Cognitive Science 1 - Master Study Guide

Hebbian Learning

- "Neurons that wire together fire together" - When learning occurs, physical connections are being made in the brain - Hebb's theory was the first testable hypothesis - Eric Kandel proved Hebb's work - Forms a neural network because of the neuron connections that are made while we learn - "When an axon of cell A is near enough to excite a cell B and repeatedly or persistently takes part in firing it, some growth process or metabolic change takes place in one or both cells such that A's efficiency, as one of the cells firing B, is increased." - Hebb, 1949

Ambiguity

- "One morning, I shot an elephant in my pajamas" - "I made her duck" - These examples require context or "common sense" to decipher the actual meaning.

Phantom limb phenomenon

- "The term refers to the phenomenon in which people who have had a limb or other body part amputated continue to feel the missing part." - It may ache, it may tingle, it may swing, it may bend. It may feel the way it used to feel when it was still part of the body. - "But how can a person continue to feel a body part that isn't there? One hypothesis offered for how these feelings come about is that the brain fashions a kind of map based on the sensations one experiences while living with one's body. A portion of the map remains, according to this hypothesis, even when a part of the body that contributed to producing it is lost." - "But it is also the case that people born without limbs, and therefore deprived of the daily experience of living with them, also feel them, as an itching or a numb sensation or a reaching-out impulse. An explanation offered for this odd occurrence is that the brain's map of body parts is not simply learned through experience but that it is innate, consisting of networks of neurons (individual nerve cells) laid down before birth." - These networks link various regions of the brain, enabling us to sense where our limbs are, to feel pain, and to learn from our experiences (Melzack, 1990). In other words, the brain comes equipped with a kind of blueprint of where the body parts are going to be so that even if they fail to develop, their "ghosts" continue to exist in the brain." - IMPORTANT SCIENTISTS IN RELATION TO PHANTOM LIMB PHENOMENON -- V.S. Ramachandran

Analogy

- 4 Stages of Analogical Reasoning: -- COMPREHENSION of the target problem -- REMEMBERING a similar source problem for which the solution is already known -- The source and target are COMPARED AND MAPPED -- The source problem is ADAPTED TO PRODUCE A SOLUTION to the target problem - Analogies are represented in the left prefrontal cortex of the brain

Embeddings

- A categorical feature is represented as a continuous-valued feature. Typically, an embedding is a translation of a high-dimensional vector into a low-dimensional space.

Norman Doors

- A door where the design tells you to do the opposite of what you're supposed to do. - A door that gives the wrong signal and needs a sign to correct it. - Doors that are difficult to use because their affordances do not match how they should be used.

The Default Network

- A network of interacting brain regions known to activate "by default" when a person is not in task -- Aka: when people are not doing anything they will begin mind wandering 3 important things to know about it - Comes upon really fast (fraction of a second) - Thinks outside of the here and now - Cognitive achievement: only species who can do this But is it good? - mind wandering mind is a unhappy mind (emotional cost) - Happiness is in the present moment

Paperclip Maximizer

- A thought experiment of an AI that is told to try and maximize the amount of paperclips in the world - The AI's optimal solution would be to ultimately get rid of all people in the world to make room for the paperclips OR turn all humans into paperclips - Shows that even though there might not be malicious intent to an AI's program, it can ultimately destroy humanity - Also reflects how AI doesn't take into account humanity's norms

Turing Test

- A way of evaluating artificial intelligence. A person would blindly talk to one machine and one person. They have to guess which one is the real human. If the person guesses incorrectly 50% of the time, the AI machine "passes" the test.

The Exemplar approach to categorization

- Accumulation of already encountered instances - If we encounter a new example that resembles something we already encountered, we assign it to that category

Action potentials

- Action potential initiates at the axon hillock; starting at the axon and propagates downwards in one direction from the cell body to the axon terminals -- Axon hillock: where the axon is met ---- Like the calculator of the neuron; collects all the excitatory and inhibitory values ---- Evaluates action potential - Electrical part is only contained within axon -- Chemical part happened at the cell body & at axon terminals;ends of the neuron

Types of bias

- Algorithm bias: problem within the algorithm - Sample bias: problem with the data - Prejudice bias: reflects existing prejudices, stereotypes, and faulty assumptions - Measurement bias: problem with how data was measured - Exclusion bias: when important data point(s) left out

Pre-attentive Visual Processes

- All within our sensory memory- We process things around us pre-attentively, automatically. Does not require conscious effort or attention. - Pre-attentive examples: notifications (red bubble), grouping text with size and width

Explicit Representation

- An explicit representation of a particular aspect of the visual scene implies that a small set of neurons exists that responds as a detector for that feature, without further complex neural processing. -- Loss of these neurons causes subject to be unable to consciously perceive that aspect directly. ---- Eg. Achromatopsia (loss of color perception), prosopagnosia (loss of face perception) - A necessary but not sufficient condition for the NCC to occur. -- NCC: minimal set of neuronal events that gives rise to a specific aspect of a conscious percept.

Consciousness as an Emergent Property

- An inquest to understand the nature of intelligence, we have found it to be like an emergent property of parallel processing of complex systems - Consciousness can be regarded as this as well.

A.I. vs. ML vs. NN vs. DL

- Artificial Intelligence: Any simulation of human intelligence in machines - Machine Learning: Systems able to learn from past data to make accurate decisions/predictions - Neural Network: Artificial neurons, learn relationships between input/output - Deep Learning: Neural nets with large number of interconnected layers, learn complex relationships

IBM's Watson

- Artificial intelligence that went up against the top players of Jeopardy - Struggled with similar ambiguous responses because considering the context of the question - Not dealing with structured database, learning language

Perceptron

- Artificial neuron that takes in multiple weighted inputs and returns a single output

Knowing and Perceiving

- As we perceive things we learn about them - Results in us talking about all kinds of knowing, understanding, and learning in terms of seeing, hearing, and feeling

Training neural networks: minimizingloss

- Attempt over and over again until you get the smallest loss function possible

Attention and Binding

- Attention can be divided into 2 types: -- Rapid, saliency-driven and bottom-up -- Lower, volitionally controlled and top-down - Binding: process that brings together rather different aspects of an object/event, such as its shape, color, movement and so on.

Analytical Engine

- Babbage's creation of a machine that could add, subtract, multiply, and divide. Regarded as the "general-purpose programmable computing machine."

Bias

- Bias can be picked up from data used in NLP methods. That bias will be represented and conveyed in the system.

Cognitive Reserve

- Bilingualism exercises a level of cognitive control when code-switching between 2 languages by requiring mental effort in inhibiting the other language - Kim et al. (1997) study

The Mind-Body Problem

- Brain is physical and can be studied objectively - But mind consists of subjective phenomena - What is the relationship between the two? - Monism: mind and body both physical. Made of the same thing. - Dualism: mind and body are separate entities. - Consciousness is a property of spiritual minds and is not open to scientific explanation.

Mental Context reinstatement

- Bringing the contexts back to mind with mental imagery - Context-dependent effect: remembering more in the same context, but you can get almost as much information if you remember the context - Studying vocab in two phonetically similar languages (Swahili & Chinyanja), minimize interference by studying them in different contexts -- First: encoded (randomized subjects and let them explore)

Language areas of the brain

- Broca's Area (left frontal gyrus) - generally for language production - Damage causes Broca's Aphasia (speech is halting, grammatically incorrect, meaningful words but in a non-fluent & telegraphic manner) - Can be helped with Melodic Intonation Therapy - Wernicke's Area (left temporal gyrus) - generally for language comprehension - Damage causes Wernicke's Aphasia (language flows with relative ease but is disordered and incomprehensible), impairment in understanding of speech, fluently connect words grammatically but lack meaning

Gradient Descent

- Calculate gradient of loss function vs. each weight to find direction of steepest descent - New weight: weight - (learning rate * component of gradient for that weight)

The Prototype approach to categorization

- Central member that is average of category and is used to compare (most prototypical member). - Some members of the category are more related to the category than others. - The typical dog, typical summer day, etc.

