[Lecture Notes] Last Third Flashcards
Brain Size: Proposals About General Intelligence
“more intelligent people have more neurons”
- problem with this is that it’s easily disconfirmed; e.g. whales have bigger brains and a certain species of whale has a better body-to-brain ratio, but they’re still doing whale stuff and we’re doing lots more stuff
- neural pruning during development: number of neurons goes down as intelligence goes up significantly; i.e. inefficient pathways are being lost as we grow
- there are lots of variations in brain size in the human species that doesn’t seem to correlate with how intelligent those people are
- note the role of development and plasticity in intelligence (i.e. neural pruning)
- development is changing how much your brain is capable of knowing, not how much it knows
- neuroplasticity is how your brain can reshape itself to change the types of things it can know
- in psychology, we’ve moved away from both brain size and speed for the most important factors in intelligence; they’re relevant, but not that relevant
Speed: Proposals About General Intelligence
“intelligent people are faster in their processing”
- some evidence to support this as people with higher IQs have faster reaction times
- part of the common sense idea of intelligence being quickness
- Luciano et al. (2005): found there’s no causal link between speed of information processing and intelligence
- why? being fast means you make more mistakes; it means you’re just doing more of something faster, not that you’re doing it better
- speed is only a quantitative measure but doesn’t pick up on the qualitative measures that would be needed for intelligent processing
- e.g. increasing computer speed was nowhere near sufficient for producing intelligence
- i.e. not just the speed of your hardware but how the software uses it
- in psychology, we’ve moved away from both brain size and speed for the most important factors in intelligence; they’re relevant, but not that relevant
Intelligence: The Neural Efficiency of Processing
- neural efficiency hypothesis (discussed in the textbook)
- Neubauer et al.: found that people with higher IQs activate less of their cortex
- the idea that we only use 10% of our brain all the time is also ridiculous
- our brains aren’t motors; an intelligent brain uses less to do more work, not putting in more energy to do the same work
- Friston, Hawkins and others have argued that what intelligent brains are efficient at is at making predictions
but there are still problems with this view:
- how do you make a brain more efficient at prediction?
- what do you predict? do you try to predict everything (like the number of socks in your drawer)?
- what scope of prediction do you use (short term vs. long term efficiency)?
- Garlick (2002): answers the first question; picks up on the connections between intelligence and plasticity that we saw in neural pruning
Qualitative Development: Neuroplasticity and Increased Functions
- remember that Perceptrons were neural networks that couldn’t do basic functions like exclusive “or”
- but as you add in neurons between the stimulus and response neurons, and remember that greatest number of neurons your brain has are interneurons, then a network gains the ability to do new things
- adding the interneurons massively increases the functions a neural network can perform
- quantitative development: a measure of how much you know
- qualitative development: a measure of what kinds of things you’re capable of learning
- i.e. compare all the information you could gather with a function vs. how many different functions a machine/system has
- the more functions you have, the more kinds of problems you can solve; plasticity drives qualitative development
- one way of understanding intelligence is if you can solve many kinds of problems in many types of domains; i.e. the more plasticity your brain is capable of/functions you have/qualitatively developed
- adding/subtracting (rf. pruning) interneurons can alter the intelligence (the functions it can perform) of a network
- Garlick (2002) argues that the way you make a network more efficient is by creating the right architecture (structure of interneurons also known as hidden units) of the network
- so what we’re measuring when we measure intelligence is plasticity of a brain; i.e. how well it can redesign its architecture through synaptogenesis and neurogenesis
but there are still problems with this view:
- simple plasticity can’t equal intelligence
- simply adding or subtracting neurons randomly will produce chaotic changes in a brain
- simply adding connections will increase how much the brain can process but not what it’s processing
- intelligence is not just quantitative but also a qualitative improvement (i.e. not just more but better)
- Garlick himself notes we need an account of appropriate plasticity
- so we need an internal standard of goodness of plasticity
- how does the brain know or measure when it’s producing appropriate plasticity?
The Generalization-Discrimination Problem: Fault with the Prediction Model
- the generalization-discrimination problem;
- as you make your predictions more generalizable, you make them more efficient in one sense because you can use the same function over and over again
- if I can make predictions about cats rather than this cat Bob, this cat Andrew, this cat Jake, etc., my brain uses less resources to come to the same conclusion
- but what happens if instead of cats, I make it all mammals, or all animals, etc.?
