Tim Behrens

1760: Bayes: You should interpret what you see in the light of what you know. 1780: Galvani: Nerves have something to do with Electricity. 1850: Phineas Gauge et al: Different parts of the brain do different things.

1850: Bois-Reymond/von Helmholtz: Cells fire electrical impulses down axons. 1880: Golgi: There are beautiful things in the brain but they are all meshed together in a gloop. 1890: Ramon y Cajal: The beautiful things are all separate things called neurons.

1900: Sherington: They connect to each other with synapses. 1900: Pavlov: Brains can learn predictions. 1910: Thorndike: Predictions of reward control behaviour. 1920: Helmholtz: The Brain is a Bayesian inference machine. 1930: Lowei/Dale: Synapses are chemical.

1940: Skinner: All behaviour can be accounted for by reward prediction. 1940: Tolman: No it can’t. You need a “map”. 1940: McCulloch & Pitt: Networks of neurons can perform computations

1950: Hodgkin and Huxley: Those nerve impulses are caused by a dance of ionic currents. There are probably some ion channels. 1950: Hebb: Synapses can store memories.

1950: Lashley: Memories are represented across the whole brain equally. 1950: Penfield: Different parts of the sensorimotor cortex do different things. 1960: Milner: There are at least two forms of memory and hippocampus is only important for one of them.

1960: Sperry: Woah spooky!! The two hemispheres are actually different people. 1960: Hubel and Wiesel: Individual neurons represent real-world things in their activity. 1960: Barlow: Neurons transmit information and should do so efficiently.

1970: Marr: Cerebellar, Hippocampal and Neocortical circuits instantiate different computations. 1970-1980 Several, including Neher and Sakman: Yes there are ion channels. 1980: O’Keefe and Nadel. Hippocampal neurons form a map of space.

1980: Hinton/Senjowski/McLelland/Rumelhart:Neural networks can optimise some cool shit. 1990: Van Essen:Blimey the visual system looks complicated. 2000: Rao/Ballard/Olshausen/Fields:Those real world things that neurons represented are actually features of statistical learning

2000: Schultz, Dayan, Montague: Dopamine neurons allow reward predictions to be learnt. 2000: Kanwisher: Different parts of the higher visual cortex do different things. 2000: Haxby: No they don’t. It’s all distributed (are we still having this same argument?)

2010: Mosers: Entorhinal cells form a better map of space. 2000-2010: Deisseroth, Meisenbock, Boyden et al: Fuck me light. 2015: Hinton/LeCunn/Silver/Hassabis and many others: Neural Networks can optimise some REALLY cool shit.

2015-2020: This neural network looks just like my neurons. 2020: Jesus that’s a lot of neurons.