Transduction the conversion of light or other sensory input into neural activity Bipolar cells This is done by the photoreceptor cells the rods and cones The rods and the 3 different kinds of cones Cone use different photosensitive pigments Rod Note the stupid design cells and optic nerve are between the light and the retina Blind spot is where the optic nerve breaks through the retina Vision Visual Processing in the Retina center mechanism surrwndmechanism GINCENTER OFF CENTER Inhibitory connections In sum ganglion cells 1 compress information 2 emphasize contrast gt good for edge detection Ganglion Bipolar Phot enstive cells Vision From Optic Nerve to Brain Stem to Thalamus to Occipital Lobes Note what s on our right will be detected by left part of retina and end up in left occipital lobe Focus on black dot in the middle Sensory Adaptation As cognitive agents we are most interested in changes in our environment Our brain processes things in such a way that things that remain constant become less and less noticed One way to do this Subtract current or predicted stimulus from current stimulus and pass on the difference to rest of brain Case Study Binocular Rivalry Binocular rivalry happens when a subject is presented with two different images one in each eye yes it requires an elaborate set up Instead of seeing a juxtaposition of the two images the subject perceives first one image then the other then the first one again etc The alternation could be explained by the predictive subtractive model of sensory adaptation when subtracting the current visual experience image1 from the input image1 image2 you are left with image2 Vision Visual Cortex Motor Cortex Somatic Sensory Area I Form 3D Position Motion Secondary primary Visual Visual Areas Cortex I I I I I I I I I I E xample Color Contrast quotmquot uquot m quot zm gg zzzzu W umquot 0 new quotozquot u A 39quot39 13 quot 039 l mambo wsz3933quot quoton 00 no 0999999 OOOOOOOQOQ a A 3 0 3 compare color ofthe 39 middle squares on top and 39 f In ront of cub e Ebbinghaus Illusion We see right orange circle as bigger than the left orange circle Ebbinghaus Illusion However when asked to imagine picking up the circle in the middle people hold their fingers apart atjust the right distance in both cases Two Distinct Visual Information Processing Streams Green Dorsal Stream lWhere Processes spatial information where things are located Note connection to Parietal lobe that deals with spatial information and contains sensorimotor cortex Purple Ventral Stream What Processes object recognition what it is we re seeing Note connection to Temporal lobe where language is located Ebbinghaus Illusion Evidence that 39what we are consciously seeing ventral stream is different from spatial information dorsal stream that is unconsciously available Judging Steepness of Slope Goodale and Milner When asked to judge how steep a slope is that they are facing may subjects overestimate But when they are asked to hold their hand unseen at the angle ofthe slope they get it exactly right Hypothesis the judgment is a semantical what judgment ventral stream the hand response a where judgment dorsal stream Blindsight Blindsight is a condition that some people have when their primary visual cortex V1 has been damaged while they can t consciously see objects they can still react in appropriate ways eg when forced to guess where a moving object is going left right up down they get it right 0 Information seems to get to the dorsal stream but not to the ventral stream What is Cognitive Science Introduction to Cognitive Science Some Definitions Cognitive science is the interdisciplinary study of mind and intelligence embracing philosophy psychology artificial intelligence neuroscience linguistics and anthropology Stanford Encyclopedia of Philosophy Cognitive science is most simply defined as the scientific study either of mind or of intelligence It is an interdisciplinary study drawing from relevant fields including psychology philosophy neuroscience linguistics anthropology computer science biology and physics Wikipedia An interdisciplinary science that draws on many fields as psychology artificial intelligence linguistics and philosophy in developing theories about human perception thinking and learning MerriamWebster Cognitive science can be roughly summed up as the scientific interdisciplinary study of the mind It results from the efforts of researchers working in philosophy psychology linguistics artificial intelligence robotics and neuroscience In Cognitive Science by Friedenberg and Silverman Some Academic History The term cognitive science was coined by Christopher LonguetHiggins in his 1973 commentary on the Lighthill report which concerned the then current state of Artificial Intelligence research In the same decade the journal Cognitive Science and the Cognitive Science Society began Wikipedia Since then more than sixty universities in North America Europe Asia and Australia have established cognitive science programs Stanford Encyclopedia of Philosophy n I Ling 3N3 Ic b0 3 391 u so E 2 Logo used by the Cognitive g Science Society and their 332 63 journal Cognitive Science 9 1 c 7 b c p 0 041 mow Cognitive Science Aspects of Cognition Cognitive science is the science of cognition which includes such things as perception action learning memory attention reasoning decisionmaking language use What about consciousness Emotions Cognitive Science as one of the Basic Sciences I predict that at some point in the future cognitive science will be regarded as one of the basic sciences in line with physics chemistry and biology Indeed I believe that cognitive technology is going to be the next revolution in technology We will have CognoEngineering autonomous intelligent machines brain interfaces and CognoEthics These are exciting times What can I do with Cognitive Science Cognitive Science can be used to analyze describe predict or even correct augment if not create minds Some specific applications are Cognitive Modeling Human Computer Interaction Artificial Intelligence Cognitive Robotics Cognitive Engineering Cognitive Modeling Cognitive Models can analyze cognitive behavior at small time scales thus eg predict the effects of texting while driving HumanComputer Interaction Cognitive Science could lead to more effective and efficient HumanComputer Interaction Artificial Intelligence Cognitive Robotics V V HandEye system in RPI s Cognitive Robotics Lab The Create the Roomba that doesn t suck Cognitive Ergonomics Designing tools with an eye for our cognitive abilities and hence the constraints it imposes on that technology cognitive ergonomically designed tools hould fit ourilities to take in and process llquot Better interfacing with Tom Cruise Cognitive Prosthetics Using EEG s brain waves measured at scalp patients that can t move their arms learn to control a cursor I on a screen Wadsworth Research Lab Albany NY Sensory Substitution Former blind Jens Naumann can now see good enough to drive slowly around in a parking lot thanks to a brain vision implant Seeing with Sound See with tongue Cognitive Augmentation Australian ArtPerformer Stelarc has a third arm which he can control using his quot39139 abdominal muscles Cognitive Science Kinds of Minds Cognitive Science can be seen as the study of mind but these minds can be human minds animal minds computer minds robot minds alien minds group minds or whatever minds In practice human mind is studied mostly We Shouldn t Focus on the Human Mind Alone An Argument What is physically What is chemically What is biologically What is cognitively possible possible possible possible QQ What is physically What is chemically What is biologically What is cognitively actual actual actual actual In other words if we merely focus on human cognition we are focusing on only a very small part of the space of all possible cognition A true cognitive science certainly if it wants to be a basic science focuses on the whole space Further Reflections l Just as considering the place of carbon in the whole Table of Elements teaches us something about carbon that we probably wouldn t have figured out by looking at carbon alone studying other forms of cognition can probably teach us more about human cognition So even ifwe just wanted to know about human cognition it still makes sense to contemplate other forms of cognition Further Reflections ll Since the space of actual cognitive phenomena is unfortunately much smaller than the space of all possible cognitive phenomena a true cognitive science is probably going to be more of an experimental science than an empiricalobservational science So while Cognitive Psychology focuses on human cognition and as such runs experiments by bringing in human subjects Cognitive Science proceeds with building and testing systems with cognitive powers quite unlike human cognition Further Reflections lll Are there going to be general principles of cognition that span much or all of the space of cognition What would those principles look like What are the dimensions parameters and units of cognition I think that there is definitely room here for a Newton Mendeleev or DanNin of cognitive science Indeed some people could become real famous by mapping the land of cognition Again these are exciting times Further Reflections IV When we do start to have a good grasp of cognition from a scientific point of view the application of this knowledge will probably be immediate cognitive technology Neural Implants Cognitive Enhancement Tools True Artificial Intelligence Science fiction now but all of this may be here sooner than we think Further Reflections V Notice that we are now in the midst probably more the beginning of biotechnology now which is exciting but at the same time a bit frightening Just look at the ethical questions that biotechnology has raised Well wait until you see what cognitive technology can do While we won t address ethical issues in this course I do think they are important issues and we raise them in the course Minds amp Machines Cognitive Science as an Interdisciplinary Study Again there is no fixed list of fields that people think should be included However certain fields are more consistently included than others Here is how I see it Primary Philosophy Psychology Computer Science Secondary Linguistics Neuroscience Anthropology Tertiary Biology Education Mathematics The Fields of Cognitive Science PhHosophy Philosophers have thought about the nature of mind for thousands of years Philosophy like science tries to understand things In fact it was in the 17th or 18th century that science grew out of natural philosophy which was a branch of philosophy that tried to understand the nature and working of the world around us Thus philosophy is what we can call prescientific philosophers are trained to ask fruitful questions to approach a subject from various kinds of perspectives and to develop the concepts ideas and vocabulary to support those different points of investigation It is only once such questions and concepts have been identified that the scientist takes over and through quantifying and measuring is able to formulate theories that make testable predictions Cognitive science being a young discipline is still very much in a prescientific stage The Fields of Cognitive Science Psychology Psychology which originated in the 19th century was the first science of mind Psychology has several branches clinical psychology developmental psychology behavioral psychology etc but the one that mostly influences cognitive science is cognitive psychology which originated in the 1950 s and 60 s Cognitive psychologists try and make functional models of the mind which can be used to make predictions of human behavior In fact much of cognitive science is driven by cognitive psychology In practice the work of many cognitive scientists is sometimes hardly distinguishable from cognitive psychology Many textbooks on Cognition or Cognitive Science are for the most part texts on cognitive psychology The Fields of Cognitive Science Computer Science Computer Science has a