- as I make my predictions more efficient, I lose the ability to discriminate between cases; I lose information about important differences
- the problem in a nutshell: sometimes, the similarities (share properties) are important and sometimes the differences are important
- so just like there are people who argue that intelligence is prediction, some theorists argue that the ability to discriminate differences is the key to intelligence
Psychologists Dealing with the Generalization-Discrimination Tradeoff
- Mercado (2008): consider visual acuity;
- acuity is your ability to see things clearly (like wearing glasses)
- you don’t want confounding and conflation of information that causes confusion in your predictions; if you can’t distinguish between two things, you can’t individually test them
- sometimes, being able to discriminate the information is more important than trying to integrate it
- the more representation power or cognitive acuity you have the more intelligent you will be
- this lines up with recent work by Wissner-Gross and Freer in that you can make a system more intelligent if it works to keep more of its options open
- this goes back to the scope of efficiency problem; efficient right now vs. efficient long term
- yet there are too many options, which lead to combinatorial explosion
- a system that can see more options, i.e. discriminate more will be more intelligent, but not too many
- there is a constant trade-off relationship and there’s no final answer
- we have two proposals for intelligence: integration for better generalization in prediction and differentiation for better acuity in prediction and long term keeping of options open
- Vervaeke and Ferraro (2013): argue that brains do both in a dynamical fashion and that is the process of realizing relevance
- see also: Vervaeke, Lillicrap, and Richards (2012)
- relevance realization as opponent processing between efficiency and resiliency (long term differentiated efficiency)
- i.e. your brain needs a lot of redundancy because the future is always changing; the more efficient, the more brittle and rigid and less likely to adapt to change it is
- your brain is constantly doing opponent processing; constantly (cognitively) evolving its fitness to its environment, i.e. redesigning itself to solve more problems
From Relevance Realization to Generalized Intelligence
- general intelligence explains:
- problem formulation
- similarity judgements
- “education” of latent information from experience
- adaption to unpredictable environments
- solving ill-defined problems
- a system becomes more complex as it simultaneously integrates and differentiates
- complexification gives you qualitative development/emergent function
- if we can see that brains moves between compression and particularization, we can see it doing relevance realization
Self-Organizing Criticality (SOC)
- Bak, Tang, and Wiesenfield (1988): sand piles; the sand piles break down as sand gets piled on it, but it builds a stronger base for a bigger pile; c.f. your brain is constantly oscillating between differentiating and integrating
- Stephen, Dixon, and associates (2009): gear diagrams for insight problem solving
- you can measure how flexibly a brain does self-organizing criticality
- Thatcher et al. (2008): phase shift (asynchrony; positively correlated with g), phase lock (synchrony; negatively correlated with g), and phase reset; the more smoothly your brain can go between these two things, the more intelligent you are
- but the plasticity issue has not been fully developed, i.e. what kind of architecture is produced and how does it relate to SOC?
- in order to address that we can turn to network theory (also known as graph theory)
Network (or Graph) Theory
- the chaotic/random network, has the least path distance, and is most efficient, but it’s also the most easily damaged
- the small-word network is way more efficient than the ordered network and way more resilient than the random network; but it’s not as efficient as the random network, or as resilient as the random network
- so if the brain is trying to get both, it should probably go for the small-world network
- and this has actually been found; the more the brain wires like a small-world network, the more intelligent they are
- self-organizing criticality (SOC) and the small-world network (SWN) support each other
- as a system fires in a SOC manner, it wires in a SWN manner and vice versa
- i.e. they reinforce each other’s development in a mutually accelerating, mutually bootstrapping fashion
Is intelligence a kind of development?
What are the three approaches to studying development?
- what we have seen is that intelligence is probably inherently developmental in nature
- it’s not just that intelligence develops; it is that intelligence is a way the brain develops
- intelligence is a kind of development
- so understanding development is crucial to understanding the nature and function of the mind
- there are three well established approaches to studying development: the Piagetian approach, the socio-cultural approach, and the information processing approach
- there’s also a new approach that we have just seen in our discussion of intelligence, viz. the dynamical systems approach
The Piagetian Approach to Development
- prior to Piaget, intelligence testing had been going on for some time but people had only been looking at the correct answers on the tests; they had been ignoring the errors are irrelevant noise
- but Piaget wondered if there were any patterns in the errors; i.e. were they random or systematic?
