Brains and Computers Flashcards
KEY POINT - IT’S NOT USEFUL TO THINK ABOUT THE BRAIN ONLY IN TERMS OF AREAS
It is useful to know the different areas of the brain, but this doesn’t tell you, on a smaller scale, how these brain areas are getting things ______. We need to know how they are _______ to understand how it all _____ (similar to knowing the parts of a car and not knowing how they go together). Knowing the areas is only useful _______.
KEY POINT - WE CAN UNDERSTAND THINGS IN TERMS OF THE NEURONAL LEVEL
Eg: Justin Harris - monosynaptic reflex (few synapses needed for this bx)
Eg: Justin Harris - neurons can learn associations to trigger a bx (only a few synapses needed for conditioned response bx)
IF YOU REALLY WANT TO KNOW HOW NEURONS WORK YOU CAN GO DOWN TO THE ION CHANNEL LEVEL
BUT SOME THINGS CANNOT BE EXPLAINED SIMPLY BY LOOKING AT A FEW NEURONS
done
connected
runs/works
superficially
A question to reflect on…what has given you the most understanding of how our brains make us smart?
From these lectures….overlaps….connecting things together…storing them via the item itself, not the location.
Also, perception of the world, movement, planning of movements using sensory information.
KEY POINT - OFTEN WE HAVE TO SIMPLIFY SOMETHING TO UNDERSTAND IT - if we can create a model or simulation for the brain we can understand it better - we’ve built it, so hopefully we understand it well.
Science can use modelling and stimulation to ______ something that it is trying to understand. If we can _____ it, it means we understand it. This does not mean we ________ it though. We don’t need _______ details.
We can also use something that is more familiar to us to understand the brain, aka a ________, such as the ________.
What TWO examples does Alex give?
mimic
build
duplicate
irrelevant
Modelling aeroplane aerodynamics - strength and wind simulation. We don’t have to have the correct size plane, what matters is how the wind interacts with it.
In electronic circuit, all we need to know is what the circuit does, and being able to predict what it does. The spatial arrangement of the wires doesn’t matter. What matters in the connectivity. Then we can draw a schematic as well.
We can also use something that is more familiar to us to understand the brain, aka a ________, such as the ________.
Computers are a useful model/simulation/metaphor to understand the brain. By looking at _______ and _______ (even by looking at how computers are NOT like the brain is really useful)
metaphor
computer
similarities and differences
KEY POINT - CARBON IS NOT WHAT MAKES US SMART. IT’S NOT WHAT WE’RE MADE OF, BUT THE CONNECTIONS THAT MATTER
When we model the brain, does it matter that it’s made of silicon and not carbon?
How is this related to the simplification of our brain as computers - the connectionist model (neural network model)?
Not really. Most people think it’s more about the connections than the ion channels. In any case the model is helping us understand the brain better.
Models represent the ability of how one neuron makes the next one fire, without the details of neurotransmitters, etc.
KEY POINT - WE DO NEED TO UNDERSTAND SOME THINGS AT THE LEVEL OF INDIVIDUAL NEURONS TO EXPLAIN BEHAVIOUR. WE HAVE TO UNDERSTAND THIS AS THESE ARE THE BUILDING BLOCKS
What can ONE or TWO neurons do? What are some examples of behaviour we know can come about from only a few neurons?
- decide whether to transmit AP (or activity) to the next neuron (can be +ve or -ve)
- the weight/strength of the vote incoming from other synapses determines this. If threshold is reached - activity is transmitted
- Reflexes
- Pavlovian learning
A simplified simulation of neurons is called a…
connectionist neural network
WHY IS THE CONNECTIONIST NETWORK IMPORTANT?
We can use the connectionist neural network to understand the brain because…
although stuff is happening at the level of the dendrites, the connection of different neurons is the important thing…
FROM THE CONNECTIONIST NEURAL NETWORK PERSPECTIVE…
Neurons have _______. Each synapsing cell has a different number of “_____”, either excitatory or inhibitory. Neurons add up the weighting from different inputs and then make a ______ about whether to fire or not.
So this model represents the ability of one neuron to make the next neuron active, ignoring _______ and action _______. It is enough to worry only about the ________ in the right learning ___________ to accomplish what we have to do.