very important role in Cognitive Science as well Computers can be used as tools to create run and test models of human cognition Additionally and probably more importantly computer science has offered the information processing concepts and vocabulary that frames much of the current thinking and theorizing in cognitive science Artificial Intelligence is a branch of computer science trying to build computational models that are claimed to be cognitive themselves The Fields of Cognitive Science Linguistics Researchers have long suspected a deep link between cognition and the use of language Language can be used to represent information and cognition is often seen as involving the representation and manipulation of information One striking difference between humans and animals is the complexity of human language does this account for the other differences in cognition between animals and humans The Fields of Cognitive Science Neuroscience Clearly the nature of our mind has a lot to do with the nature of our brain and studying the anatomy and workings of the brain can provide us insight about the mind But what exactly is the relationship between the mind and the brain Are they the same Does knowing everything about the brain tell us everything about minds Do all minds even require brains The Fields of Cognitive Science Anthropology Anthropologists have all kinds of interests and insights into cognition Cognition and Communication Cognition and Social Groups Cognition and Evolution Cognition and Culture Unfortunately this area has been long neglected by practicing cognitive scientists eg cognitive linguistics has traditionally paid little attention to this areal but some more recent theoretical developments have the potential to change this The Fields of Cognitive Science Biology Biology can provide important insights into cognition as well Seeing cognition as subservient to the evolutionary survival of biological organisms and species Biologists can provide insights about the brain and other physical aspects underlying cognition nervous system whole body But what can cognition be without evolutionary pressures The Fields of Cognitive Science Educa on The relation between Education and Cognitive Science is reciprocal Researchers in education reveal things true about cognition Vice versa the findings of cognitive science can be applied to improve education The Fields of Cognitive Science Mathematics This relation has been little explored so far but there may be some deep connections between cognitive science and mathematics Obviously mathematics can be applied to analyze cognitive models Discrete mathematics in planning and reasoning Calculus and differential equations in perception and action But it is possible that the complexity found within cognitive science requires different kinds of mathematics eg to analyze and get a grasp on complex neural networks Indeed just as other branches of science pushed the creation of new branches of mathematics so may cognitive science Also cognitive science can inform us about the nature of mathematical thinking Robotics A Good Candidate for Integration While cognitive science is informed by many fields it is important to try and integrate all the results of these fields into one unified whole Also as argued earlier cognitive science will probably have to rely heavily on the experimental approach Hence it is important to build various different kinds of cognitive systems that can be analyzed from all these different perspectives Robotics seems to be a good candidate for this The Cognitive Science department offers a course Cognitive Robotics Cognitive Science at RPI In 19911992 RPI s Philosophy and Psychology Departments merged to form the Department of Philosophy Psychology and Cognitive Science In 2003 this became the Department of Cognitive Science one of only about 15 in the world In 2004 we created a PhD program in Cognitive Science In 2010 the BS in Cognitive Science was approved Yes you can now major in Cognitive Science About 20 faculty Several Laboratories CogWorks Lab Cognitive Modeling RAIR Lab Artificial Intelligence and Reasoning PandA Lab Perception and Action Virtual Reality HumanLevel Intelligence Lab Cognitive Architecture Lab 39 Co nitive Robotics Lab And various research groups Renssclaer Majors Minors and Concentrations Majors PSYC PHIL COGS When you dual major with PHIL or PSYC or COGS required courses can count towards HampSS requirement Minors Cognition minor in PSYC Logic Computation and Mind minor in PHIL Minors in COGS forthcoming Concentrations IT has Cognitive Science concentration GSAS has Cognitive Science Concentration Core Curriculum for BS in Cognitive Science Minds amp Machines Introduction to Cognitive Science Introduction to Logic Experimental Methods and Statistics Cognitive Psychology Behavioral Neuroscience Cognitive Neuroscience Introduction to Artificial Intelligence Cognitive Modeling Programming for AI and Cog Sci Sensation and Perception Structure of Language Philosophy of AI Knowledge Belief and Cognition Metaphysics amp Consciousness Undergraduate Thesis What Can I do With a Cognitive Science BS Rich Skill Set ProgrammingModeling Empirical Data Collection and Evaluation Complex Systems Analysis Critical Thinking Communication Careers industry academics in Computer Science Al robotics Psychology Cognitive Modeling Human Factors Philosophy Cogno Ethics lT HCI Decision Sciences Economics Anthropology Social Sciences Education Law etc Possible Dual Majors COGS CSC COGS MATH COGS GSAS COGS PSYC I am interested in the BS in Cognitive Science What should I do Talk to the Director of Undergraduate Studies in Cognitive Science Bram van Heuveln heuvebrpiedu Brain Basics Introduction to Cognitive Science Basic Anatomy 4 Main parts of brain Cerebrum Cognition reasoning planning decision making complex behavior Cerebellum Coordination of behavior posture balance walking Limbic System emotion and memory Brain Stem basic body regulation breathing heartbeat sleep most basic sensorymotor control eg eye movement Brains of Different Animals Cerebrum 4 lobes Frontal Lobe central sulcus executive control reasoning decision making sense of ethics parietal lobe Parietal lobe integration sensory information 3 visuo spatial pro occipital mOV lobe cessing ement numbers lateral sulcus 39 Occipital lobe I cerebellum telmgljra medulla VISUaI processmg o e oblongata Temporal lo exterior of the cerebrum from the left side be auditory processing semantics language The cerebral cortex or neocortex is the outer layer of the cerebrum This is the gray matter This seems to be where the cognitive computational action is Underneath the cortex is white matter which seems to serve more supporting functions eg connections Cortex so matosensory motor cortex quot cortex I 39 x visual cortex auditory cortex Motor and Somatosensory Cortex trunk arm hand leg fingers lt and thumb I f t eye gt oo face I litoes Motor cortex tongue head trunk arm hand Qal leg fingers y lt lIfoot and thumb Ftoes genitals t Somatosensory cortex ongue Here we find very specialized areas modules of the brain speci c areas responsible for speci c functions Question Is the rest of the brain mind like that Eg are there specific areas of the brain responsible for all knowledge regarding grandmother This figure represents how much of the cortex is devoted to the different parts of the body Limbic System Thalamus relay station of sensory input to cortex exception smell Hippocampus formation of memory Amygdala Emotional salience of stimuli Hypothalamus basic Jquot I drives 4 F s breathing heartrate and more Phrenology 19th century brain science Assumed localization of cognitive function personality traits really Bumps or dents on head were thought to reveal more or less neurons devoted to that cognitive function Modern Brain Nonsense The 10 of Your Brain Myth Most people only use 10 of their brain The suggestion being that if you could somehow unlock the extra power of your brain you could be so much smarter Which is why you should buy this book or product or take this 7 week course for only 499 Total myth First off all this makes no sense evolutionarily Indeed when doing brain imaging we find activity all over the brain at all times Where does the myth come from Probably from exactly those brain images Brain Scans CAT or CT Computed Axial Tomography Xray mostly to find suspected brain damage PET Positron Emission Tomography Radioactive chemicals inserted in bloodstream MRI Magnetic Resonance Imaging Place subject in powerful magnetic field and measure differences in frequency Good spatial resolution low temporal resolution fMRI highertemporal resolution lowerspatial EEG Electroencephalography Measure electric fields by placing electrodes in scalp Good temporal resolution terrible spatial resolution Neurons ll iquot V Human brain 1OO billion 1011 neurons 1OO trillion 1014 neural connections Dozens of different neuro transmitters Why And why neurotransmitter in the Neuro ammo l v s l tl nds first place why not a hardWired Storagcvuaiclc m a 39 mnlmrnng chemoelectrical connection newmnsmmsx r f Doesn t fit a simple neurons fire or not is like 1 s and 0 s in a computer story is Dell membrane Caz membrzmu ls informationprocessing clone 5333an swam ligaw F quotum I by chemicals From Behaviorism to Computationalism Introduction to Cognitive Science The Biological History Argument Science seems to show that there was a time when the universe was just a purely physical soup out of which life and mind somehow originated But if the mind is something nonphysical how can that be How can you get something nonphysical if you start with purely physical stuff Similarly consider one s biological development starting at the moment of conception As far as science can tell a just conceived egg does not have a mind and in fact there will be no mind present in the growing fetus or baby for until quite a while So again where does one s mind come from Does it somehow pop into existence out of nowhere Or does it get bestowed upon us by some unexplainable and unfathomable supernatural being for which we have zero scientific evidence More on the Biological History Argument In fact in both cases it doesn t seem like we can even indicate at what time a mind does come into existence This suggests that seeing the mind as a thing that entities either have or don t have may not be a very good idea in the first place In fact what this does suggest is the idea that the mind only gradually develops at the same time as the biological organisms develop In fact this argument can be seen as an important elaboration on the neural dependency argument it seems that minds develop gradually in sync with the development of the brain Life and Mind Potential Parallels Biologists have a hard time trying to define life There is no universal definition of life there are a variety of definitions proposed by different scientists Same is true for mind There are many typical aspects to life such as growth homeostatis response to stimuli and reproduction There are many aspects to the mind such as perception reasoning learning memory While many organisms have all of these features sorge have only a subset of these Are they alive or no Similar questions can be asked about the mind Artificial Life and Artificial Intelligence A branch of computer science called Artificial Life creates computer programs that have all of the features that biologists typically ist Should they be called alive Many people object as they are not carbonbased and no oozecomesoutwhensteppedupon But is that important Are we prejudiced when it comes to declaring things alive Artificial Intelligence presents us with the same debate Being Pragmatical about Life and Mind In practice many biologists really don t care very much about any exact definition of life they find that they can just study all these different aspects of life without having to give an exact definition of life itself Maybe the same is true for mind Maybe it is more useful to just study perception reasoning use of language etc without