Abstract

A recording of this talk is available at https://www.youtube.com/watch?v=i9GMuk0gffM and further resources can be found at https://gitlab.com/apgoucher/DLAE (including complete source code).

The question of whether a mechanical automaton could reason as we do has been posited for as long as such machines have existed. Real progress in realising machine intelligence, however, has eluded us until very recently with the advent of deep learning. But why has it taken so long? Does intelligence inherently require vast swathes of memory and lightning-fast multiprocessors? Or could it equally well be embodied by, say, the very first computer ever envisaged?

We transport ourselves two centuries into the past to relive the conception of the Analytical Engine: an immersive Victorian adventure marked by lavish soirées, gargantuan contraptions, and promiscuous poets. We continue this story by investigating how a 21st-century approach to algorithmic intelligence can be implemented on a 19th-century machine, encountering many obstacles along the way.