Please enable JavaScript.
Coggle requires JavaScript to display documents.
Lecture 1 - Introduction - Coggle Diagram
Lecture 1 - Introduction
-
Why Neurones Matter
Tri-Level Hypothesis (Marr)
- Neurones function at three levels
Levels
- Computational level
- Algorithmic/representational level
- Physical level
Computational Level
- What does system do & why
- Take information & use that to build percept
Algorithmic/Representational Level
- How does system do what it is does
- What representations does it use
- What processes does it employ to build & manipulate representations
- What info needed to be represented & how, & how to manipulate representations to analyse stuff
Physical Level
- How is system physically realised & implemented
- Literal & physical implementation
- Other levels can be described
- This level practical but not interesting in understanding how system works
Functionalism & Physical Level
- Functionalists argue understanding cognition does not require knowledge about physical level
- Only neural implementation
Do Neurones Matter (Colheart, 2004)
- Neurones don't matter in wider context of cognition
- No facts about brain activity can be used to confirm/refute some information-processing model of cognition
Do Neurones Matter (Fodor)
- The neurones are active anywhere
- Neurones matter but knowing where matters less???
- Understanding physical level crucial if goal is intervention
- But deeper problem with functionalist logic
Why Neurones Matter
- Computational level generally much too vague to be useful
- Algorithmic & physical levels rarely separable in algorithmic sense of what they represent
- Algorithms don't necessarily generalise
- So just because know principles in one system does not follow another system relies on same system
- Cognitive psych relies on black-box models
- But any behaviour can potentially be realised by infinite number of algorithms
- To find out which one brain uses, must operate black box
- At end of day cognitive neuroscience about how human & animal brain work
- Otherwise psychology indistinguishable from robotics
Black Box Model
Black Box Model
- Abstract description that takes info as input & produces some kind of output
Black Box Models in Psych
- Don't know what goes on inside model
- Try to create most complex, accurate answer
- Assume that signal is brain function, but signal is the measure of brain function
- But can create infinite no. theories
- Just because model works cannot conclude that's how brain works
- Model that comes from behaviour not necessarily translates to what goes on in brain
Significance
- Do care about how brain does what it does
Questions
- What is right level to study the brain
- What do we need to measure to understand how the brain works
Example: Levels
- Clocks & time
- At computational level do same thing
- At algorithmic level ask how they are represented
- But this representation cannot be differentiated between physical forms (all forms show time physically)
Example: Generalisation
- Flight
- Plane & bird aerodynamics are similar & the same type of physics that govern
- Plane & bee aerodynamics are different & diff type physics that govern
Sentience
- Watson vs Brad vs Ken
- Google Chatbot
- ChatGPT & Turing Test
Google Chatbot
- Language Model for Dialogue Applications (LaMDA) claimed it is a person
- Is aware of own existence
- Is capable of emotions
ChatGPT & Turing Test (Turing, 1950)
- Routinely passes
- Implications of what that tells us about brain
- If generative AI models are good model for brain function
- Sounds for all intents & purposes like human being
- Set tone & AI responds in that tone
Turing Test (Turing, 1950)
- Tests ability of machine to exhibit intelligent behaviour
- Machine passes if is indistinguishable from human in behaviour
Can We Learn from Them
- AI works through predicting what comes next
- Evidence that awareness an after construction
- Own algorithm that predict certain behaviour like LLMs
Example: Jeopardy Watson vs Brad vs Ken
- Computer created to play game
- Conclude computer have no consciousness
- Designed for specific purpose
Why Neurones Matter
- Cog neuroscience about how human & animal brains work
- So do care about how brain works so physical level matters
-
-
-
Problems & Limitations
Problem w. Methods
- Most don't work in humans
- Those that work in humans provide very limited information
Limited Information
- Indirect measures
- Population-based
- Can only measure correlates of neural activity
- Averaged over many neurones & over time
- But single-unit measurements suggest activity of individual neurones is significant for cognition & perception
- Level of observational detail is orders of magnitude
- Slower/coarser than those of neuronal processing
Evolution as a Friend
- Animal models
- Comparative studies
Comparative Studies in Human & Non-Human Primates
- Most informational experiments generally based on single-unity recordings in non-human primates
- Experiments in humans often mainly confirmatory
- Details of mechanisms only revealed by direct neuronal recordings
Studies: Comparative Studies
- Mapping visual cortext
- Role of MT
Study: Role of MT for Motion Perception (McKeefry et al.,)
- Q:
P: Human & non-human primates
- M: TMS of human cortical areas V5 & V3A
- R: Induced deficits in speed perception
- Changed percept of motion in response to stimulation in non-human primates similarly in humans
Study: Mapping Visual Cortex (Larsson & Heeger, 2006)
- M: fMRI
Ps: Human & non-human primates
- R: Found similar visual areas & sig differences in human & macaque
Animal Models
- Brains similar to non-human primates
- So can learn about brain function
- But limit to how far can get using animal models
- Evolution eventually diverge
- May have to start making educated guesses before cross-referencing w. human studies
Problem w. Complexity
- Connections not random
- Form highly organised neuronal circuit
- Approx. 100,000,000,000 neurones in brain
- Each connected w. average 1000 other neurones
- Approx. 700,000,000,000,000 synapses in brain
- Each neurone makes average 7000 synaptic connected w. other neurone
Principles of the Brain
- Brain may rely on small no. computational units (neural circuits) adapted for diff purposes
- So may be enough to understand principles governing each unit
- But brain may also solve each problem differently
-
-