trying to define mind or cognition or intelligence itself Many practicing cognitive scientists see the mind as a multidimensional abstraction as implemented by the brain So again the mind is not a thing but it does have a purely physical basis The mind can be scientifically studied just like other abstractions such as economies can be studied Labels Concepts Facts and Science Asking whether or not a virus is alive or whether a computer can think may be less of a factual matterthan most of us may think Ultimately life and mind may bejust that linguistic labels expressing certain concepts And concepts can change In particular concepts can change because it may be useful to change them Why Because concepts allow us to make sense of the world parse it give explanations make predictions and sometimes we realize that a change in concepts may be even more effective in making sense of the world around us In fact this is notjust some wishywashy semantical issue but something we do in science Remember what happened to Pluto So maybe the question to ask is not What is a mind or What has a mind but rather What would be useful to consider a mind or consider having a mind The Beginnings of Psychology Introspectionism Presumably inspired by the successes of physics chemistry and other sciences during the 19th century the first systematic scientific investigations of the mind began The early views of the mind saw the mind as consciousness mind consciousness spiritsoul Consciousness was studied through introspection put human subjects in different situations or conditions and have them report on their conscious experiences Moving Away from Consciousness One obvious problem with introspection is that it is not an objective or thirdperson measurement How can these reports be trusted Indeed for this reason some people refused to call this even scientific Also Freud s Psychoanalytic Psychology for all its dubious theories did develop the notion that there is an unconscious mind and that unconscious mental processes in fact do a lot of work Important move away from mind consciousness Psychological Behaviorism During much of the first half of the 20th century the dominant school of thought became Behaviorist Psychology Behaviorists stated that since introspection is not a reliable method the mind must be studied through the observable behavioral dispositions The Stimulus gt Mind gt Response chain should be studied through the Stimulus gt Response chain only Problems Not all mental states are caused by stimuli Not all mental states cause observable behavior Philosophical Behaviorism Psychological behaviorism has a counterpart in the philosophy of mind philosophical behaviorism Whatever goes on inside the agent obviously contributes to the agent having those properties but doesn t constitute them 80 Mental properties are behavioral dispositions To be intelligent is to behave intelligent Compare a car s speed traction and maneuverability We can look under the hood of the car to try and explain why the car is fast but its fastness is constituted by its behavior to quickly go from A to B Problem What about our inner mental life our thoughts feelings sensations etc MindBrain Identity Theory Acknowledging the problems with behaviorism and probably pushed by advances in technology that allowed for brain imaging techniques researchers went back inside around mid20th century But this time the focus of study was the brain not consciousness and so it could still be seen as a proper science One view of the mind became known as the MindBrain Identity Theory Mental states are physical states of the brain To have a beliefX is to have certain brain neurons fire Problem the mind weighs 3 pounds Carbon Chauvinism One fruitful way to illustrate the problem with mindbrain identity theory is to ask why does having a thought or being in other some mental state require such a specific physical configuration In short what is so special about carbon Couldn t there be other beings made out of completely different materials that have thoughts and beliefs Philosophical Functionalism What makes a chair a chair is that we can sit on it The physical material of the chair is irrelevant Chairhood is multiply physically realizable Similarly the functionalists state that the mental states of an agent can be defined relative to an abstract causal system as implemented by that agent s sensory apparatus motor control and mediating mechanisms brain but that could potentially be implemented by other physical means as well eg we could replace a neuron with a prosthetic neuron and as long as it would function the same way your mind remains Brain or Behavior Functionalism can be seen as a kind of compromise between behaviorism and mindbrain identity theory Like behaviorism and unlike identity theory the emphasis is on the functionality of things Like identity theory and unlike behaviorism we are going to look what goes on inside of us The mind can only be understood in terms of the brain and its behavior Multiple Realizability Functionalism allows for completely different kinds of entities to be intelligent as the relevant abstract causalfunctional organization can be implemented in various ways Can computers be such entities The Problem of Consciousness While this conception of the mind seems to work well for such things as reasoning learning and memory there is one aspect of the mind that still gives us trouble consciousness Consciousness is subjective Consciousness is anything but an abstraction Indeed many of the doubts that dualists have about mind being physical are really doubts about consciousness being physical New Dualism is a modernday version of dualism lt no longer sees the mind as some kind of nonphysical soul or spirit and believes that much of the mind such as reasoning memory etc can be studied scientifically However they make an exception for consciousness and say that consciousness may be some kind of nonphysical property Objection to Functionalism Qualia Let us call the phenomenal aspects of consciousness qualia Qualia seem to be problematic for functionalism and thus for computationalism Absent Qualia Zombies are creatures that have the same abstract causal organization that we have and thus behave just as we do but which have no qualia Inverted Qualia lnvertoids are creatures that have the same abstract causal organization that we have and thus behave just as we do but which have an inverted spectrum eg when we see green they see red Cognitive Psychology The Mind as Information Processing Functionalism was very much driven by the development of the modern computer Question What is the functionality that the brain implements and that gives rise to a mind Answer It is informationprocessing functionality the ability to take in perception store memory and process thinking information So around the 1960 s inspired by ideas from psychology philosophy and computer science cognitive psychology emerged Functionalism Chairs and Computers We can be functionalist about chairs What makes a chair a chair is not what it is made out of indeed you can have wooden plastic or metal chairs but that you can sit in it ie its functionality But there is no way that we can program a computer so that it becomes a chair chairhood is not a functionality that can be implemented by computer program Computationalism Cognition can be defined in terms of information processing Perception is taking in information from the environment MemoryBeliefsKnowledge is storing information Reasoning is inferring new information from existing information Planning is using information to make decisions Etc Informationprocessing can be done through computations Therefore cognition is computation Attention Introduction to Cognitive Science Overview A few more things on perception Attention Description of Attention What is attention What are some features of attention Models of Attention How does attention work What is the underlying mechanism Larger theory of attention What is attention for o How does attention fit into the overall cognitive architecture Constructive Perception Our visual system is not like a camera The Inverse Problem of Perception The Blind Spot Background knowledge effects perception Color constancy Size constancy Expectations wishes fears effect perception Flash experiment N rays Classroom attack The Attention Selection Problem Perception as Inference The Inverse Problem of Perception My visual system brain needs to deduce from the sensory data what is out there it needs to reconstruct what it was that led to the raw sensory data However this Inverse Problem is underspecified the same set of raw sensory can be generated by infinitely many different scenarios out there Hence I can only make an inference to the best explanation at all times The Selection Problem I can look at the same image and consciously choose to pick up on certain features as opposed to others Or 2 people can look at the same thing yet see something different Neisser s experiment of 2 overlaid baseball games Selection problem how do my visual system select what features to pick up on TopDown Visual Processing Many people believe vision or any other form of perception to be a purely bottomup process starting with the raw sensory input we recognize edges bars blobs and other basic optical features These features can then be combined eg via the recognitionthroughcomponent approach to recognize complex objects There is neurological evidence for such processing but there are also reasons to believe that the processing doesn t only go from raw image to interpreted image and that there are also topdown factors involved Imagery Neurological evidence suggests that I use same resources for direct perception as well as imagery seeing something in my mind s eye So why don t I get confused How do I know the difference between a direct perception and an imagined one Again a topdown process would easily explain this a bottomup process has a much harder time with this Active Perception It seems that contrary to popular belief perception is not a passive kind of data collection process passing interpreted snapshots on to higherlevel processing Rather it seems that higherlevel processes actively drive perception to a considerable extent in that it asks the visual system to look for certain things in the environment that the higherlevel processes need Attention Some Features of Attention Attention is selective I can t attend to everything that s going on around me Attention is divisible I can pay attention to multiple things multitask Attention can shift voluntary I can consciously switch attention from one thing to another Attention can shift involuntarily Sometimes my attention is drawn to something whether I want to or not Example A Cocktail Party Suppose you are at a Cocktail Party with many conversations taking place around you Selective Attention you can t attend to all conversations at once Divisible Attention you can converse while getting a drink at the same time Voluntary Attention Shift you can consciously shift your attention from one conversation to another Involuntary Attention Shift you suddenly hear your name being spoken by some people across the room Filter Models of Attention Filter theories of attention try to explain why attention is selective Attention is directly linked to perception Many models of perception see perception as some kind of process from raw data to interpreted information Eg vision raw retinal image gt edgesblobs gt shapes gt objects Filter theories of attention postulate that there are certain filters along this informational pathway that only makes certain information pass but not other Where along the pathway are these filters Early Filter Models Some people proposed that there are filters fairly early on in the pathway Evidence Dichotic Listening Task Set Up Subjects had headphones with different input on each ear Subjects had to pay attention to one ear s input Result Subjects could tell that something was being said in other ear as opposed to say music being played and also whether the speaker