BCOG 458 Exam One

What is cognitive science?

• The study of intelligence

• How animals and other machines acquire, represent, and use information to solve problems 

• How minds (human and other, living and non-living) work 

• A hybrid discipline comprising…

  • Experimental psychology, Computer science, Philosophy, Mathematics, Linguistics, Neuroscience, Evolutionary biology, Engineering, Robotics, Educational psychology, Anthropology, etc.

How does The Mind work? How do minds work in general?

• “It’s a bunch of heuristic algorithms that solve problems.” 

• Heuristic definition: “a mental shortcut or practical rule of thumb that helps people make decisions, solve problems, or learn things quickly without doing deep analysis. It offers an approximate solution.” 

• Heuristic algorithms purposefully trades accuracy, completeness, or precision for speed. 

What is (general) Intelligence?

• A partial answer: “The ability to exploit constraints in order to solve problems in the service of your goals.”

  • What are constraints? 

    • In this sense, they are not rules or restrictions. Constraints are properties of the universe that make it possible for you to solve problems. Gravity is an example, because it is a constraint on life on earth. 

Consider a game of chess. What is the goal, and what are the constraints?

• Goal: to capture your opponent’s pieces and win the game

• Constraints: the rules of chess + the configuration of the board

Consider the rabies virus. What is its goal? Is it intelligent?

• Goal: to reproduce

• The virus attacks the nervous system, more specifically the mechanisms that control salivation. The virus is therefore in the saliva. It then attacks the mechanisms in the brain that control anger and rage, prompting the animal to thus bite other organisms and spread the virus. Is this an example of exploiting constraints to meet a goal?

What roles do generalization, inference, and learning play in terms of general intelligence?

• Generalization and inference allow for adaptability. They help answer the question, “What if the constraints/environment change?”

• Learning solves problems flexibly in the service of your goals. It is what separates animals from single-cell organisms

Does sentience play a role in intelligence? Consciousness? Self-awareness?

• These questions are one of the un-computables. They’re not possible to know

Describe the importance of relational generalization.

• Relational generalization means you can reason explicitly about relations and perform analogical reasoning (solving matching-to-sample tasks based on relationships like "same" or "different") 

• Relational generalization is what separates humans from other animals

  • However, certain animals (corvids, toothless cetaceans, and possibly elephants) have also been shown to demonstrate relational generalization as well

What is the mind?

• The software that arises from the brain's hardware

  • The mind derives from the activity of the brain

• A complex mind (the human mind) is a collection of systems, each solving a different problem

• A mind is a device that exploits constraints to solve problems. They can also learn and generalize. If they can reason about relations explicitly, that makes them even better at exploiting constraints to solve problems.

Why are psychology and neuroscience different sciences?

The laws of the mind do not equal the laws of the brain.

True or false: “Learning” and long-term potentiation (LTP) are synonymous

False. LTP is just one way to implement learning

The mind is to the ______ as phenotype is to __________.

• Brain, genotype

• The phenotype derives from the genotype, but…

  • Mutation operates on the genotype

  • Natural selection operates on the phenotype

    • Laws of natural selection (laws of adaptive advantage) do not equal laws of genotype (laws of organic chemistry)

What ideas did René Descartes have that gave rise to Cartesian Dualism?

• He used his 4 Proofs to prove that the mind is its own thing 

  • Separate from bodies and other physical objects, which are subject to natural laws 

  • He posited that there are 3 types of ‘stuff’ in the universe: 

    • #1. is matter and #2. is energy, and they are both governed by natural laws 

    • #3 is the mind, which he claimed was a fundamentally different kind of stuff because minds have free will. They are not subject to–or understandable in terms of—natural laws. 

What is Cartesian Dualism?

The philosophical concept proposed by French philosopher René Descartes stating that the human being consists of two fundamentally distinct kinds of substances: an immaterial, thinking mind and a material, physical body

Why is Cartesian Dualism a bad idea?

• Nowadays, we argue that minds are different from brains. This isn’t to say that they’re not subject to any natural laws at all. We are instead trying to say that the natural laws that govern the mind are simply different from the natural laws that govern the brain. 