Neurons and the Brain They Call Home

- Cerebral cortex: the outer layer that directs our motor and cognitive functions - Nervous system: two parted, complex system that coordinates actions and sensory information by transmitting signals to and from different parts of its body. -- Peripheral nervous system: ---- It is made up of afferent nerve fibers, which carry signals from the senses to the brain, and of efferent nerve fibers, which carry messages from the brain to muscles and glands that tell them to move or to secrete. -- Central nervous system ---- Consists of the spinal cord and the brain - The cells that make up the nervous system are called neurons -- "Neurons vary in size, from very small to perhaps 3 feet long and are made up of three parts: the cell body, or soma; branching systems of dendrites; and an axon. ---- The soma contains the cell nucleus, which in turn contains the genetic material (deoxyribonucleic acid [DNA] and ribonucleic acid [RNA]) of the neuron. ---- Dendrites are thin ribbons that extend from the body itself, like strands of bubblegum pulled from the main mass. A membrane surrounding the soma receives information from other cells, and little bumps, or dendritic spines, on the dendrites do the same. ---- The axon is a single, longer branch that also originates in the soma. Near its end it branches into little terminal boutons or end feet. -- When neurons are not at rest, they are "firing," producing what are called nerve impulses. These are electrical charges that travel along the axon to the axon terminals. The terminals are the transmitters of information to the membranes of other neurons, not in the course of touching them—because typically they don't—but rather across a small space or gap, called a synaptic cleft. - Glial cells (glial is Greek for glue) make up the white-space in the brain and have the same neurotransmitter receptors that neurons do, and are necessary to the processes neurons carry out: "neurons are utterly dependent on [them] to fire their electrical impulses and to pass messages to one another across synapses." - "The cortex consists of four major sections, or lobes: the frontal (memory/emotions), the parietal (sensory), the occipital (visual), and the temporal (hearing). The whole is divided into halves, the cerebral hemispheres, which are connected by a thick bundle of fibers called the corpus callosum."

Noam Chomsky

- Challenged behaviorism by implementing linguistics into psychology. Claimed that linguistic psychology could not be represented under the behaviorism framework. - Hypothesized universal grammar - Innate in brain and can be applied to any language - Principles and parameters - basis of universal language - Switches are set depending on exposed language.

First and second language acquisition

- Children don't solely learn language from their environment. They make sentences they've never heard before (suggests language innateness). - When children learn languages, they acquire rules as well. - In the process of distinguishing sounds, children also categorize them. - The younger, the better at distinguishing sounds in language. - At some point in 1st year, babies distinguish crying from other vocalizations of feelings and needs - cooing stage. - Babbling stage - making sounds in real sentences "mama," "baba," etc. - Levels of grammar develop at certain ages. - Real-life social contact is necessary to develop language. Second: - The earlier the age of exposure to a second language, the better acquisition of it. - When bilingual people hear/speak one language, both languages activate. There is some mechanism that cancels the other one out. There is some sort of competition between the two languages. - Bilingualism increases executive control - ability to inhibit language. - There is a period where concepts are translated to sound. - Brain uses a wide variety of its parts for word retrieval but some areas are specific to concepts actions faces, etc.

CRUM

- Computational representational understanding of mind - Thinking can be best understood in terms of representational structures in the mind and computational procedures that operate on those structures (Thagard) -- Thinking is performers by computations operating on representations -- Example: MATHEMATICS ---- Numbers = representational ---- Adding, subtracting, multiplying, dividing = computational - We (our minds) think in concepts (schemas), propositions (declarative knowledge), rules (procedural knowledge), analogies (reasoning, problem solving), images (visual imagery), logic -- Locations/ Represented in the brain: ---- Concepts: basal ganglia, prefrontal cortex ---- Propositions: left hemisphere ---- Rules: basal ganglia, prefrontal cortex ---- Images: occipital lobe

Brain imaging techniques

- Computer Axial Tomography (CAT) - Magnetic resonance imaging (MRI): measures soft tissue structure my aligning protons with a powerful magnet - Positron emission tomography (PET) - functional magnetic resonance imaging (fMRI) a version that shows changes in brain activity over time - transcranial magnetic stimulation (TMS): a noninvasive method to excite neurons in the brain: weak electric currents induced in the tissue by rapidly changing magnetic fields

Expert Systems

- Computerized advisory programs that imitate the reasoning processes of experts in solving difficult problems - Knowledge base systems: computers designed to become experts in a given field of interest. - Two parts: knowledge base and engine. - Must have enough data to make a decision.

Birth of AI: John McCarthy and Claude Shannon

- Conference that occurred in 1956 - Claude Shannon: originators of concept of coding theory (understanding communication systems, entropy) - John McCarthy: combined info theory with the idea of logic History of AI - 1940-1950: Early Days: Circuit model of brain & neurons Turing suggested that designing intelligent machines was possible Nothing was done about it, no programming done - 1950-1970: Excitement: 'Look, Ma, no hands!' Reinforcement Learning (RL) with programs that played checkers fairly well People accepted that AI really was logically doable - 1970-1990: Knowledge-based approaches Found out logic was not very useful, nor practical Have systems not only reason, but have a lot of knowledge ex. Programming info about how flights work (distance, time, location, etc.) when building system that allows you to book a reservation on a flight Useful to an extent, but not in all domains aka Common Sense Reasoning (how do you program knowledge about things falling due to gravity) - 1990-2012: Statistical Approaches + subfield expertise Techniques made to combine knowledge with learning so you dont have to put all knowledge into system (graphical models) Initial knowledge (prior) and then the system can experience more and "learn" Useful, but required a lot of computing - 2012-Present: Excitement: 'Look, Ma, no hands again?' Decided to go through with big data, big compute, and deep learning Used text and image data Computing became cheaper and more useful (cellphones)

Harvard's Implicit Association Test

- Conjunction Fallacy: bank teller example. Basically implicit bias. - We tend to put things in certain categories based on what we already know about them.

Neurocomputational Theory of Consciousness (Paul Thagard)

- Consists of 3 elements: - Biological/Neurological -- Cell damage in the brain affects consciousness - Electrical -- Epileptic seizures affect electrical signals. -- Different electrical signals while sleeping, which is a subconscious state. - Chemical -- Chemicals like caffeine affect consciousness - alertness.

Linguistic Relativity - Frames of Reference

- Coordinate systems used to specify the location of objects with respect to other objects -- Relative: relative to a person -- Intrinsic: using elements of the object itself to describe the location -- Absolution: as implied (e.g. A is west to B)

Phrenology

- Created by Franz Gall - "Science of the mind" - Pseudoscience that is based off of the shape and size of one's skull and equating it to intelligence - Correlated size of eyeballs with the ability to learn; studied brain functions - Localization of function - Essentially thought that brain structure could determine one's personality - 3 basic principles: -- Brain is the sole organ of the mind -- Basic characteristics and intellectual traits are innately determined (nature vs nurture argument) -- Differences in individuals = differences in their brains

Cognitive disorders

- Damage to the dorsal stream: -- Akinetopsia (motion blindness) -- Optic ataxia (deficit in visually guided reaching) - Damage to the ventral stream: -- Achromatopsia (color-blindness) -- Prosopagnosia (face-blindness to familiar faces; affects 2.5% of the world's population)

Neural reinstatement of study context

- Decoding reinstatement of VR contexts (through MRI scans, you can encode which environment the subject is in) - New sample of dual context participants performed same learning protocol, but day 2 testing was in an MRI scanner - When patients mentally reinstate themselves, they shed a facilitation (conquering reinstatement) - Representational similarity analysis - context reinstatement template (high RSA means you're reactivating the original learning contexts well, which means higher recall) - Mental reinstatement of VR contexts improves recall of a foreign language - Cortical reinstatement of context-specific representations - Distinctive environmental contexts may help learners compartmentalize knowledge, reducing interference - Reactivate visual representations

Dark Patterns

- Def: Intentionally using the principles of design against the user and making difficult designs to lead the user to a certain action. Features of interface design crafted to trick users into doing things they might not want to do, but benefit the business in question. - Ex. deleting an amazon account - Roach Motel: Design to make it easy to get in but hard to get out. Example of this is specifically deleting amazon - Coined by Harry Brignull

Affordances

- Def: The perceived and actual properties of the thing, primarily those fundamental properties that determine just how the thing could possibly be used. It's both how we perceive what we can do with objects and their actual properties. - Ex. button affords pushing, a switch affords flipping, a knob affords rotating - Digital affordances: this concept of affordances applies not only to physical objects but also to digital interfaces. - Ex. iOS7 button looks more like a button.