was male or female But they could not tell whether it was English or Russian or nonsense eg reversed speech Interpretation Basic physical features speech vs music pitch were processed but not much more Hence a filter fairly early on Objection to Early Filer Model Objection Certain discriminations presumably only happen fairly late In the pathway Your name being spoken at a cocktail party Neisser s overlapping baseball games So other models were proposed Late Filters Multiple Filters Movable Filters Leaky Filters Etc Much research done see book but no clear picture has emerged Resource Models of Attention Try a different tack focus on the divisibility aspect of attention Attention is seen as a resource a kind of mental energy or capacity that is in limited supply Hence while you can divide distribute your attention over several things you can t pay attention to everything Interference Experiment Subjects wore headphones had to pay attention to story told in one ear But subjects also had to pay attention to Condition A Words spoken in other ear Condition B Words presented on screen Condition C Pictures presented on screen Result Performance was worst for condition A best for Condition C B was in between Interpretation There was more task interference in condition A presumably because this involves some of the same neural resources BottomUp vs TopDown Filter views have more of a bottomup feel the filters let certain information through or not and they decide what you involuntarily become aware of Resource models have more of a top down view on attention you can voluntarily decide what to pay attention to whereas Another Tack Let s look at attention from yet another perspective and ask the following question Even though attention is divisible to some extent why is attention in most cases focused on one thing Especially in the light of our brain being able to do powerful parallel processing without any problem this seems like a interesting question Attention and the Body One possible answer is that if we were to attend to several things our body may have to react in incompatible ways eg one cognitive activity will want to make our body do one thing and another something different In short we re physically being pulled in two different ways which we can t physically accomplish So even though from a neurological computational or purely psychological point of view it should be perfectly possible for attention to be much more divisible than it is the physical consequences it has for our body make it so that we have to focus on just one thing Attention and the Mind Indeed maybe the fact that we have one mind or at least have a sense of having only one mind is a natural result of having one body In other words having one mind may be an illusion forced upon us by having one body Indeed sometimes the illusion can t be held up and some people seem to have multiple minds from clinicalabnormal psychology multiple personality disorder from neuroscience the two hemispheres of the brain doing quite different things Divided Attention and MultiTasking So how divided is attention really Certainly when highorder cognitive processes are involved it is next to impossible to divide attention can someone really do a calculus problem while having an intense debate with some about some complex issue And is the younger generation really better at multitasking Can multitasking be learned or trained An interesting finding Last year researchers from the University at Stanford published their findings on an experiment involving multitasking They found that selfdescribed heavy media multitaskers were worse multitaskers than selfdescribed light or non multitaskers This seems to go against the view that multitasking can be learned But what explains this MultiTasking as lntenNeaving Here s a possibility maybe much of what we call multitasking is really intertasking the interweaving of several tasks not unlike the way a modern computer simulates parallel processing by quickly going back and forth between different processes However such interweaving requires high amounts of logistics organization and focus So maybe heavy multitaskers are really people with less focus and more easily distracted Neural Networks Introduction to Cognitive Science Representations and Information Processing On the view of computationalism cognition is informationprocessing of some kind But what kind In particular how is information represented so that information can be Stored if important Recalled if relevant Integrated if helpful Updated if needed And all that quickly and efficiently The Jets and the Sharks Nam Gang Age Edugau an Man39ml Occupaziim at Art Jar 40 s JH Sing Pusher Al Jets 3039 LE Mar Burglar an Jets 3039s COL Sing Boukis Classical representation y e Jen or 1H Sing Buckie Mikc 1m 30 m Sing okl Information is organized Jim JP 20 s 1 Div lliilgliir 39 39 39 39 mg Join 2039s 5 Mur imam k9 a lm SyStem39 Item John Jets 20 s JH Mar Burglar by item Doug Jet 3039 Hs sang Buckle Lance Jet 20 s JH Mar Burglar Geurge Jets 2039s JH Dlv Burglar pane iei 2039 145 Sing Baukl Fred Jet 2039 HS Sing Push r Ordering and indexmg helps en 1m 2039 COL Sing Pusher Ralph 1m 30 in sing Pusher to 300955 Informat39on l hil Slimks 30 s COIL Mar Pusher Ike Sharks 3039 LR Sing Boolua Nick Sim 3039 HS Sing Push Dun Sharks zo39i COL Mm Burglar Still how would you flnd the Ned Shark 30 COL Mm Becki ii i Karl Sharks 4039s HS Mar Eouki Name Of that Shark Who 5 m Km sharks 1039 HS sing Burglar his 40 s has a HS education Ear Sharks 40 HS Mar Bur i u Rick Elia m as Div 131mm is Single and is a Burglar 7 Slimlu 30 s Mnri Pusher Nun shn 3039 Sing on ie DI Sharks 3039s Div Pusher Fl39gurz 21 39 L39 L From 1 I wu i a a i McClellnnd Reprinted by petmlulun Damian A p 439 r L um bH L UK Connection ist distributed representation A dl Fxgu 22 McClaUand s 1981 Jeu and Shark network Each gang membzr is rzpresented by on parsnn unit center am a mama to the appropriate name and Prawn 39 V t V Items are grouped linked by semantical associations Information is accessed by content rr uf rh Cagnilm 3mm Society Copyright 1981 by L L McClelland What are Neural Networks Neural networks are architectures inspired by the brain Neural networks consist of Nodes the neurons Connections between nodes the axons l dendrites Moreover Nodes have a certain activation In some networks it is simply on or off In other networks the activation can take on any value ie nodes can be more or less active Connections have a certain weight A positive weight means an excitatory connection negative is inhibitory In some networks the weight is simply either 1 or 1 o In other networks the weights can take on any value How do they work Through the connections activations and weights nodes will activate or deactivate other nodes If two nodes are connected by an excitatory connection then they will try and activate each other Eg the Ken node is positively connected with the Ken node the Sharks node the 20 s node etc If the connection is inhibitory the nodes will try and deactivate each other Eg the 20 s node is negatively connected with the 30 s node The more active a node is the more it will activate or deactivate the node it is connected to Similarly the bigger the weight the greater the effect 7100 quot I quotAr1quotname 90 L ART person 8 0quot aet quotCIydequot name mquot ClYDEperaon m 0 MIKEapersan E 505 aln205 40 mm 3639 Imus Tillh m mm 40 j W i 7 e 26 1 i Aquot I I n Jr 71 7 L M 20 30 40 EU 60 H 80 Cycdes Farm 23 The In vatjnn he Irma cycles at some DI the units in the Jen and Sharks nahlurk after the unit for Artquot nilm is activated by an external input A iva nns 30 40 Cyclas Eigqu 24 Th nctimiun vluea acmu cycles fur mm nnd pt nn unlm uf various punt H in pun Ickiun mn nnly via hz nrann Inlitn WordLetter Perception Network by McClelland and Rumelhart Q If v eyed 1 V witsai w 39 v we 39 Il i i i quot d l 2 I H E r L a F E H I 39 dwig i i m LJKLMNDPER 9 ETUVWXYZ 13 N9 39 fA Stet of letters ahnd gtqrgg tg 34 oiihiizufzeit positions ivutior C1D word actlvations unrk aLvion lattequot an m vations E in READ lLlFl calcine EFF l l l 1 i l Word Advantage Effect Letters are more quickly recognized in the context of a word that when presented in isolation This is a human cognitive phenomenon and the network can explain it Neural Networks and Cognitive Science Neural Networks have several features are of interest to cognitive science Neural plausibility Neural Networks have certain features in common with real brains Decentralized and Parallel processing Like our brain processing in different parts at the same time Pattern Completion and Correction Networks can handle incomplete or even incorrect input Automatic Categorization and Prototyping Categories emerge from dynamics of the network Moreover some instances are more prototypical of a category than others Graceful Degradation Changesdamage to the network only gradually effects performance Fuzzy boundary between declarative knowthat and procedural knowhow knowledge No clear distinction between data and processing InductionLearning Networks can learn and are able to pick up on statistical regularities Neural Networks and Computation Do neural networks fit into the scheme of computationalism Well it doesn t follow the classical computational schemearchitecture see next slide However like other computational systems features such as physical size location or matter are irrelevant to the implementation of a neural network it is only the abstract causal organization that matters Indeed neural networks can be implemented on your laptop so they are computational systems Moreover it can be shown that the computational powers of a neural network equal that of more traditional computational systems Neural Networks Computation vs Classical Computation Some important differences In classical systems computation is done serially one instruction at a time in a central location CPU but in neural networks computation is done in parallel and is decentralized The representations in classical systems are symbolic natural language sized such as cat on and mat whereas in neural networks they are subsymbolic activationsweights represent lowlevel features that are hard to put into words if at all In classical systems there is a clear separation between datasymbolsrepresentations on the one hand and algorithmsprocesses defined over those objects on the other hand But in a neural network this distinction is far less clear if existent at all And finally it can be hard to make sense of the computations of a neural network even if a network successfully accomplishes some task it can be hard to explain how exactly that happened this problem obviously mirrors the problem of relating neuroscience to cognitive science Creating Neural Networks The fact that it is so hard to understand how neural networks accomplish a certain task means that it is hard to create a network that accomplishes some task How many nodes to use How to connect them And what weights should be put on these connections Neural Networks Basic Learning Algorithm There are learning algorithms for neural networks that roughly work as follows Start with random weights And now keep doing the following until you get desired performance Present an input to the network The difference between the actual output and the desired output is the error Figure out how each weight is contributing to the error you can t do this in absolute terms but you can figure out if you should increase or decrease the weight in order to decrease the error Adjust the weights accordingly Generalization from Cases The interesting thing is that neural networks can generalize from their training set to never before seen cases Eg you can train it to recognize Bob in certain pictures of Bob and when you then give it a new picture of Bob it can recognize that as Bob as well Some Neural Network Examples Neural Networks have been trained to Recognize faces see handout Character recognition the US postal service uses neural networkbased software that can read handwritten Zip codes with 99 accuracy