Eg: Irina has already spoken about this in the ____ _______ Model to shown how knowledge is organised in the brain. You don’t have a list of a magpie or a list of canary features. Instead, you have the features represented as ______ and the ______ of connections between these nodes is what tells you what you know about the bird (does it fly? is it yellow).
thresholds
votes
decision
neurotransmitters
action potentials
connectivity
environment
PDP Network
nodes
pattern
WHAT MAKES INTELLIGENCE?
There are ______ ways to accomplish intelligent behaviour. We can use our knowledge of ______ to help us theorise how our brain does things. We already know we only need a _____ neurons to do something smart (eg: hebbian synapse)
multiple
neurons
few
KEY POINT - IN SIMPLIFIED NEURAL NETWORK THRESHOLDS ARE IMPORTANT TO GET THINGS DONE
The linear activation rule is when you simply pass on overall votes/simulation but this is not enough for…
For this, we need…
decisions
on/off switch OR threshold level of activation
–> so the weights of neural inputs are summed and ONLY if they reach the threshold for that neuron does it fire.
Eg: you either think about something, or you don’t. You don’t think 30% about one thing, 20% about another, etc
THERE ARE MANY WAYS TO ACCOMPLISH INTELLIGENT BEHAVIOUR
Eg: alphabetising a list
It is not obvious how a being (eg: humans, computer) accomplishes a task. It helps to have an understanding of it’s _________. We can use our knowledge of neurons to help ________ about how the brain completes tasks.
An algorithm is simply a series of _____ followed to achieve a _____.
capabilities
theorise
steps
goal
A NAIVE THEORY OF HOW THE BRAIN ACCOMPLISHES INTELLIGENT BEHAVIOUR IS THAT….
Different parts of the brain are dedicated to different functions (eg: run, bark, pant, etc) and that these areas are…
segregated
We know this this not true - often the same neurons are involved in both! Lots of overlap.
NOW LETS START TO SEE HOW SYSTEM CAN BE DESIGNED IN BRAINS VS COMPUTERS - NAIVE THEORIES
BRAIN
Attention can:
- ______ the representation of one voice compared to another.
- ______ rustling in the room
- Engage
- ______ (from the lecturer if there’s a loud noise)
- motivate ______ (to see what the noise is)
- THE INITIAL BOX-AND_ARROW PSYCHOLOGICAL THEORY OF ATTENTION, initial theories by psychologists summised that there are different _____ of attention (eg: interrupt function, moving functioning, alert function, etc) in different parts of the ______, that are organised in a ____ - and - ______ diagram. If you were designing a computer program or software, this might be how you do it (have different parts do different parts of attention), and you would set up _____ to have it happen in the right order.
BUT the brain doesn’t work this way. There are many neurons involved in each of the different aspects.
enhance
inhibit
disengage
movement
parts
brain
box-and-arrow
rules
NOW LETS START TO SEE HOW SYSTEM CAN BE DESIGNED IN BRAINS VS COMPUTERS
- If there is not one bit of the brain that does one function, this means functions are _________. There is not just one bit that does one particular _______.
Attention doesn’t happen because you’re running a particular set of ______, but rather, it’s because of the summation of ______ and ______ on the retinal layer and _______ activity rules. Then, in layers further up, there are neurons responsible for certain things (eg: orientation, colour, etc).
- Neurons representing the same location ______ excite each other.
- Neurons representing different locations ______ each other.
- Neurons involved in recognition also are involved in lateral _______.
- lateral inhibition causes units on all _____ that represent a single object to become ______.
- _______ cuing - this cue _______ the location, helping subsequent target to win the competition sooner
distributed function rules excitation inhibition threshold
mutually inhibit inhibition layers active Posner pre-activates
BRAIN vs COMPUTERS - HOW KNOWLEDGE IS REPRESENTED
Irina Harris’ memory slide describing parallel distributed processing demonstrates that information about bird is ______ across networks, and the “bird category emerges from the _______ of all instances. Alternatively, the “emu” representation doesn’t overlap as much as the representations of other birds, and can cause a _____ response.
A computer does not do this. It would have ______ of emu features, magpie features, etc. The _______ of features do not help computers get the response ______.
–> MAJOR DIFFERENCE BETWEEN COMPUTERS AND BRAINS - COMMON ELEMENTS HELP THE BRAIN (reuses the neurons). BUT THIS DOES NOT HELP COMPUTERS.
shared overlap lists overlap slower
lists
overlap
faster.