Brain Machine Interfaces (BMI)

- Direct communication pathway between the brain and some external device - Training an intelligent assistant to move

Visual Pathways

- Dorsal (the "where" information, spatial information) - Ventral(the "what" information, information about objects) - Superior collucus (subconscious visual processing; responsible for 10% of visual processing)

PET

- Dynamic/Functional Imaging - Positron Emission Tomography - Requires injection of the radioactive isotope - Measures blood flow in the brain - Bad Temporal resolution (30 second delay) - Good Spatial resolution - Invasive

The ENIAC

- Electronic Numerical Integrator and Calculator used in WWII. - Was a very large machine that completed specific calculations for beneficial strategies.

A Neural Substrate of Prediction and Reward Science

- Even when you can only predict the reward but it hasn't occurred, dopamine still spikes - Temporal difference signal = how good a status is based on how far away in time it is to actually receiving the reward (ie. ACTUALLY being in that good state) - The example given - First time = didn't know reward was going to occur (ie. no prediction) so there was no early spike in dopamine - Second time = got a sign that reward is going to occur (ie. had prediction) (eg. bell rings for dog means it's about to get food) so there was an early spike in dopamine when prediction occur (ie. when the bell rang)

Neural Language Models

- Example = predicting words based on all the words that it has seen before - Huge vector that represents with embedding

Fairness, Accountability, and Transparency

- Fairness - select representative data sets with good variance - Accountability - be prepared for consequences resulting from biased model - Transparency - explainable AI, SHAP (Shapley Additive Explanations). Not a black box

Alan Turing

- Father of AI and theoretical computer science - Cracked the enigma code used by the Nazis, helping allies understand enemy wartime communication - Designed Turing Machine, a mathematical model that defines an abstract machine - anything that can be performed with algorithms can use this machine - Designed turing test to evaluate artificial intelligence and how good it is

Recursion

- Feature of human languages - Using a finite set of rules in our grammar, we can potentially build an infinite number of sentences

Localization of Function

- Franz gall - 4 lobes and their function -- frontal lobe: motor activity, planning/decision making -- parietal lobe: sensations - touch, pain, temperature -- temporal lobe: auditory system -- occipital lobe: visual system

What does each lobe do?

- Frontal Lobe: Reasoning, Planning, Speech, Movement, Emotion & Problem Solving - Parietal Lobe: Movement, Orientation, Recognition, Perception of stimuli - Occipital Lobe: Visual Processing - Temporal Lobe: Perception, Recognition of auditory stimuli, Memory & Speech

fMRI

- Functional magnetic resonance imaging - functional/ dynamic - Shows changes in brain activity over time - Picks up and detects oxygenated blood in the brain - Very strong magnets (measured in teslas) - Bad temporal res., great spatial res. - Non-invasive

Why we can train neural networks with many layers cheaply?

- Geoff Hinton's work resulted in ingenious simplifications and engineering of artificial neural networks - 2014: came up with way to solve networks by using non traditional computer architecture - Rather than using a computer, he used graphics units (cheaper, you can get more of them, cheap hardware) - Made it available to everybody, so ML took off

Glial cells

- Glial cells are much more abundant than neurons - Albert Einstein's brain had more than the average number of Glial Cells - 3 different types -- Oligodendroglia: provides insulation (myelin) to neurons ---- makes up 76% of glial cells -- Astrocytes: star-shaped cell ---- provides physical and nutritional support for neurons ---- 17% of glial cells are astrocytes ---- "mother cell" ---- help digest parts of dead neurons along with the microglia -- Microglia: digest parts of dead neurons ---- make up 7% of glial cells ---- "janitor cells" (they clean up after the neurons and get rid of unnecessary waste) ---- Diseases such as Parkinson's and Alzheimer's are connected to a microglial cell deficiency

Reinforcement Learning

- Goal is to maximize some kind of reward over many time trials - Model learns to achieve a goal by interacting with its environment - Agent does an action (moves to another state) and updates its policy depending on the consequences - Reward = immediate primary sensory feedback from a particular state - Value = long-run total reward the agent can expect to make from a particular state; agent tries to maximize value function

A.I. as machines that think and act

- Goals of AI: We want to make machines that... - Think like humans - Think rationally - Act like humans - Act rationally - Developing intelligent machines allows us to explore the principles that make minds work - And test theories about the properties of the mind

Case Study: Prescription Bottles (Clear RX/ScriptPath)

- Good design of Prescription Bottles: Safety push down feature, hard to mistake, transparent - Bad design of Prescription Bottles: difficult to open & find important info., inconsistency, hard to read, confusing (looks similar to others) - Improvements with Clear RX: easy to identify, logo smaller, important info. larger, one label, color ring to differentiate, attached to the info sheet, improved readability, and text

Retrograde Amnesia vs. Anterograde Amnesia

- H.M.'s Hippocampus case study: brain is structures in a from that allows it to be responsible for consolidation & transfer of information from STM to LTM -- Had damage to hippocampus = anterograde amnesia: inability to retain new information subsequent to damage (but able to recall past) -- Retrograde amnesia: can't recall past events

Theory building

- Help us think about old problems in new ways - To construct theories is everywhere in science - Nonnative species: - Comparing nonnative species to invading armies is harmful - Demilitarization: talk about intervention instead: control expansion, limit population growth, engineer new ecological balance

Multi-layer feed forward neural network

- Hidden layer is where most of the thinking happens - Learns by backward error propagation. Weights shift based off of trial and error - Multiple neurons connected to a wider network. Forming a network, input layer → hidden layer → output layer

Case Study: crime as beast/virus

- Highlights the misuse of a metaphor - Thibodeau and Bordosiky want to see how wording would affect participant's outcome - Found that punishment is more severe when crime is viewed as a beast - Known as a metaphoric snowball - only need to use the metaphor once to determine what further information you are likely to seek - Beast as source domain, crime is a beast entails that they are inherently immoral and cannot change, who is hunter and who is prey and what is the punishment in this idea - Virus as source domain: inherently neutral, can change, can be reformed/cured, doctor has a responsibility to help not harm

Embodied Cognition (from Chapter 11 Reading)

- Holds that all aspects of the mind rely on the interaction of the body with its environment.

Sternberg's Approach

- How does sternberg regard intelligence? A: Goal directed adaptive behavior What is a more detailed belief surrounding sternberg's regard of intelligence? A: Intelligence is when someone is able to make the most out of their environment and everyday life, not just in a laboratory setting. - What is Sternberg's triarchic theory of intelligence(1985)? A: There are three different components of intelligence. First is being able to make the most out of/ adaptations to an environment, for example making changes to a marriage one is unhappy in. Second is the ability to deal with novel situations and be able to create an automatic response to it, for example driving. Third is involves intelligences to the performance of tasks. - What is the automatizing of a task? A: The ability to do a thought consuming task with only "part of your mind". This is developed with repeated practice. - What are the three components involved in thinking and performance of tasks- a component within the triarchic theory of intelligence? What are these processes known as? A:(1) metacomponents- processes used to plan, monitor and evaluate problem solving. (2)- performance processes, implement commands of metacomponents. (3)-knowledge-acquisition components, gaining new knowledge. - What is the formulation of intelligence testing? A: Gaining insight. - What is tacit knowledge? Is it another part of intelligence like sternberg is describing? A: Tacit knowledge is basically common sense. It is another part of intelligence like sternberg is describing. What is sternberg's belief on tacit knowledge? A: His belief is that tacit knowledge is something that success in life depends more on rather than explicit knowledge. - What is the view held explicitly/implicitly on knowledge? A: One is already born with a certain capacity of knowledge, but the environment can nurture and encourage more of that knowledge. Intelligence is seen as a condition for progress. It is widely accepted in the discipline.