Translate written text into speech Net Talk Control power plants Control robotic limbs Predict behavior of aircraft in response to pilot s commands used in flybywire jets ac F e ec I I ognltlon Netw y R l ary CO I k b G e I Pals Num ion Ir mom 3 ms 5 a Bandquot 0 r hand C39sMA my Two39 Face use an cans a Luer 039quot nFm when 6 was new 6 Face Recognition Network Performance Achieved 100 accuracy on training set Of testing set set of never before seen of pictures 100 correct in determining face vs nonface 98 correct in determining name of familiar face ie face of someone who had a different picture in training set 81 correct in determining gender of unfamiliar face Over 70 correct in determining name of familiar face when 15 of picture was blackened Fm 9 Male subreglon 44 I 39 ifng C quot0 4 N r gpmmyplcm iamola face meander ambiguous face 40 c V V cariccliurei of your face V d a n n e K Ac vm ion Level A Quinine quot Sweet Cellquot Some Issues with Backpropagation Learning Cottrell had to present has training set 1000 s of times to the network to reach the performance it did Cottrell told the network what was right or wrong and reached in to change the weights supervised learning So how do we humans engage in oneshot learning usually unsupervised Do we need to tweak some parameters or is a completely different kind of learning called for Neural Networks More General current Shortcomings Neural Networks are good for perception and control with possibly some very shortterm prediction all pretty low level cognitive tasks Neural networks do not seem to be very good at more highlevel cognitive tasks such as complex reasoning or problem solving things that more traditional symbolic Al is good at Indeed while neural networks seem to have the potential to explain lowlevel cognition that characterizes much of animal cognition it is not clear how it can be used to explain humans higher order cognitive faculties such as logical reasoning or longterm planning and prediction Situated Cognition to the Rescue Suppose we situate the neural network in a symbolic environment and embody the network with perceptual apparatus and motor control so it can interact with its environment Maybe the network can learn the kinds of perceptionaction sequences l that end up manipulating symbols in fruitful ways and thus engage in higherorder cognition An interesting blend of Connectionism and Situated Cognition resulting in something that can be described as a Classical symbol system The Turing Test Introduction to Cognitive Science Can Machines Think The Behavioral Repertoire Argument Arguments for the possibility of thinking machines or intelligent computers often take the following form An entity is intelligent if it displays certain behavioral repertoires X Computers can be programmed to display those behavioral repertoires X Therefore computers can be intelligent Objections to this Argument While this argument is deductiver valid some people doubt it is wellfounded Holow Shel Objection Premise 1 is questionable Just because something displays certain behavioral repertoires X doesn t mean that it is intelligent maybe it just behaves as if Behavioral Shortcoming Objection Premise 2 is questionable I doubt that you can program a computer to do X Computing Machinery and Intelligence Turing 1950 I propose to consider the question quotCan machines thinkquot This should begin with definitions of the meaning of the terms quotmachinequot and quotthinkquot But instead of attempting such a definition I shall replace the question by another which is closely related to it and is expressed in relatively unambiguous words The new form of the problem can be described in terms of a game which we call the 39imitation gamequot The Imitation Game Machine Interrogator Human I believe that in about fty years time it will be possible to programme computers With a storage capacity of about 109 to make them play the imitation game so well that an average interrogator will not have more than 7 0 per cent chance of making the right identi cation after 5 minutes of questioning Alan Turing 1950 The Turing Test Today the Imitation Game is usually referred to as the Turing Test If a computer can play the game just as well as a human then the computer is said to pass the test and should be declared intelligent Some Initial Observations on the Turing Test The Turing Test attributes intelligence purely on verbal interactions Is that ok Well physical characteristics size weight agility etc don t seem to be relevant as far as intelligence goes so that seems right However shouldn t we have to open up the computer program and see how it works to make this kind of determination Then again do we ever open up other human beings to determine whether they are intelligent Hmm maybe Turing has a point The Turing Test and the Behavioral Repertoire Argument Indeed Turing s strategy seems to fit the behavioral repertoire argument we started with Specifically Turing s version would be Anything that behaves in such a way that it passes the Turing Test is intelligent Computers can pass the Turing Test Therefore computers can be intelligent Why The Whole SetUp But if we re after a certain behavioral repertoire why does the Turing Test have such a complicated setup Why did Turing pit a machine against a human in some kind of imitation game That is if Turing is trying to determine machine intelligence purely based on the interactions the interrogator is having with the computer s responses to certain questions why not have the interrogator simply interact with a machine see what it is or is not able to do and determine whether or not the machine is intelligent based on those interactions So why not The SuperSimplified Turing Test Interrogator lt gt Machine Answer Bias The mere knowledge that we are dealing with a machine will bias our judgment as to whether that machine can think or not as we may bring cebrtain preconceptions about machines to the ta e For example knowing that we are dealing with a machine will most likely lead us to raise the bar for intelligence What it can t write a sonnet Aha I knew it It s not intelligent By not knowing who or what is on the other end such biases and raisingof thebar is eliminated in the TuringTest OK but still why not The Simplified Turing Test Interrogator lt gt Machine or Human Note this is exactly how many commentators talk about the Turing Test Level the Playing Field Since we know we might be dealing with a machine we still raise the bar for the entity on the other side being intelligent In fact I bet that with this setup probably a good number of humans would be declared to be machine Through his setup of the test Turing made sure that the bar for being intelligent wouldn t be raised any higher or lower for machines than we do for fellow humans Thus the Turing Test levels the playing field between humans and machines A Definition of Intelligence Some commentators see the Turing Test as a definition of intelligence And many people have subsequently commented on the shortcomings of the Turing Test as a definition of intelligence This definition would amount to some kind of philosophical behaviorism But most of us think that while being intelligent causes the behavior it does not consist in the behavior Also this definition would be a real sloppy definition Who is the interrogator How long is the conversation What is the conversation about How does the interrogator decide Not a Definition Turing himself clearly did not intend to propose a definition of intelligence In his paper Turing readily acknowledges that one could have intelligent beings not being able to pass the test simply by not having a human like intellect May not machines carry out something which ought to be described as thinking but which is very different from what a man does This objection is a very strong one but at least we can say that if nevertheless a machine can be constructed to play the imitation game satisfactorily we need not be troubled by this objection A Sufficient Condition for Intelligence Most commentators therefore interpret Turing s statement as saying that if a machine passes the Turing Test then it is intelligent ie that passing the Turing Test is a sufficient condition for intelligence since intelligence is a necessary condition to pass it but not a necessary one and hence it is not a definition ln logic We have P gt But not I gt P Same Sloppiness And A Question But as a sufficientcondition for being Intelligent the Turing Test suffers from some of the same problems as before such a criterion would still amount to a subjective judgment based on imprecisely defined behavioral criteria nshort this seems to be a rather sloppy criterion Why would Turing not exactly known for his sloppiness propose such a sloppy test Cheap Tricks Eliza A psychotherapist program developed by Joseph Weizenbaum in 1966 Eliza used a number of simple strategies Keywords and precanned responses Perhaps I could learn to get along with my mother gt Can you tell me more about your family Parroting My boyfriend made me come here gt Your boyfriend made you come here Highly general questions In what way Can you give a specific example Eliza and the Turing Test Many people conversing with Eliza had no idea that they weren t talking to a human So did Eliza pass the Turing Test Or is itjust easy being a psychotherapist Eliza wasn t really tested in the format that Turing proposed Still it is interesting that humans were quick to attribute humanlevel intelligence to such a simple program Maybe in a real Turing Test a relatively simple cor ieruter program can trick the interrogator as we The Loebner Competition Modern day version of the Turing Test Multiple judges rankorder multiple humans and multiple computer programs from most likely to be human to least likely to be human Loebner has promised 100000 for the first computer program to be indistinguishable from a human Thus far Loebner is still a rich man occasionally a judge will rank a program above a human but on the whole the judges systematically rank the humans above the computer programs An OK Test After All Apparently it is quite difficult to pass the test When put to the real test interrogators can see through superficial trickery So it seems we could say that if something does pass the test then there is at least a good chance for it to be intelligent In fact if we are turning this into an inductive argument anyway the sloppiness of the test isn t a huge concern either we can now simply adjust our confidence in our claim in accordance to the nature of the conversation So is this maybe what Turing was saying Contrary Views In his paper Turing goes over a list of Contrary Views on the Main Question Machines Can t be conscious Can t some specific ability eg be kind use language properly enjoy strawberries and icecream etc Can t make mistakes Can t be creative Can t learn Can t do other than what they re told Lady Lovelace Machines can t be conscious have feelings or have emotions Turing s reply to this objection is that this response is most likely the result of a generalization of machines that we have encountered in our lives so far all of which do indeed lack these qualities However it is not clear that in the future machines couldn t have these qualities So without any further support for the truth of these claims this objection really doesn t work Also how do you know if a machines isn t conscious We don t really know this for other humans either OK Machines can do X but they can t do Y fill in anything for Y Turing s reply to objections of this kind is that we should be careful not to require unreasonably much from the machine before we declare it to be intelligent Many people are bad at playing chess or writing poetry so if some machine can t do this that doesn t automatically mean that the machine isn t intelligent In fact by using the Turing Test Turing wanted to make sure that the bar wasn t raised any higher for machines than we do for fellow humans Also in saying that machines can t do Y we may once again be generalizing from existing