Towards a Less Artificial Intelligence

Speaker: Christos Papadimitriou

Location: Off-Campus, Room Flatiron Institute

Date: Thursday, March 9, 2023

Today's AI still lags behind brains in important dimensions.  Can we make progress through artificial systems which adhere to the basic tenets of neuroscience -- for example, do not rely on backprop?  I will present an abstract model of the brain entailing brain areas, excitatory neurons, Hebbian plasticity, local inhibition, and long-range inhibition and disinhibition.  This model is a Turing-complete hardware language, and can also be seen as a novel kind of software-based neuromorphic computation, which can be simulated efficiently,  through lazy evaluation, at the scale of tens of millions of neurons and trillions of synapses.  I will describe experiments with this model, including current work on a biologically plausible language acquisition system: A neural tabula rasa of areas and fibers which, on input of modest amounts of grounded language, will create a mature language organ with a lexicon, semantics, parser, comprehender, and generator.

There will be a happy hour reception at Flatiron afterward. Make sure to register on Eventbrite before you attend!

Registration: https://www.eventbrite.com/e/ml-nyc-speaker-series-christos-papadimitriou-tickets-563828334727