Gardner's Approach

- How is Howard Gardner's perspective of intelligence related to localization of function? A: Gardner has a more explicit approach to the localization of function. - What is Gardner's definition of "frames of mind"? A: The diversity of human range and talents. - What is Gardner's view of intelligence in ordinary living? A: All intelligence are interdependent and work in harmony. - What does Gardner focus on when trying to arrive on an understanding of the nature of intelligence? A: He looks at unusual ways or underdeveloped abilities in order to arrive at an understanding the nature of intelligence. - What is savant syndrome? A: A rare condition where one is gifted in a particular region(being able to play complex piano pieces by ear) but is not more intelligent than the average person in general. - What is on the set of intelligence by Gardner? A: 1. Linguistic intelligence-poets demonstrate this. 2. Musical intelligence-composers demonstrate this. 3. Logical-mathematical intelligence- scientists and mathematicians demonstrate this. 4. Spatial intelligence- the ability to see the same path despite being thrown disorientating obstacles. 5. Bodily-kinesthetic intelligence- demonstrated by actors who have timing with their body or dancers. - What have studies shown in relation to language and music with brain damage? A: If there are lesions on the brain, language loss may mean that musical ability remains unimpaired. Musical ability loss may mean that language is unimpaired. - What is spatial intelligence? A: Spatial intelligence is being able to recognize different visual experiences of any entity even if the perspective has changed. - What is the relationship between brain damage and spatial intelligence? A: It is distinct from any other type of intelligence. There are more difficulties in spatial representation and orientation if brain damage has occurred. - What is intrapersonal intelligence? A: Intrapersonal intelligence is the ability to access one's own feelings in life. Being conscious of one's own emotions. - What is interpersonal intelligence? A: Interpersonal intelligence is the ability to make distinctions in emotions among other people. For example, identifying when others are in particular moods. - What is something generally accepted about intrapersonal and interpersonal intelligence? A: It is something necessary to be successful in life and follows a more practical type of intelligence. - What is a deficiency in conducting interpersonal relationships maybe known as? A: This follows the lines of Asperger's syndrome and mindblindness. - What is another type of intelligence posed by Gardner to include in his set? A: naturalist intelligence. The ability to recognize plants, animals, natural environment. This is adopted in materialistic society by being able to distinguish between brands. This follows the newer concept of evolutionary psychology. - What are another two types of intelligence posed by Gardner to include in his set? A: Existential and moral intelligence. They have not been as clearly discussed yet. Not as important right now. - What does research on visual-spatial intelligence point towards? A: Orientation and size are performed better by the right hemisphere rather than the left. - What happens with long-term motor training(i.e. practicing musician) to the brain? A: There is a strong association between structural differences and music status- there is an increase volume of gray matter in musician's brains. - How should the findings of sternberg and Gardner be viewed? A: They should be looked at as important foundations of research-not as explanations of intelligence.

Single-cell/multi-cell recording

- Hubel and Wiesel created single-unit recording and trans-synaptic tracing - Multi-Cell: brain recording -- Measures neuronal activity -- Measures multiple neurons at once -- Invasive - Single: brain recording -- Measures neuronal activity -- Invasive -- Good spatial and temporal resolution

Inductive Bias Problem

- Humans enter the world with predetermined inputs like color vision and motor movements - Our nature vs nurture is nothing like how machines evolve - We are not born as a blank slate

Koch & Crick's framework for consciousness

- Hypothesized the thalamus to be the possible NCC (Neural Correlate(s) of Consciousness). - Thalamus connects with multiple parts of the brain. Together, as a whole, consciousness is given. - In conclusion, consciousness may be the product of specialized consciousness neurons. - Done with studies with optical illusions.

Turing Machine

- Hypothetical machine that can solve any problem given an algorithm. - Operates on tape of 1's and 0's (tape is hypothetically infinite). Its value was the abstraction that served to present the possibilities of artificial intelligence. - Descendant of the Analytical Engine.

Linguistic Relativity

- Idea that if we all speak different languages, we all think and perceive the world differently (relativity) - Notice that determinism & relativity are like both sides of the same coin

Linguistic Determinism

- Idea that the language that you speak determines how you perceive the world (Sapir-Whorf Hypothesis)

Searle's Chinese Room argument and the main replies

- If a machine can convingly simulate an intelligent conversation does it necessarily understand? - References a database without truly understanding to spit back out responses - Syntax is not equivalent to semantics - Stimulation of the mind does not = understanding the thing Suppose: - We built a machine which behaves as if it understands chinese - It passes the turing test (ex: convinces a chinese speaker that they are talking to another chinese speaker) - Searle sits in a room with an English version of the algorithm the machine uses. He uses the textbook to spit out a chinese answer without actually understanding Act Intelligently is NOT think intelligently - There is no difference between Searle and the computer program in this scenario. Each follows a program, receiving an input and producing an output. Replies 1.) Systems Reply: a.) Concedes man in room doesn't understand Chinese b.) However, if man is but a part (a CPU) of a larger system which includes databases, memory, etc., we can argue that this complete system understands Chinese c.) So, although the man himself doesn't understand Chinese, the whole system does 2.) Robot Reply a.) Concedes man in room doesn't understand Chinese b.) However, argues that normal people only know language due to external experiences c.) So if we attach a robot system to the man in the room such that he can interact with external stimuli, the robot as a whole will be able to learn and understand Chinese, similar to normal people 3.) Virtual Minds Reply 4.) Intuition Reply 5.) Other Minds Reply 6.) Brain Simulator Reply a.) Redesign AI to simulate neuronal activity of a native Chinese speaker when speaking Chinese b.) This way, the AI can understand the same way the Chinese speaker does (literally)

Interactionist theory of meaning: concept of grasping

- Imagination (like perceiving is embodied and structured by our interactions with the world and society through our brain, bodies, and mind - Interactionist theory of meaning is a result - Neural exploitation is a big part of human cognition

Virtual memory palace

- Imagine yourself in a familiar place or landmark - Method of Loci - discovered that if the visual images of things to be memorized were placed in sequence along imagined journey they are easily recalled later - Order doesn't matter - Higher number = more error - Performance correlated with object placement - Spatial memory is likely a scaffolding - Volitional placement of the objects resulted in improvement in verbal recall - Some subjects in the control group might have used object-location binding strategy to bolster list encoding - VR paradigms provide a unique opportunity to examine the influences of environmental contexts on memory and harness the power of memorable contexts to scaffold with a frame of reference - Conflicts: practicality and affordability - Context could include state of mind as well

Unconscious / Implicit Bias

- Implicit bias is an unconscious and unavoidable element of the human experience.

Animats

- Insect Robots/Microbots: -- Examples of robotics & locomotion - The Subsumption Architecture -- Brooks' insect robots use layers of direct, reflex-like coupling between sensors and motors

Implications of AI: Technical

- Intra and Inter state levels -- Dialogue between the two -- Intrastate set of issues with Project Maven - Project Maven -- Lots of Google employees quit in protest -- Contributing to precision military operations that end up in the deaths of people on the other side of the planet -- Many employees did not know what they were working on -- Google had to take itself off the project -- Using pixels to identify targets -- AI, AI ethics as intrastate issues --- When Russia "accidentally" leaked nuclear bomb information, it leads to strategic, stability, and perception issues - Research and Scientific Challenges -- Adversarial Input ---- When images are pixelated, it is harder to identify objects ---- When this technique was used in project maven, lives were at risk -- Data Poisoning ---- You can inject alternative data to manipulate results -- Model theft ---- Easy to purchase, but often there's mistakes or slight flaws ---- Hard to tell what has changed within the pixels - Responsible AI development and regulations -- Existing regulations are insufficient to ensure responsible AI development - Reward hacking -- Distributional shifts -- Doesn't work in the real world that much -- Negative side effects (effects of emergent properties) -- Brittleness of systems itself -- Interpretability in details - Algorithmic persuasion and its cognitive effects. -- Relying on AI and ML to make your decisions -- Seeing emergent properties -- Systems learn from each other (interplay) -- Learning beyond human intelligence

Lab Experiment vs. Survey vs. Field Study

- Lab Experiments: In human-centered design, lab experiments are often in the form of usability tests of a prototype. - Typically involves giving participants tasks in order to measure user errors, satisfaction, etc. - Con: not realistic - Surveys: Questionnaires to measure and categorize attitudes, or collect self-reported data. - Better for representing larger samples to measure differences between groups of people and to identify change over time. However, many variables are affected. - Field Study: Going out into the "field" where the system is actually being used to investigate how people use it. - Capture things that you may not have outside of a lab, very realistic. - Con: time-intensive. - All methods are flawed, depending on the study, we may try to use multiple methods to fill in any gaps.