machines rather than make any kind of argument for the in principle impossibility for machines to do Y Lady Lovelace Objection Probably the most common objection to machine intelligence Machines can only do what we tell them to do which is usually followed up by the program isn t intelligent the human programmer isl This takes several forms Machines can t make mistakes Machines can t be creative Machines can t learn or adapt Machines Can t Make Mistakes Of course this is a weird kind of objection to the possibility of machine intelligence because what is so unintelligent about not making mistakes Anyway it is a proper objection to the claim that a machine could be able to pass the Turing Test because supposedly a machine would always give itself away by its lack of mistakes or by how inhumanly fast it is able to correctly solve math problems Response Machines Can make Mistakes However machines sometimes do make mistakes due to a bug in the program say In fact it is easy enough to program a machine such that it does give the wrong answer to certain kinds of questions and so that it does take a long time to give that answer Of course since we don t want a machine to make mistakes we try to ensure that machines don t make mistakes So we rarely see machines making mistakes But that doesn t mean that all machines are like that we may once again be making a bad generalization based on the kinds of machines we see around us A Paradox How can Machines make Mistakes How can machines make mistakes given that they follow some deterministic routine algorithm or program Turing This paradox is easily resolved It is indeed true that machines do exactly what their underlying mechanism routine or program dictates them to do However as a result of that routine machines may end up getting the wrong answer make the wrong decision or do the wrong thing Machines can t be Creative The point about machines not being able to make mistakes is often related to this one the common ground being that machines can only do what they are told to do The mistake in this objection is again that while this statement is true from the perspective of the underlying program it is not clear that a machine couldn t do anything new or creative when looked at from a higher level Indeed look at Deep Blue Deep Blue beats every human in chess but that would be impossible if Deep Blue couldn t do any better than any of its programmers Machines can t Learn Again the underlying thinking here is that a machine can only do what it is told to do and hence not do anything new and hence not learn However we know this claim to be false because there are plenty of machines that do learn Eg there are machines that learn to play chess learn to walk learn to diagnose diseases etc Again we only look at the machine from the underlying mechanicalprogramming point of view Yes the machine will follow some program and not deviate from it However as a result of doing so it can learn In general then we can program a machine to make mistakes do creative things and learn There is no contradiction there Analogies between Machines and Humans One could say that humans are like machines subject to strict laws of nature we can t do anything other than what nature forces us to do And one could say that humans are programmed by other humans through education etc Indeed just because the programmer is intelligent does that mean that the program is automatically not intelligent How does that follow Another Question If Turing s point of his article was to propose a test or criteria for intelligence then why are none of these objections about the validity of this test In particular given the nature of the test one would expect a whole bunch of Hollow Shell objections and as we saw that is indeed what we got from the commentators due to tricks or due to the subjective nature of the judgment something can pass the test without being intelligent But at best Turing s own list of objections seem to be Behavioral Shortcoming objections In fact some of these objections don t even seem to really and directly address the behavioral repertoire that would be required to pass the test Indeed almost all of Turing s paper seems to be a defense of the possibility of machine intelligence in and of itself So what was Turing s real point of the paper Passing the Test Also if Turing really would be more concerned with Behavioral Shortcoming Objections then why is it that Turing hardly makes any effort to argue that machines can pass the test In his paper Turing merely lays out the principles of computation and discusses the notion of universal computation but Turing never directly addresses how this relates to passing the test Presumably Turing thinks that passing the test requires nothing more than some kind of information processing ability which is exactly what computers do Yet Another Question But if that is true then it seems that Turing could much more easily have argued as follows Intelligence requires nothing more than some kind of information processing ability Computers can have this information processing ability Therefore computers can be intelligent Indeed this is exactly how most proponents of Al make the argument today So Why didn t Turing make this very argument Why bring in the game at all In Summary The Contrary Views make it clear that Al opponents think machines can t do certain things but Turing thinks they can But the Turing Test doesn t seem to be able to shed any light on this issue itjust doesn t seem to be at the center of this whole debate 80 If Turing really wanted to propose a test for machine intelligence why not propose a test that much more directly and objectively tests certain abilities that both parties can agree on to be relevant to intelligence And If Turing wanted to defend the possibility of machine intelligence why even bring up such a sloppy test at all Indeed What was the point of Turing s paper My Answer I propose that the convoluted setup wasn t merely a practical consideration to eliminate bias in some strange game but rather the point of his article which is that ifwe put a label intelligent being on other human beings based on their behavior then we should do the same for machines So right or wrong our use of slapping the label intelligent onto things human or othenNise should at least be consistent Imitation Game vs Turing Test In other words I think it is likely that Turing never intended to propose any kind of test for machine intelligence let alone propose a definition Interesting fact In his original article Turing uses the word pass or passing 0 times test 4 times and game 37 times The Turing Test as Harmful In fact I believe that seeing Turing s contribution as laying out a test is harmful The harm is that we have been thinking about the goal of Al in these terms and that has been and still is detrimental to the field of Al Eg In Essentials of Artificial Intelligence Ginsberg defines Al as the enterprise of constructing a physical symbol system that can reliably pass the Turing Test But trying to pass the test encourages building cheap tricks to convince the interrogator which is exactly what we have seen with Eliza Parry and pretty much any entry in the Loebner competition This kind of work has advanced the field of Al and our understanding of intelligence exactly zilch So I think we really should no longer refer to the Turing Test as the Turing Test How to Read Turing s Paper So what did Turing really mean Ultimately this is an issue of history and not an issue we as cognitive scientists need to be concerned about Better questions to ask are What if anything can we learn from Turing s paper What would be a fruitful interpretation of his paper Well there are many interesting parts of the paper especially in Turing s responses to the Contrary Views I also believe that seeing Turing s paper as laying out a genuine test is harmful not helpful Instead I believe a fruitful reading of his paper is to see the Turing Test as a statement about the use of the word intelligence Artificial Flight and Artificial Intelligence Imagine going back 100 years when the Wright Brothers had their first flight We can imagine people say Well but that s not real flight There is no flapping of the wings But over time we realized that it is from the standpoint of using concepts that help us think and make sense of the world around us a good idea to consider airplanes as really flying Maybe the same is true for intelligence In Turing s Words The original question Can machines think I believe to be too meaningless to deserve discussion Nevertheless I believe that at the end of the century the use of words and general educated opinion will have altered so much that one will be able to speak of machines thinking Without expecting to be contradicted Alan Turing 1950 Memory Introduction to Cognitive Science What is Memory and What is it for Memory is the capacity to retain information as picked up from the environment Memory forms the basis for learning reasoning prediction and decision making it allows a cognitive agent to better deal with situations it finds itself in In other words memory makes one an experienced agent allowing for higher chances of survival Types of Memory Sensory Memory Sensory Buffer Very shortterm on the order of tenths of seconds a few seconds at best Unconscious Modalityspecific ie specific to a sense visual auditory etc Capacity determined by senses ShortTerm Memory Shortterm on the order of seconds to minutes to hours without distraction Conscious Modalityspecific Low capacity 56 items or more LongTerm Memory Longterm some memories last lifetime ie basically indefinite Unconscious Often less modalityspecific see next slides Large indefinite capacity Sensory Memory We seem to have a kind of buffer specific to each of our senses that is able to hold raw sensory data Iconic Visual Memory Experiment by Sperling Subjects shown 3 rows of4 letters for short time 100 msec In free recall subjects remember about 4 out of 12 letters Prompted to report any one specific row subjects could report all 4 out of 4 letters So all information is there but by the time information enters conscious recall much information is lost Echoic Auditory Memory Echo of what was said or heard Phone number Spelling ShortTerm Memory Typical shortterm memory task in Cognitive Psychology recall of word lists Some experimental results Subjects can hold only about 5 or 6 words or other items in memory Chunking combining multiple items into one item can help Items can be are held in place through rehearsal phonological loop Distractions or other changes in attention or cognitive tasks quickly eliminates short term memory Lists of items are better remembered through elaboration eg imagine taking walk from dorm room to classroom and associating landmarks along the way with items in list gt mnemonics LongTerm Memory Cognitive psychologists have proposed many different types of or distinctions within long term memory Episodic specific vs Semantic abstract Declarative knowthat vs Procedural knowhow Explicit can be consciously recalled vs Implicit not How distinct are these different types Is there a clear qualitative difference eg are they represented differently or in different places in our brain Or are these differences more quantitative ie is there some continuum along which these different types are located Short Term vs Long Term In fact some cognitive psychologists wonder whether there is even a clear distinction between shortterm and long term memory Maybe there is really just one kind of memory ie one kind of way in which memories are created and stored and what we call shortterm and longterm are just different ways in which memories behave themselves over time Evidence for more Fundamental Distinction Double Dissociation Cases of amnesia shortterm memory is fine but doesn t make it to longterm memory Patients whose longterm memory is ok but can t form shortterm memories Drugs affecting shortterm and longterm memory performance in clearly distinct ways Possible neurological distinctions Neural