Nancy Kanwisher
Last modified on May 27, 2022
Talk she gave to the Symbolic systems program on May 27.
A roadmap for research
There are many different areas to the brain:
- face area
- scene area
- visual word area (ppl don’t have this response until after they learn to read— and only in languages they know!)
- auditory speech area
- language area
“what is the relationship between language and thought?” > thought ! language! two diff. brain regions respond much differently to thought and language. also left-hemisphere stroke patients — they can still think.
Causal role?
fMRI is nice…but what’s the causal role of these regions? We don’t just want correlations of brain activations & activities.
We need experiments where we “poke” part of the system – intervention. Schalk et al.
If he’s looking at a face, the face changes. If he’s looking at something else, it adds a face to that object.
“Poking the face area” results in weird, weird face stuff happening to brain patient
Stimulated color regions – he saw a rainbow (wtfff)
What other specializations?
not their appearance, sensations, etc.
NOT every region has a highly specific function – large swaths of brains are “multiple demand” regions.
3P Social Interactions region – seeing other people interacting with each other.
Physics regions – reasoning about the physical world.
Interesting frame – data doesn’t need to be collected “For” a specific hypothesis – instead, they collected data while people listened to 165 everyday sounds. Looked at the components of the neural response.
Six “components” of cortical response to sounds
One component: responds to speech.
Another component: music!
Why do we have music?
Music “co-opts” speech/language mechanisms – NO. still a mystery.
Lol the textbook picture of the brain is so messy in so many ways – it’s not always neatly divided into left/right hemisphere. Many, many functions are bilateral, whereas others are right- or left-lateralized.
- there’s a pink patch that REALLY selects for thinking about others’ thinking
Developmental origins?
Face area takes years to arise, requires visual experience…?? – interesting question of nature vs. nurture. Actually, Rebecca Saxe found that you can get face responses in INFANTS. Heather Kosakowski scanned 83 awake infants.
^Hyo question: what does the “pretrained” network in the brain look like? A: we don’t really know.
My question: where is the supervision “signal” coming from in the brain?
^markman question: can we correlate the actual training regime of NNs with development? A: Seems much more sus.
congenitally blind subjects respond to touching faces!!!
Evolutionary origins?
some stuff is shared (common ancestor) –
face-color-place: wtf?? those regions are “sandwiched” together in monkeys and humans.
What’s represented and computed in each region?
Astonishing success of convnets…analogs between them and the brain perception. Model of how vision “might” work in the brain. (Whoa, that’s fantastic to hear a NEUROSCIENTIST say that. We can consider them a model of the brain!)
Ratan Murty – CNN-based models of FFA, PPA, EBA? Concretely – get CNN activations, and see if you can predict the fusiform face area responses. really really high correlations.
Oh interesting – then you can test the trained model on all of Imagenet, and see what it responds to…ONLY faces. damn.
LMAO can feed things into GANs too.
Why does the Brain have category-selective regions in the first place?
Old hypothesis: face specific regions needed, since face recognition needs diff features than object recognition.
Well…the dual-trained network does just as well on both – they “lesioned” the network and found that the network had spontaneously separated itself into two separate networks – one for object detection, one for face detection.
There are diff feature spaces necessary for different problems – brain learns to segregate those into different systems.