Difficulty of language

- Language is a complex social process. - Tremendous ambiguity at every level of representation. - Modeling is AI-complete (requires first solving general AI). - Speech acts ("Can you pass the salt?") - Conversational implicature ("The opera singer was great, she sang all the notes.") (sarcasm, etc.) - Shared Knowledge ("Warren is running for president").

Implications on language

- Language makes use of concepts. Concepts are what words, morphemes, and grammatical constructions express. Indeed, the expression of concepts is primarily what language is about. If we are right, then: - 1. Language makes direct use of the same brain structures used in perception and action. - 2. Language is not completely a human innovation. - 3. There is no such thing as a "language module." - 4. Grammar resides in the neural connections between concepts and their expression via phonology. That is, grammar is constituted by the connections between conceptual schemas and phonological schemas. Hierarchical grammatical structure is conceptual structure. Linear grammatical structure is phonological. - 5. The semantics of grammar is constituted by cogs—structuring circuits used in the sensory motor system. - 6. Neither semantics nor grammar is modality-neutral. - 7. Neither semantics nor grammar is symbolic, in the sense of the theory of formal systems, which consists of rules for manipulating disembodied meaningless symbols.

Single Context learners

- Learned language in the same virtual world

Dual Context learners

- Learned languages in different virtual worlds - Advantage in interference (high intrusions, less likely to recall word in different language) - Subjects who studied the two languages in two different virtual contexts showed better long-term retention than subjects who studied the languages in the same context

Learning and Memory

- Learning in a nervous system requires a change in the biochemistry of synapse, synaptic plasticity - Sea slug case study: differences in the brain for short and long term memory -- Short term memory- linked to functional changes in existing synapsis/connections -- Long term memory- associated w/ the change in number of synaptic connections

Stages of iteration

- Like the Scientific Method- We design something, prototype it, and iterate quickly - failing fast so we don't spend too much time and effort on a design that may not work. - In Human-Centered there are examples of iteration and evolution (ex. Landline to cell phone)

Prototype effect

- Listing: prototypes listed first - Direct rating: prototypes rated higher than others. - Asymmetric reasoning - non-central members are judged as being more similar to central members than the other way around. - Contrastivity

MRI

- Magnetic resonance imaging - structural/static - Uses a strong magnet to measure soft tissue by the alignment of protons - Provides static images - Can be used repeatedly - Bad temporal resolution - Good spatial resolution - Non-invasive

How design and cognitive science are related?

- Main disciplines overlapping: Psychology, Computer Science, Social Science - Human-Centered Design (def: → observation → idea generation → prototyping → testing →) applies findings (about cognitive science, cognitive science is about understanding the brain/mind and its capacity) and what we know about the mind to build human-centered products and services. Understanding the human mind is fundamental so we can better design for people.

Q-learning

- Maximize Q function at every step - Q(s, a) depends on the reward r and the resulting state s' from doing a while in state s - For given state s, choose a that maximizes Q - Move to s' and repeat until you reach a terminal state (goal achieved or failed)

Challenges of VR-based learning platforms

- Memory is context-dependent -- Will knowledge acquired in a virtual environment transfer to real world knowledge? ---- (Protocols in social settings) - Recall -- A changed environment can hurt recall -- Think scuba divers -- Imagining the original study setting can help retrieval of information

Memory & Design

- Memory- information is input, processed, encoded, retrieved, and output. The limitations of human memory have many implications for design. - Working memory: Design- 7 +-, Chunking ex. verification code 417 722 - LTM: Recognition Over Recall, Design- ex. pictures to match the concept

Cognitive Metaphor theory

- Metaphor is a cognitive process developed in early and used in everyday reasoning - Metaphor reasoning manifests in language - Structuring one concept(the target) using another concept (source)

Artificial Neuron

- Models biological neurons (has cell body, threshold function) - differences: does not have myelin - Threshold Linear Unit by McCulloch-Pitts - Inputs 1's or 0's - Mapping from neuron - model of biological neuron - Types of activation functions that influences output - Threshold, piecewise linear, step, etc

Current use of AI

- Movie AI (computer-like brains with human bodies) - News AI (Elon Musk speaking on society being overran by "robots") - Real AI (Google Search, machine translation in language, using Maps to navigate)

Neurogenesis in the Hippocampus

- Neurogenesis: the creation of new neurons - Hippocampus highly effective to cortisol -- Chronic stress leads to decreased neurogenesis - Not all forms of stress is bad: we have high levels of cortisol when we wake up and decreases throughout the day

Mirror Neurons

- Neurons that fire both when action is performed and when it is only observed - Important in imitation and language acquisition, understanding the actions of other people, putting yourself in other people's shoes, empathy, etc. - Some researchers also propose that they simulate observed actions and thus contribute to theory of mind skills

Universal Grammar

- Noam Chomsky proposes a set of rules intended to explain language acquisition in child development - Something must be innate for language acquisition - not just learning from the environment - Everyone has innate "light-switches" that turn on and off depending on the language they exposed to - Main language groups: - SVO (English, Russian, Mandarin) - 42% - SOV (Japanese, Hindi) - 45% - VSO (Tagalog, Biblical Hebrew, Irish) - 9%

EEG

- Non-invasive imaging technique that measures electrical activity of neurons and has high temporal resolution - Measures gross electrical activity of the entire brain - Event related potential (ERP) is a response to a representation of a stimulus - Attaches electrodes to the scalp - Invented by Hans Berger - Good temporal res., bad spatial res.

TMS

- Noninvasive - Transcraneal Mangetic Stimulation - Uses magnetic fields to stimulate nerve cells in the brain - Good temporal resolution, bad spatial resoltuion

Language Typology

- Organizing languages based on criteria of interest -- Helps to isolate the contribution that comes from the child

Cognitive Decline

- Our brain's memory declines around the 3rd decade of our lives - As we age our memory worsens

Bermudez Reading Overview

- Overview: talking about how neuroscience has been important in cog sci recently and how it became important - Cognitive scientists used to think that study of the functional systems of the brain should be done separately from studies of the actual hardware that lets those things happen - top down approach of thinking about cognition - Marr thought in order to understand visual system, must identifying algorithms the system uses to process information (ex processing basic characteristics and spacing of objects in an environment). Top down - Leslie Ungerleider and Mortimer Mishkin used bottom up approach

Enactive Cognition

- Perception is a form of embodied action - Cognitive structures and processes emerges from an agent's sensorimotor engagement with an environment Perception consists in perceptually guided action - Held and Hein kitten experiment -- Actively walking kitten that received visual stimulation as a result of self-movement developed normal depth perception and paw-eye coordination -- Passively suspended (but moving) kitten that received identical visual stimulation did not Cognitive structures emerge from recurrent sensory-motor activity Continuity between life and mind, brain as a dynamical system - Organisms are autonomous agents that actively generate and maintain their identities, and thereby define their own cognitive domains Neurophenomenology - Enactivists are starting to combine advanced neuroscience & advanced lived experiences/3-rd person imaging methods with 1st-person lived experiences -- E.g. Using fMRIs on experienced mindfulness practitioners to observe onset of their spontaneously arising thoughts/mind-wandering

Critical Period

- Period where language ability is at its peak (0-12 years) - If linguistic experience is missing during this period, language ability is impaired - Case studies: - Victor the "wild child" - Genie

Information Theoretic View

- Person 1 thinks of concept x. - They encode x in language. => encode(x) - Person 2 acquires the encoded concept. - Person 2 then decodes the encoded concept. decode(encode(x)).