activation shortterm vs neural connections longterm Hippocampus shortterm vs neocortex longterm Relationships Between Types of Memory Early Model Sensory ShortTerm LongTerm Memory L Memory V L Memory Attention Rehearsal Interference Working Memory Later models of memory added the notion of Working Memory Working Memory can be seen as a work bench useful for all kinds of cognition What would be placed on such a workbench would obviously be effected by attention and interference And rehearsal would be a way of keeping the items on the workbench although a rather uninteresting cognitive activity Indeed reasoning planning etc also make use of working memory and are much more Interesting Relationships Between Types of Memory Later Model Sensory ShortTerm LongTerm Buffer L Memory Memory Memory RecallRetrieval Computer Similarities Notice how the modernday computers have very much this architecture Sensory buffer keyboard buffer mouse buffer Working Memory RAM Plain Memory ROM Indeed much of the recent focus on working memory was probably inspired by the computer Another way to look at this Thinking about informationprocessors from a practical engineering point of view we found it useful to have buffers a workspace and long term memory Therefore is it all that surprising that Mother Nature came up with something similar Computer Dissimilarities Computer working memory seems much bigger than human s working memory Maybe working memory requires attention and attention is as we saw earlier not very divisible possibly due to human cognition being embodied and situated Or is there much more going on than meets the conscious eye Shortterm vs Working Mmeory And what happened to shortterm memory Some researchers have suggested that working memory and shortterm memory are the same or at least that working memory uses shortterm memory working memory shortterm memory attention On the other hand while many shortterm experiments may turn out to be working memory experiments eg the 5 or 6 item shortterm memory capacity limit may well be a working memory attentional limit there still seems to be a kind of memory that lasts many hours or even a day or two definitely longer than working memory but shorter than longterm memory Memory and Sleep and Dreams Maybe shortterm memory is where we collect a day s events and when we sleep and dream we sift through this and assimilate some of it and discard the rest Lack of sleep certainly decreases memory performance actually it decreases pretty much all physical and mental performance Perfect vs Imperfect Memory Computer longterm memory is perfect human longterm memory is imperfect Barring unintended hardware and software problems if we tell a computerto remember something it will not only remember it but it will do so completely accurately V th the exception of a few people humans both forget things as well as misrememberthings Memory is selective Little of what we are consciously aware of ever make it to longterm memory Memory is leaky Much of what was once in memory fades away False memory Sometimes what we remember didn t happen at all But is there may a good reason for having such imperfect memory Selective and Leaky Memory A possible good reason for memory being selective and leaky is that only certain things may be deemed important to remember as far as the agent s functioning and survival goes Indeed if everything was remembered then maybe there is too much information to sift through in order to make quick decisions Experiment Subjects had to watch video Half had arms in freezing ice water while watching Half had not The first group had better recollection of video Possible explanation events were deemed more important as situation was one subjects would like to avoid in future Indeed typical longterm memories are often unusual or emotional events vacation trip performance fight etc False Memories Famous experiment by Elizabeth Loftus Subjects saw video of car accident A week later subjects were brought back One half was asked How fast was the car going when it bumped into the other car Other half How fast was the car going when it crashed into the other car Second half estimated remembered speed higherthan first half They also recalled seeing glass laying on the road after accident even though there was none Courts are relying less on eyewitness testimony But what s the point of having false memories Memory is Constructive Rather than to ook at memory as a place of storage maybe It IS more useful to think of memory as a process of reconstructionwhen usefulorneeded Cues from the environment will trigger experiences from memory that agent will try and apply to current situation Since no two situations are exactly alike some amount of abstraction will have to take place in order for such a generalization to work 80 agents will try to match their memory with current situation gt constructive memory and on the perception end gt constructive perception Such a matching process while useful may produce false memories however 80 maybe false memories are the unfortunate sideeffect of a useful process Vision Introduction to Cognitive Science Overview The Myth of Perfect Perception Object Perception Space Perception Motion Perception Integration The Checker Board Shadow Illusion The Checker Board Shadow Illusion The Myth of Perfect Perception Many most people regard perception as something that is under normal circumstances perfect That is as long as it s not dark we re not drugged we re not wearing funnycolored glasses and we re not looking at a highly artificially constructed scene that provokes some specific visual illusion we see things exactly as they are So the idea is that we are looking at the world through a perfectly clean window as if our eyes are a perfect camera our ears perfect microphones etc False Perception must be lnferential Constructive Perception We can look at Perception as an Inversion Problem What is happening out there causes there to be raw sensory data My visual processing system brain thus needs to reconstruct what it was that led to the raw sensory data it needs to invert this causal process However this Inverse Problem is underspecified the same set of raw sensory data can be generated by infinitely many different scenarios out there Hence I can only make an inference to the best explanation at all times All perception is a construction Our perceptual apparatus works kind of like a detective given the clues raw sensory data it tries to tell a plausible story The Blind Spot There is a region in the back of the eye s retina where there are no light receptors So we don t see anything in this region But we don t notice this close one eye right now do you see a gap No What is going on The brain takes a best stab at what might be going on In fact it isn t so much filling in as that still sug est that the rest of the perception is pe ect but the brain works with the raw data it is provided with and constructs a perception from that Background Knowledge Background knowledgebeliefs drive perception believing is seeing Color constancy Cutouts of trees and donkeys made of same color material but perceived as different color in accordance to our beliefs Size constancy even though approaching objects would seem to get bigger and bigger going by the mere raw 2D projection on retina we never perceive them as growing in size because we know or at least based on background experience have strong reasons to believe that they remain the same size Expectations Fears and Wishes Corridor Experiment subjects were told to walk through dark corridor and stop when they saw a flash of light almost invariable they stopped at some point or other but there never was actually a flash of light Monsters under the bed in the closet etc Blondlot Case NRays Devils Angels Aliens Jesus quot quot54quot 3a Clearwater FL USA A htt9iesusgancom Jesus Pattern Bias It is important for our survival to recognize patterns in nature Patterns allow us to make predictions and control our lives Moreover it is better to be safe than sorry better to recognize a nonface as a face than not to recognize a face Hence our brains have pattern bias when there may or may not be a pattern it is likely to say that there is a pattern Object Perception Object Perception and Object Recognition Normally object perception is talked about as object recognition However as we just saw perception is constructive it parses up the world in a way that is in some way useful Indeed rather than tables and chairs being out there and that we subsequently recognize as such it may be better to simply talk about stuff being out there that we parse perceive as tables and chairs That said from now on I ll talk about recognition 5 The Problem of Object Recognition We accurately and effortlessly recognize objects in our environment even though the raw image on the retina changes dramatically due to vana onsin Position Size Color Surface texture Viewpoint Exemplars of a given class How do we do that RecognitionbyComponents Biederman inspired by Marr defined 36 geons that qualitatively I differ along 5 dimensions crosssectional curvature symmetry aXIs curvature size variation aspect ratio Straight gt Curved Cross section Equal gt Cross section greater RampR Refectional Aspect Ratio Symmetry Equal Axis greater gt Aspect Ratio 4 I g arEtlplyo Straight gtCuned Sweeping Axis Constant gt Expanding Crosssectional Size Constant Exp amp Contract Crosssectional Size RecognitionbyComponents Many most All objects can be specified as spatial arrangements of primitive volumetric components called geons geometric ions The 36 geons can generate over 150 million 3geon objects like an alphabet uucf f 31quot J 33 A 3 Yquot 1 3 f M 2 J P 39 391 I y l fro View invariance of geons each different geon has its own key properties at the level of the 2D full primal sketch geons can be identified directly from image these nonaccidental properties are invariant over different views 0 iv a d 39BP SKS CYUN JEHS39 o 3 31 ul 3 axallr cogss O mmin slrmg t 10333 3913 h I Ji 9 g 39 i i I71 C 0 n nrYvvzsm 0 P pmnln cume cdgn39 lc39z39i 1 i 39 8 oxar arm unnccs 2 Lirng Y wxzccs 1quotiit ud i ldcu nil0 103 with a few exceptions called accidental views highly unstable Space Perception Space Perception Object recognition is about perceiving what Space perception is about perceiving where Cues Another myth about perception Many people think that space perception is solely due to the fact that we have two eyes However if you close one eye you can still perceive 3D quite well eg you have little problems navigating your environment How is that possible It turns out that there are many cues that we use to perceive space Convergence Accommodation Occlusion Etc Types of Cues What does the cue require Monocular or Binocular Does the cue require one eye or two eyes Optical or Ocular External or Internal Is the cue related to the visual image optical or is it related to the physiology of our eyes ocular Static or Dynamic Does the cue work for a single snapshot or does it need the image to change Information Present in Cue What does the cue tell us Absolute or Relative Does the cue tell us exactly where eg how far away some object is located or does it tell us where that object is in relation to other objects Quantitative or qualitative If a cue is relative does it say that something is for example exactly twice as far way as something else or does it merely tell us that something is further away than something else Note all absolute cues are quantitative What is the effective range of the cue Some cues only work well for close by objects whereas others work better with a far distance Convergence Eyes rotate to fixate objects of interest Angle of inward looking relative to head midline varies systematically with distance 80 how much the muscles have to rotate the eyes provides a clue as to how far the object is Accommodation Lens bulges as we focus on nearer objects Tension in ciliary muscles varies systematically with distance Occlusion Closer objects hide or partially hide distant objects gt hidden objects