- most intelligence testing was being done on children for academic reasons
- systematicity and the competence/performance difference; performance is what you’ve done, competence is what you’re capable of doing
- between your competence and performance are processes that implement your competence, e.g. all the processes that you need in order to speak
- performance errors (are random and) do not tell us about constraints on our competence, e.g. you can’t speak properly because you’re drunk or asleep
- systematic errors reflect limitations in your competence, e.g. you can’t speak properly because of brain damage
- performance errors are unsystematic because they’re due to idiosyncratic factors of circumstance in individual lives, e.g. Peter couldn’t sleep last night and is currently napping; this tells us nothing about why Susan currently can’t speak well
- Piaget found systematic errors in children’s performance; systematic errors reflect limitations in competence
- e.g. the egocentricity error, the conservation failure
Piaget: Schemas
- Piaget argued that these patterns of error indicated that the child’s thinking was being limited, but also enabled, by certain schemas
-
schema: a way of organizing information so that it makes sense and problems can be solved
- c.f. they’re ways of reorganizing problems
- for example, in failures of conservation, children are using a schema that works according to centration
- their attention is centered upon one variable (how much space is taken up by the candies) that they find super-salient, and that blocks (rf. learning theory) their relevance realization
- Piaget’s idea was that although these schemas were causing errors, they were actually adaptive and helped the child makes sense of its environment and solve its problems
- given the environment and problems a small child faces, the more space = more stuff heuristic can often make good sense
- what looks like bizarre behaviour to adults is actually intelligent behaviour from the child’s POV; this works both literally and metaphorically, based on how children’s bodies are structured
Piaget: Assimilation and Accomodation
- Piaget argued that these schemas were formed by two opponent processes
-
assimilation: the brain tries to fit information into its existing schemas
- the schemas have worked and are metabolically expensive to form and maintain; so use them and try to make it work
- to use language we have already developed: it is efficient to refuse existing schemas by making information fit in them even if it is somewhat distorting of the information
- however, overuse of assimilation leads to too much distortion
- i.e. there’s too much confounding and conflating of variables and confusion, too little cognitive acuity, and so the cognitive system has to differentiate its processing
- for Piaget, accommodation: when the brain has to differentiate its processing and make new schemas
- this is the second of the two main processes that drive the development of schemas
- there is opponent processing between assimilation and accommodation
- there are stable periods in which a balance between assimilation and accommodation have occurred and this system is well adapted to its environment (the system is behaving intelligently)
- the development of cognition will show a step wise function, also known as a punctuated equilibrium
- for Piaget development builds up from sensory-motor interaction into symbolic representational thought that then develops into more logical thought, that then develops into abstract logical thought
Piaget: The Four Stages of Development
- (Chapter 10.2) he thought that there were four basic stages of development:
- sensory motor stage; 0-2 years, main accomplishment is object permanence
- pre-operational stage; 2-6 years, main accomplishment is symbolic thought
- concrete operational stage; 6-11 years, main accomplishment is logical thought
- formal operational stage; 12 years and on, main accomplishment is abstract, hypothetical, logical thought that is used in science
- there are some major assumptions in Piaget’s theory that have been challenged:
- the whole of the mind moves from one stage to the next
- the timing of stages is invariant
- the direction of development is one way, i.e. from sensory motor to pre-operational
- there are reasons for doubting all of these assumptions
Baillargeon (1987): Object Permanence
- did a series of famous experiments with 4.5 month-olds testing for object permanence, which for Piaget is not in place until around 2 years of age
- how do you test the cognition of such young children? answer: habituation
- if you have object permanence, you expect the object to stop; if it goes over it, you would be shocked
- if babies have object permanence, they won’t show any interest if the plank stops, but they will if it goes all the way down
- this is exactly what Baillargeon found; at 4.5 months, babies are startled when the planks go down
- however, this is not full object permanence; it’s not very accurate
- but it does show some kind of object permanence and that it’s achieved way more sensory-motor competence has been completely
- and just because there is object permanence in this sense, it doesn’t mean that all the other abilities Piaget associated with it have been achieved