Current approach to AI: Neural Networks

- Phase of doing Machine Learning based AI that requires ton of data & computing - Wasn't until 2015 when more data was able to be obtained that Neural Networks performed better than any other network - Neural Networks: artificial neurons connected to each other to learn the relationships between inputs and outputs

What A.I. can and can't do

- Play a decent game of table tennis? YES - Drive safely along a curving mountain road? YES - Drive safely along University Avenue? MAYBE - difficulty navigating traffic - Buy a week's worth of groceries on the web? YES - Buy a week's worth of groceries at Trader Joe's? NO - difficulty recognizing items and navigating through crowds - Discover and prove a new mathematical theorem? MAYBE - Converse successfully with another person for an hour? NO - difficulty adhering to human language conventions - Perform a complex surgical operation? MAYBE - depends on the type of operation - Unload a dishwasher and put everything away? YES - Translate spoken Chinese into spoken English in real time? YES - Write an intentionally funny story? NO - difficulty adhering to human language conventions

Lakoff's Brain Concept article

- Proposes that the sensory-motor system has the right kind of structure to characterize both sensory-motor and more abstract concepts - When mapping concrete source and abstract target in metaphors, target mapping also utilizes the sensory motor domain

Neurological Disorders Related to Consciousness

- Prosopagnosia and other agnosias (prev notes) - Blindsight -- Something wrong with attention system - not eyes -- Detect but not aware of objects -- Damage to main visual pathway (which could be related to consciousness) -- There is a parallel path that wasn't damaged that is responsible for attention grabbing. -- "Like reptile vision" - Visual Neglect: nelnet of one visual side (usually left side) -- Prismatic magnetic therapy: glasses to bring left field to right field. -- Constraint movement therapy (CMT) - sling on weak side.

Outfielder Problem

- Question posed: how does an outfielder catch a ball? - Brain-bound Prediction Strategy -- Brain-computation based --- They watch the batter hit the ball --- Perceives the initial speed & angle --- Computes where the ball will land using the math of projectile motion --- Runs (in a straight line) from where they are to where the ball will land - Using a move-able body -- Linear Optical Trajectory --- They run in a curved path that offsets the curved path of the ball, so that ball appears as if it were tracing out a straight line (rather than a parabola) --- By moving this way, the outfielder ends up at the right place at the right time to catch the ball

Case Study: fighting/racing climate change

- Race: stakes not as high, just need to wait until end, absolutism/everything is either fine or not, fans (plants & animals) not affected by race - War: Complex results, results happening constantly, emotional and physical effects on bystanders - Affects how we reason about complex issues Important to know how to - Identify metaphors, analyze

Implications of AI: Societal

- Reality vs. Conjecture - Ethical Conundrum -- Companies are profit-driven -- Autonomy - if machines are told to eliminate based on threats from data - Human in the loop for high stakes decision making -- Set systems up to verify the claims being made -- Third party auditing company -- Humans should be in the loop when it comes to life or death -- What's scarier: a machine that kills or a machine that tells people to kill?

Embedded Cognition

- Reducing cognitive load -- The cognitive load that a task requires can be reduced when the agent embeds themselves within an appropriately designed physical or social environment -- The environment can contribute/impact cognitive tasks - Expert tetris players -- Expert tetris players tend to rotate Zoids physically, rather than mentally --- This strategy offloads cognitive difficulty onto expert environment

Activation Function

- Restricts the value of the output to a certain range - E.x. Value greater than 0 => return 1, value less than 0 => return 0 - Adds non-linearity - Gives model ability to learn non-linear patterns (ex. concentric clusters)

Methods: Rule Based Systems, Probabilistic Models, Neural Networks

- Rule Based Systems: Interprets given information on previously defined rules. - Can't generalize something new. - Probabilistic models: classify new data based on word values. - Neural Networks: Modeled off biological neural systems.

SOAR

- Rule based cognitive system - Created by Allen Newell - Production memory= rule based learning (if-then) - Knowledge based reasoning - Used to create a system that has the same computational and cognitive abilities as humans

ACT-R

- Rule based cognitive system - Created by John Anderson - Cognitive model that involves the transfer of working memory to production memory and declarative memory

Brain Mapping Techniques

- STRUCTURAL Brain mapping techniques: where components are located in space -- computed tomography (CT): 2D rendition -- computer axial tomography (CAT): X-ray beam rotated around a slice to construct images ---- invasive (according to lectures) because of introduction of radiation ---- subjects the body to radioactivity of X-rays ---- used to image bone, brain, organs -- magnetic resonance imaging (MRI): protons in the body emit radio signals which creates detailed images of organs ---- noninvasive - FUNCTIONAL Brain mapping techniques: functions performed over time -- electroencephalography -- event-related potentials (ERPs) -- positron emission tomography (PET) -- functional magnetic resonance imaging (FMRI)

Neural Substrates of LTM

- Semantic memory: linked to the limbic cortex - Hippocampus mediates consolidation of episodic memory - Basal ganglia & motor cortex associated with procedural memory function

Atkinson & Shiffrin's Multistore Model of Memory

- Sensory Memory -- Iconic memory: a brief persistence of a visual impression (think icon/visual) -- Echoic memory: sensory memory that processes auditory information (when auditory stimulus is heard it is stored in memory to be processed and understood) - Working Memory/Short-Term Memory -- Current information monitored and manipulated (working it: can be received by senses or retrieved from LTM) -- Capacity can be increased by chunking (grouping into meaningful wholes) -- Rehearsal also helps with the duration you remember Ex: phone number, social security number - Long Term Memory -- Declarative: explicit recall memory (aware of it) --- Semantic: facts about the world --- Episodic memory: specific personal events --- Procedural: implicit memory (not entirely aware of it) skill based ex: driving

Neural Correlates of Consciousness

- Specific parts of the brain responsible for consciousness. - Can be one thing or multitude. - Unanswered, but some hypotheses.

The Feature approach to categorization

- Specify those characteristics of a category that are both necessary and sufficient for membership in that category - Bachelor: adult, human, male, unmarried - Needs to be a precise match - All features must be included for category to be adequate

Roger Sperry's Split Brain patients

- Sperry's experiment found that each hemisphere of the brain (left and right) have their own set of sensations - Performed experiment on monkeys to see the relationship between the left and right hemisphere of the brain - Each hemisphere can operate independently from the other and they both have their own functions - There is one stream of consciousness between the two - Found that what one sees in the opposite line of sight determines how they express it -- Example: Patient who has undergone a lobotomy is shown a ring in his right visual field and a key in his left visual field. He is only able to articulate that he saw a ring because speech is controlled by the left hemisphere and the connections are made contralaterally -- Information is not trapped in the right hemisphere, the patient knows what they saw, they just cannot articulate it. He is able to pick up what he saw using his contralateral hand

Corpus callosotomy

- Split brain experiment - word on right field of view, patient would answer the word on screen because left hemisphere is dominant for verbal processing - word on left field of view, patient would answer nothing because right hemisphere cannot share information with the left, patient is unable to say but can draw it

Theory of cogs

- Start off by calling the premotor cortex a "secondary" area → "an area not directly connected to sensors or effectors, but which provides structure to information going to effectors or coming from sensors - Narayanan's hypothesis, abstract aspectual concepts have the following properties: - They are neurally simulated in a secondary area with no active connections to a primary area - Their inferences are computed via that simulation - They characterize concepts in the grammar of a natural language - These concepts are general, and can apply to any special-case concepts when there are active connections to primary areas - When there are such active connections, the general concepts are an inseparable part of the structure of the special case concepts - Lakoff has proposed a generalization of the list above → any concept that has the properties in the list above is called a cog - In the theory of cogs, all concepts of grammars of natural languages should have the properties given in the list above - In other words, they should be computed in secondary areas - This may include primitive image-schemas (containment, source-path-goal, force dynamics, orientation schemas, etc) - Overall, the theory of cogs says: - Because neural structures in secondary areas are inseparable in behavior from the primary structures that they are connected to, they characterize generalizations that are inherent in and inseparable from special cases - The learning of general cases is not the acquisition of new structures - Rather, it is the inhibition of the connections between the secondary + primary areas - The generalizations are inherent in the special cases that are learned first → we learn the control of inhibitory connections

Perception and Steven's Law

- Steven's Law- S is perceived sensation, I is physical identity. - Ex. When doubling the physical area, humans perceive the area as much less than double.

