DTC Journal Club – Friday 10th June 2016

Speaker:  Sander Keemink

Title: AI and the Pesky Videogames

Abstract:  One by one they fell. Chess, they said, chess is a human game, it will be decades before an AI exists that can beat the best. Well, but, actually, Go, they said, Go, is the difficult one that will take decades yet to crack. Even Arimaa, a game specifically designed to be difficult for AI’s, has beaten the dust long ago.

The year is 2016. Games have been entirely conquered our future robot overlords. Well, not entirely…. One game (or rather genre) still holds out against the artificial intelligences. And life is certainly not easy for the AI’s that have made the first attempts at winning the game of Starcraft, Overmind, Krasi0, Skynet and others…

 

Speaker: Agamemnon Krasoulis and Matt Graham

Title: Could a neuroscientist understand a microprocessor?

Abstract: The articles ‘Can a Biologist fix a Radio?’ (Lazebnik, 2004) and ‘The tale of the neuroscientists and the computer: why mechanistic theory matters’ (Brown, 2014) provide slightly tongue-in-cheek yet thought-provoking critiques of the research methods used in molecular biology and neuroscience respectively. Both use thought-experiments of considering what would happen if experimentalists of various flavours attempted to use the standard experimental techniques of their fields to tease apart how a well understood artificial system (transistor radio and desktop computer) works.

Both articles come to a similar conclusion that there is a need for a greater use of standardized frameworks for describing and constructing mechanistic models in biological research which provide a common and unambiguous basis for researchers to communicate ideas to each other and build upon each others research. Examples such as the common visual and mathematical languages of circuit theory in engineering and the standard model in physics, are held up as examples where such shared languages and models have facilitated the progression of physical sciences.

What those two articles lacked however was an actual instantiation of the proposed thought-experiments. In their just released pre-print, ‘Could a neuroscientist understand a microprocessor?’, Jonas and Kording, go that extra step and apply various neuroscience analysis techniques to a transistor level simulation of a simple microprocessor to try to tease apart how it functions. In our talk we will introduce the model organism (MOS 6502 microprocessor) and standard behaviours (Donkey Kong, Space Invaders and Pitfall) studied by the authors and some of the conclusions they draw from their results.

Given the personal experience (and strong views) of many in the DTC of working at the interface between the methods and practices of biological and physical science research, we hope for there to be some interesting discussion of the issues raised!