are seen as more distant Familiar Size Familiar size If an object s size is known then the object s distance can be determined Relative Size Relative size If two objects of known similar sizes occupy different visual angles then the object that occupies the larger visual angle must be closer Height in the visual field Distant objects are higher in the visual field gt objects that are higher in the visual field or more generally closer to the horizon are perceived as more distant Atmospheric Perspective Particles of dust water and pollution cause us to see distant objects as less sharp gt Fuzzy objects are perceived as being more distant note how in this example light and dark gives us a big clue Linear perspective Lines that are parallel in the scene converge with distance on the image plane Texture gradient Elements that are equally spaced in a scene occupy smaller visual angles as distance increases gt surfaces with elements that occupy smaller and smaller visual angles are perceived as receding in depth 7 Motion Parallax Distant objects move more slowly ie have lower optical velocity than nearby objects T i 39 A 4 v r quotM m 39 Stereopsis Hold L index finger at arm s length Hold R index finger at 12 arm s length Close R eye and align both fingers with distant point D quot Fixate distant pont Close L eye and open R eye Fixate distant point again Binocular disparity the difference in the images on the two eyes due to 50 I separation of two eyes in space difference in depth of objects in scene Motion Perception The problem of motion perception How are properties of the 3D world structure and motion of objects specified in the changing 2D image as further complicated by eye head and body movement Biological motion Czechslovakia 19603 Black Light Theatre Gunnar Johansson 19703 Point light walkers N n ms m g u NL t tt 13s Ni 7k m3 NV 4 r wu LAW 1 rv Biological motion Main results Static image perceived as meaningless configuration of dots like a constellation of stars Dynamic image perceived as person in motion in as little as 200 ms Subjects could accurately discriminate different activities walking dancing doing pushups different gaits walking jogging running limping different genders different individuals self vs roommate weight of a lifted box 4 n in final ii The Barber Pole Illusion Rotating Poles The Correspondence Problem 15 lt2 A Perceived as B Perceived as Rigidiy Rotating Plasticaily Deforming The correspondence problem is to figure out which pixel from an earlier snapshot corresponds with which pixel from a later snapshot This is much easier to do in A as the end points of the straight ends can be put into such a correspondence For B we don t know which point in the later snapshot corresponds to which point in the earlier snapshot Hence our brain gets confused Correspondence Problem for Barber Pole Easily t gt This is the original identified spin to right point but there is point no easy way to 39 identify or track Re it as such identified point whoops Or do we correspond the whole red patch with the whole other red patch Integration The Problem of Integration We perceive the world as one whole But we get bombarded with separate pieces of sensory information all the time How do we put it all together The temporal correspondence problem was one example of this problem of integration But there already was a spatial correspondence problem for static visual input due to the fact that we have 2 eyes which pixels from the left image correspond to which pixels from the right Also notice that integration not only takes place over separate pieces of visual information that also across our different senses Some more examples of integration follow The Binding Problem Neuroscientists have found some very specific and separate parts of the visual cortex performing some very specific tasks Discriminating shapes Discriminating colors Discriminating textures Etc But how does all this information get together again How for example is it that we perceive a red pen next to a blue notebook How did the red go with the pen and the blue with the notebook Why not the other way around Why not 4 different objects Spatial Cue Integration If there are multiple cues then it makes sense that they get somehow integratedcombined In particular some cues only give us information about 1 particular object not about all objects present in the visual field So the visual system somehow integrates multiple sources of depth information into a single coherent consistent representation How does it do that An Example of Integration Binocular Disparity tells us relative distance Convergence gives absolute distance but for fixated object only Binocular disparity distF 2distC Convergence distF 10 Modified weak fusion distC 5 The Ames Room Cue Conflict The Ames Room shows that cues can be in conflict and this may well happen in natural conditions as well where one cue says one thing yet another cue says something different How does the visual system deal with this Possible ways of dealing with this Dominance assign some kind of priority to certain cues over other cues Compromise Take some kind of average Example Dominance Perceived depth is determined by the most dominant cue available Less dominant cues are ignored Perspective information dominates familiar size Integration and Conflict Probably the question of how cue conflict is solved is best seen in the context of integration as a whole conflict or not But lots of research still needs to be done here Attention Introduction to Cognitive Science Overview A few more things on perception Attention Description of Attention What is attention What are some features of attention Models of Attention How does attention work What is the underlying mechanism Larger theory of attention What is attention for o How does attention fit into the overall cognitive architecture Constructive Perception Our visual system is not like a camera The Inverse Problem of Perception The Blind Spot Background knowledge effects perception Color constancy Size constancy Expectations wishes fears effect perception Flash experiment N rays Classroom attack The Attention Selection Problem Perception as Inference The Inverse Problem of Perception My visual system brain needs to deduce from the sensory data what is out there it needs to reconstruct what it was that led to the raw sensory data However this Inverse Problem is underspecified the same set of raw sensory can be generated by infinitely many different scenarios out there Hence I can only make an inference to the best explanation at all times The Selection Problem I can look at the same image and consciously choose to pick up on certain features as opposed to others Or 2 people can look at the same thing yet see something different Neisser s experiment of 2 overlaid baseball games Selection problem how do my visual system select what features to pick up on TopDown Visual Processing Many people believe vision or any other form of perception to be a purely bottomup process starting with the raw sensory input we recognize edges bars blobs and other basic optical features These features can then be combined eg via the recognitionthroughcomponent approach to recognize complex objects There is neurological evidence for such processing but there are also reasons to believe that the processing doesn t only go from raw image to interpreted image and that there are also topdown factors involved Imagery Neurological evidence suggests that I use same resources for direct perception as well as imagery seeing something in my mind s eye So why don t I get confused How do I know the difference between a direct perception and an imagined one Again a topdown process would easily explain this a bottomup process has a much harder time with this Active Perception It seems that contrary to popular belief perception is not a passive kind of data collection process passing interpreted snapshots on to higherlevel processing Rather it seems that higherlevel processes actively drive perception to a considerable extent in that it asks the visual system to look for certain things in the environment that the higherlevel processes need Attention Some Features of Attention Attention is selective I can t attend to everything that s going on around me Attention is divisible I can pay attention to multiple things multitask Attention can shift voluntary I can consciously switch attention from one thing to another Attention can shift involuntarily Sometimes my attention is drawn to something whether I want to or not Example A Cocktail Party Suppose you are at a Cocktail Party with many conversations taking place around you Selective Attention you can t attend to all conversations at once Divisible Attention you can converse while getting a drink at the same time Voluntary Attention Shift you can consciously shift your attention from one conversation to another Involuntary Attention Shift you suddenly hear your name being spoken by some people across the room Filter Models of Attention Filter theories of attention try to explain why attention is selective Attention is directly linked to perception Many models of perception see perception as some kind of process from raw data to interpreted information Eg vision raw retinal image gt edgesblobs gt shapes gt objects Filter theories of attention postulate that there are certain filters along this informational pathway that only makes certain information pass but not other Where along the pathway are these filters Early Filter Models Some people proposed that there are filters fairly early on in the pathway Evidence Dichotic Listening Task Set Up Subjects had headphones with different input on each ear Subjects had to pay attention to one ear s input Result Subjects could tell that something was being said in other ear as opposed to say music being played and also whether the speaker was male or female But they could not tell whether it was English or Russian or nonsense eg reversed speech Interpretation Basic physical features speech vs music pitch were processed but not much more Hence a filter fairly early on Objection to Early Filer Model Objection Certain discriminations presumably only happen fairly late In the pathway Your name being spoken at a cocktail party Neisser s overlapping baseball games So other models were proposed Late Filters Multiple Filters Movable Filters Leaky Filters Etc Much research done see book but no clear picture has emerged Resource Models of Attention Try a different tack focus on the divisibility aspect of attention Attention is seen as a resource a kind of mental energy or capacity that is in limited supply Hence while you can divide distribute your attention over several things you can t pay attention to everything Interference Experiment Subjects wore headphones had to pay attention to story told in one ear But subjects also had to pay attention to Condition A Words spoken in other ear Condition B Words presented on screen Condition C Pictures presented on screen Result Performance was worst for condition A best for Condition C B was in between Interpretation There was more task interference in condition A presumably because this involves some of the same neural resources BottomUp vs TopDown Filter views have more of a bottomup feel the filters let certain information through or not and they decide what you involuntarily become aware of Resource models have more of a top down view on attention you can voluntarily decide what to pay attention to whereas Another Tack Let s look at attention from yet another perspective and ask the following question Even though attention is divisible to some extent why is attention in most cases focused on one thing Especially in the light of our brain being able to do powerful parallel processing without any problem this seems like a interesting question
Are you sure you want to buy this material for
You're already Subscribed!
Looks like you've already subscribed to StudySoup, you won't need to purchase another subscription to get this material. To access this material simply click 'View Full Document'