Implications of AI: Strategic

- Strategic stability, perception and the security dilemma -- Exists when adviseraries lack a significant incentive to engage in provocative behavior -- Deterrence theory -> survivability = hardening, concealment, redundancy (idea that you can survive the first strike) -- A capability that is almost effective might be even more dangerous than one that already works (may be destabilizing) -- Movement too far, fast, or without appropriate consideration could generate instabilities that adversely affect international security -- AI directly impacts the security dilemma ---- Security dilemma - actions thought necessary that actually decrease security

Stress and its Effects on the Hippocampus

- Stress decrease/shorten dendrite length and inhibit neuronal communications -- When chronic stress it activates HPA axis, this can cause a type of neurotoxins to the neurons

Strong vs. Weak A.I.

- Strong AI: artificial systems that really "think" and have real understanding, making them indistinguishable from human - Weak AI: artificial systems that act intelligently and mimic certain aspects of human

CAT

- Structural/static - Scan is according to the amount of x-ray absorption - Good temporal resolution - Bad spatial resolution - Non-invasive - Computer axial tomography

Multimodality of action concepts

- Structure of argument takes this form: - Information Structure: information structure needed for conceptual structure of "grasp" is available at neural level of sensory-motor system - Includes semantics, aspectual structure, and hierarchical category structures

Cleveland & McGill

- Studied human perception of data visualizations by running experiments. - Work was foundational to the field of designing data visualizations and graphs. Thye detailed the common cognitive tasks that happen when somebody reads a chart and they evaluated how well subjects performed on these tasks depending on the feature of the graphs. - Lots of ways to represent data: point on a number line, bar chart, pie chart. - Bar Charts vs. Pie Charts - 54 participants were shown 20 different graphs (10 of each) and asked which is larger and the percentage. - Results showed bar charts performed better, pie charts were more inaccurate. - Finding was that a line graph is better for clarity.

Habituation Studies

- Studies that exploit babies getting bored: use babies' attention and focus - Put trackers to see attention span (looking at two of the same faces until they're bored, then adding one new face and seeing how long they fixate on the new thing, they look away when they're bored with the stimuli) - How the child reacts to stimuli - Habituation = bored - Language influencing category construction

Categories and Prototypes - Levels

- Superordinate: Overarching category. Fits lots of things. -- Ex: furniture. - Basic level: overall perceived shape, single mental image, general motor program, shortest word, first used by children. -- Ex: chair, desk. - Subordinate: specifics of basic level categories. -- Ex: leather chair, wood chair, L-shaped desk, etc.

Types of Machine Learning

- Supervised Learning: Task Driven - Classifying and Detecting (Is this a picture of a cat or dog? Yes/No) - Labeled - Unsupervised Learning: Data Driven - Finding patterns in data (Do people that buy books also buy the movie on the same day? Clusters on Graphs) - Unlabeled - Reinforcement Learning: Learn from rewards and mistakes - Interacts with environment and makes a decision/judgment - Unlike supervised, which would penalize any bad thing, reinforcement penalizes actions based on how bad they are - AI never being shown how to walk yet it came up with ways to move and walk

The Anatomy of the Brain and Primary Visual Pathways

- The Primary Visual Cortex is in the Occipital Lobe, the point of arrival for information from the retina. - Visual information from each eye is transmitted by the optic nerve to Lateral Geniculate Nucleus (a sub-cortical area of the forebrain) and then to the primary visual cortex. - Right visual field information is processed by the Left Hemisphere. - Three parts of the mammalian brain are forebrain, midbrain and hindbrain -- forebrain: --- forebrain is the largest out of the three parts --- cerebrum takes up most of the forebrain ---- cerebrum is divided into left and right hemispheres ---- important for cognitive processing and motor processing ---- cerebral cortex, or gray matter, is the outer layer of each hemisphere ---- four lobes in the cerebrum

Sapir-Whorf hypothesis

- The Sapir-Whorf hypothesis is also known as the relativity hypothesis: holds that the distinctions within a particular domain expressed in a given language will not be the same as those in any other language. -- the realm of color is used to support this hypothesis, as color varied arbitrarily in each language -- Whorf's point was that the Hopi language reflects a different worldview that our language lacks the means to express - linguistic determinism: the strong version of the Safir-Whorf hypothesis: holds that our first language determines the way we think - linguistic relativity: distinctions found in a given language aren't the same as those in any other language

The Typological Prevalence Hypothesis

- The developmental ordering of words/partition labels in language learning should reflect their cross-linguistic prevalence -- All else being equal, within a given domain, the more frequently a given way of categorizing is found in the languages of the world, the more natural it is for human cognizers, hence the easier it will be for children to learn

The role of presence in retention

- The sense of experiencing a virtual world as a place that one is actually inhibiting, rather than something one is watching on a screen - Forgetting you are in a virtual setting - The dual context results was only significant in those who experienced a high degree of presence in the virtual environments

Learning rate

- The size of the steps we take in the direction of the local minimum - How drastically we adjust the weights to minimize the loss function

Supervised Learning

- Think about studying for an exam with previous exams as practice. After the exam you check your answers with the solutions - Given data and expected output, see if model is doing well - Doing well = optimizing a loss function (distance between observed and expected values) - Ex. (expected - observed)^2 (mean-squared error)

Unsupervised Learning

- Think of watching a youtube video and the algorithm outputs more of that type of content - Discover and explore structure in data without an expected output - Goal: to discover and explore structure in the data - Why? - Data isn't always labeled - Expensive labels - Exploration - Reduce noise and complexity

K-Means

- Type of clustering (identify groups in data) - Pick k centroids, assign each point to the group whose centroid is closest, adjust centroid based on new point - Ideal k: where there is a sharp/steep fall on the graph of Sum of Squared Errors vs. k (elbow method)

Two Cortical Visual Systems

- Ungerleider and Mishkin first proposed the two visual systems hypothesis -- from the primary visual cortex (PVC), visual information takes either a ventral or dorsal route depending on on the 'type of information' it is - dorsal route: the "where" information -- information relevant to locating objects follows a dorsal route from the PVC to the posterior parietal lobe - ventral route: the "what" information -- information relevant to recognizing and identifying objects fallows a ventral route from the PVC to the temporal lobe - Ungerleider and Mishkin studied cognitive impairments due to brain damage from experiments on monkeys - bottom-up explanation

Google Duplex

- Uses neural networks to interpret speech. Given inputs include context and speech. Google Duplex then makes decisions based on these inputs. - ASR: Automatic Speech Recognition. That's how the agent was able to understand the human. Then pass the context of conversation to the neural network (neurons with layer of 'hidden layers'). After passing through the neural network it is converted to text to speech (TTS).