COMPUTERS AS A METAPHOR OF THE BRAIN
The default way to model the brain is to think about how ________ do it. This is because computers are the only thing we have (or can make) that does similar things to us. We don’t really have a good sense of the brain from this, but we can look at the _______ between brains and computers, and this can also help us understand the brain.
The computer can be a good metaphor, but can also be _______.
Eg: brains never really get “____”. But a computer memory can…
computers
computers
difference
misleading
full
BRAIN VS COMPUTER - KNOWLEDGE AND MEMORY
In network memory (brain) the same _____ (neurons) are used for many lists (eg: tofu, chicken, etc), as each of the units are connected to each other
- PROS - less ______, Generalisation makes it work _______ (can get things done ______) due to overlap. This generalisation happens ______.
eg: faster at writing down cheese once you write milk - -> extracting common elements/associations
- CON - these are more susceptible to _______.
eg: parts of the network might become activated that you don’t want to become activated simply because they are connected to each other.
Computers - separate lists with each of the units
PROS - separate lists for things means less _________.
CONS - doesn’t matter how many lists are remembered…common elements doesn’t make it any ______ (no generalisation)
eg: no faster at writing down cheese once you have written milk
units
space faster faster automatically interference
interference
faster
BRAINS vs COMPUTERS
MEMORY AND INFORMATION PROCESSING
Recognition, memory, and computing power in the brain are _______ - they are made up of the _____ units (neurons). But in the computer these are _______ (the memory is separate from the computing power). All information processing has to go through the CPU.
intertwined
same
separate
CONNECTIONIST BRAIN VS COMPUTER
CONNECTIONIST
- individual concepts are widely ______
eg: there are lots of neurons representing lentils and they’re various elements (including their ______ with similar things like black beans). - content-_________ (you identify items by the items themselves and their features, not by location AND once you activate one aspect of it, it activates various elements of that _______ that are related to it (activation spreads).
- Naturally/automatically _________ (WISDOM?)
- Some capacity _______ (there can be interference as memories accumulate)
- -> BUT
- Prone to _________ (esp. when retrieving memories)
COMPUTERS
- memories _________ (in one area) and completely _______ from everything else.
- no _________
- -> BUT
- _____-consuming (looks though list in a serial manner - can’t look up things via the item itself)
- no __________
- capacity ________ (once it’s full, that’s it!)
distributed overlap addressable network generalises limitations interference
localised separate interference time generalisation limited
Why is it that our brain never seems full like a computers?
Network memory is distributed across thousands/millions of synapses
New memories can degrade old ones, but it can also build on them, strengthening elements in common while degrading others (interference)
So you never hit a limit. Instead, you constantly degrading old memories and making new ones (and making new synapses in the process).
When things have elements in common we remember them more. When things are not in common they degrade (eg: playing an instrument - continuously refined)
COMPUTER VS BRAIN
- basic unit
- communication between units
- number of units (which has more?)
- speed of messages (which is faster?)
- time for single computation (which is faster?)
- storage
- robustness
- energy required
- transistor (CPUs) vs neuron
- electrical wires vs synapses and neurotransmitters
- computer sends things quicker
- brain has more neurons than the computer has transistors (CPUs)
- computer is faster (1 GHz vs 200 Hz)
- localised (manipulation separate) vs distributed memory (manipulation integrated)
- minor injury = catastrophic failure vs continually adaptive, graceful degradation
- 45 watts vs 20 watts
Early computer scientists trying to prove how _______ computers were would try and set them up against humans in doing maths or chess.
Computers win a chess game by computing all possible moves in ______. It makes many, many calculations on each position on by one. Computer uses a lot more energy to do this.
Whereas a brain does this in _______. We suspect the brain is able to ______ something about patterns, situations, and what things have in common. Over your lifetime it assesses things in a content-addressable way and uses memory from previous chess experiences.
During chess matches, a computer would evaluate hundreds of ______ of chess positions per second, but the human was only thinking about a ____…and it barely beat a human.
- But what about _______? Even if they can beat them they can’t do this…
- conclusion is work ______, not _____
intelligent
serial
parallel
learn
millions
few
generalisation
smarter
harder
Rodney Brooks worked out that getting computers to do math and chess is not the most important and difficult thing…what’s important is how they interact with the ________.
environment