• “Does this mean the mind is not physical?!” 

  • Yes: Real and lawful, but not physical 

  • Like algorithms, information, mathematics, laws, music, money…

  • However, for the mind to operate, it must be instantiated physically 

What are the main questions that the “Computational” section of David Marr’s levels of analysis asks?

• What problem is it solving, and why? 

• What are the constraints on its solution? In other words, what makes this problem solvable at all? 

• What, in general terms, is the nature of the problem that is getting solved, or the function that’s being computed? 

What are the main questions that the “Representation & Algorithm” section of David Marr’s levels of analysis asks?

• What information does the system represent about the problem, and how does it represent that information? 

• What does it do with that information? 

• What algorithm is it running on the information in order to get something useful out of it? 

• What is the input to the system, what is its output, and what stages did it go through in between? 

What are the main questions that the “Physical Implementation” section of David Marr’s levels of analysis asks?

• How are these representations and algorithms realized in the hardware of the device itself (e.g. in the neurons of the brain, the silicon of a computer, etc.) 

True or false: David Marr’s levels of analysis are constrained bottom-up.

False. They’re mostly constrained top-down, meaning you must answer the questions in the “Computational” section first

• Marr argued that you must first understand what problem a system is trying to solve before looking at how the physical hardware implements it.

What is the very first part of the cortex that responds to visual stimuli?

V1 (primary visual cortex)

Where in the brain do neurons begin to respond to whole objects?

The inferotemporal cortex

What is a receptive field/a neuron’s tuning curve?

The set of all the things in the universe that cause a particular neuron to become active. There’s usually a specific thing in the universe that the neuron responds the most to, but it may respond lightly to other things as well. 

What is an equivalence class?

A psychological and computational concept that describes how the brain groups completely different receptive fields together based on learning, logic, and functional meaning.

• For example, your brain groups the written word "CAT", the spoken sound /kæt/, and a picture of a feline into one equivalence class.

How does light detection in the retina work?

• Light moves to back of retina (photoreceptors) then back through bipolar/horizontal cells, then ganglion cells, then to brain

• 10 million photoreceptors per eye, 1 million ganglion cells

  • Compresses data 10:1 before it reaches the brain

• Activation of photoreceptors is inversely proportional to the amount of light present in each location

• Photoreceptors respond to light and dark at each location, ganglion cells respond to contrast (larger scale, combination of light and dark)

Describe the computational goals of contrast detection.

• The goal: Surface boundaries are useful indicators of object boundaries. Find them! 

• The problem: The retinal image tells you about luminance (brightness) at each point, not about surface boundaries 

  • You get a luminance map as an input, but you want a boundary map instead

• The Constraints:

  • Smoothness: reflectance tends to change smoothly within surfaces, and abruptly between surface boundaries  

  • Projective geometry: Adjacent points in the world project to adjacent points in the image. Thus adjacent points in the image tend to be adjacent in the world

    • Light travels in straight lines 

    • Information is lost from the 3D projection of the world to the 2D projection to a lens 

  • Therefore, abrupt changes in luminance in the image tend to correspond to boundaries between surfaces in the world. But there are no guarantees.

• New Goal: Find abrupt changes in luminance (contrast) in the image

Which set of graphs (top or bottom) is a better representation of how we detect luminance boundaries?

The top one

Which set of graphs (top or bottom) is a better representation of how we detect depth/distance?

The bottom one

Why did David Hubel and Torsten Wiesel win the Nobel Prize in 1981?

They used a microelectrode in a monkey’s brain to prove that neurons respond to particular locations, orientations, line thicknesses, etc. in the visual field. They disproved the prior belief that neurons responded to simple dots of light.

What mathematical value corresponds to edges in the visual field?

Zero crossing of the second derivative (of luminance)

What is the algorithmic solution for computing a contrast map (in terms of Marr’s 3 levels of analysis in relation to contrast detection)?

Compute the contrast map by convolving the luminance map with a Difference of Gaussians Operator (DOG) at multiple locations and scales