Specialized cells in Visual Cortex

- V1 (the primary visual area of the cerebral cortex; it is the first stage of the processing of visual inforamtion) -- Processes visual infro from the LGN (lateral geniculate nucleus), which is located in Thalamus in the Visual Cortex; then the processed information is sent to higher visual areas -- P ganglion cells (shape) - layers 3-6 -- M cells (motion) - layers 1, 2 -- Non M - Non P cells, aka kinocellular (color) - all layers -- Simple cells (orientation detectors) -- Complex cells (directionality) - V3 - V5 (extrastriate cortex)

Language deprivation cases

- Victor "wild boy Aveyron" was found at 11 to 12 years old in the wild without any human contact prior. Couldn't pick up any language when found. - Genie, who lived in total isolation until the age of 13, was also unable to learn language. - Isabelle lived in isolation until 6, but she was able to learn language and even catch up to other children the same age. - Hearing imparied children also have a critical period through sign language. - Agrammatic aphasia - aphasia of grammar. - Broca's Aphasia - speak haltingly, omit grammar, have trouble getting ideas out. (nonfluent aphasia) - Wernicke's Aphasia = Speech flows with ease but contains many wrong words. (fluent aphasia) - Melodic Intonation Therapy (MIT) - nonfluent aphasia patients are taught to sing sentences as speech is only affected by the condition.

Relativist / Universalist Theory of How Humans Gain Knowledge (Nature vs. Nurture)

- We acquire knowledge/concepts/learn language through our interactions with the world (primary scenes). - But we also learn from what is taught to us (teaching english babies kkita in habituation studies). - It's a mixture of both, and should be emphasized as that.

Language Acquisition

- We learn in a sequence - Study how babies map meaning onto form - Is language first or is meaning first? - Children learn cognitive primitives first, but also we know it's a cognitive primitive when it emerges very early

Extended Cognition

- We think not only with our brains, but with our bodies, the tools and technologies we use and the spaces in which we learn and work - Otto and Inga (the parity principle) -- Inga hears about MoMA. She forms desire to go see it. She thinks for a moment and recalls MoMA is on 53rd Street. She goes and visits -- Otto has antegorate amnesia. He hears about MoMA. He forms desire to go see it. He consults his notebook to find MoMA is on 53rd Street. He goes and visits -- Both have the belief that MoMA is on 53rd Street, just that INGA had that belief in her mind while Otto has that belief in his notebook -- Otto's beliefs are stored in his notebook, and beliefs are a part of the mind. Therefore, it can be argued that a part of Otto's mind is in his notebook

Primary Scenes, Primary Metaphors

- We're all born into roughly the same environment (gravitic, land-based, surrounded by moveable and immovable objects), many of us with pretty similar bodies (fleshy, symmetrical) and w pretty similar perceptual systems - Because of our similarities, we also notice similarities in our lives and we notice correspondences between things very early - Primary Scenes: the sources in a primary metaphor, and it helps us to structure our language and reasoning - Primary Metaphor: basic connection that exist between subjective or abstract experiences - Primary metaphor in primary scenes

Behaviorism

- What goes on in the mind that is not directly observable or measurable is not an appropriate field of research. Only thing that can be studied in psychology is behavior. - Brain is black box.

Associative Learning in Neural Networks

- When an axon of cell A is near enough to excite a cell B and repeatedly or persistently takes part in finding it, some growth process or metabolic change takes place in one or both cells such that A's efficiency, as one of the cells firing B, is increased - When 2 neurons fire together, they wire together

Hebbian Learning

- When two joining cells fire simultaneously, the connection between them strengthens - Cells that fire together wire together - We learn based on coincidence of input - This is different from how AI and ML learn

Applications

- Why do humans need language? - Communicate ideas. - Represent complex, imperfectly-designed concepts. - Perform logical reasoning.

Representation of Language: words, syntax, semantics, discourse

- Words: Basic element of speech or writing. - Syntax: Arrangement of words (structural) - Semantics: Meaning and relations of words (functional) - Discourse: Linguistic expression beyond the boundary of the sentence.

Connectionism

- a type of information-processing approach that emphasizes the simultaneous activity of numerous interconnected processing units

Quantity and Verticality

- as there is more of something physical, it takes up more (vertical) space - this correspondence between QUANTITY and VERTICALITY results in us reasoning about all changes in quantity as changes in verticality, even when there is no actually object taking up space - When setting up a metaphor, remember to capitalize the source and target

Brain recording and stimulation techniques

- brain's electrical activity measured -- single cell recording -- multiple unit recording ---- Invasive techniques of measuring and recording electrical activity that occurs in a specific part of the brain, or across the brain - brain recording technique -- electroencephalogram (EEG) measures the gross electrical activity of the entire brain ---- Non-invasive, nice medium ---- BUT, there's a lot of noise (interference from the skull, movement, motion artifacts that interfere with signal) -- Event related potential (ERP) is an EEG recording in response to a stimulus

Tulving's Model

- contrasts two types of declarative memory: semantic, episodic -- semantic memory: memory of how to do something. a mental thesaurus or organized knowledge a person possesses about things like rules, formulas, algorithms; about words, verbal symbols; manipulation of symbols, concepts, relations -- episodic memory: receiving and storing information about temporally dated episodes/events and the temporal-spatial relations - example: memory of your late arrival to class due to a flat tire constitutes your episodic memories while the knowledge that a flat tire is an justifiable excuse derives from semantic memory

Constructive Memory

- memory is not always accurate as seen in contradictory evidence brought in by witnesses, amnesia victims -- failure to remember how/when they required retrieved pieces of an episode can cause people to arrive at some illusory memory - "the constructive nature of human memory" is a discussion that takes note of the fact that memory consists of constituent elements distributed widely across the brain. -- retrieval of a past experience involves a process of pattern completion; a subset of the features comprising a particular past experience are reactivated; the features of a given episode must be coherently linked

Synaptic transmission

- neurons communicate with each other at synapses - exocytosis: action potential results in the release of neurotransmitters across the synaptic cleft - Neurotransmitters bind to receptors in the dendrite of the receiving neuron creating IPSPs or EPSP which are summed at the axon hillock - Hebbian Learning: neurons that fire together wire together -- eric kandel showed in sea slugs' reflexes using classical conditioning

Retinotopic organization

- two types of visual fields- -- left half of each eyeball is the left visual field; -- right half of each eyeball is the right visual field

Google LaMDA

-An open-ended conversational AI application - In demos, it took the role of a person or an object during conversations with users - LaMDA synthesizes concepts from the training data to ease access to information about any topic via live conversations

Marr's three levels

1.) Computational (infromation-processing systems can be analyzed in terms of the problems they can solve) 2.) Algorithmic (representations and processes by which these systems solve the probelms; the "software") 3.) Implementational (the physical instantiatin of these representations and processes; the "hardware") - Marr's model of vision -- image->primal sketch (based on feature extraction)->2.5D sketch (intermediate stage)->3D sketch (stage we can view images in our mind's eye)

How language is special?

1.) Relies on mental representations (trouble for behaviorisms) 2.) Particular strengths and weaknesses: great for things that aren't there (future, hypotheticals, etc.) 3.) Uniquely human: suggests an innate basis (trouble for empiricism) 4.) Potentially infinite: can always produce novel sentences

Paperclip Maximizer Thought Experiment

AI to maximize number of paperclips either kills all humans or turns every human to make more room for paperclips

CRUM vs 4E Cognition

CRUM: - Cognition is a kind of information processing that consists in the syntactically driven manipulation of representational mental structures. Cognitive processes are realized by brain processes only. 4E Cognition - Cognitive phenomena are in deeply dependent on the physiological details of an agent's body, an appropriately structured natural, technological, or social environment, and the agent's active and embodied interaction with this environment.

Embodied Cognition

Cognition depends directly on the body as a function whole, and not merely on the brain - Conceptual Metaphors -- Grounding Metaphors (of Arithmetic) -- Yield basic, directly grounded ideas and usually require little instruction -- E.g.: ARITHMETIC IS OBJECT COLLECTION -- E.g.: ARITHMETIC IS MOTION ALONG A PATH - Linking Metaphors -- Yield sophisticated ideas, sometimes called abstract ideas. These require a significant amount of explicit instruction ---- E.g. Number as points on a line (number line) ---- E.g. Geometrical figures as algebraic expressions ---- E.g. Operations on classes/sets as algebraic expressions

The 4E Approach to Cognition

Embodied, Embedded, Extended, Enactive Cognition

Narrow vs. General vs. "Superintelligence"

Superintelligence - exceed human intelligence / cognitive capacity (can think on its own)