Conkers (the game) and Deep Learning

[ by Charles Cameron — offered to 3QD ]

.

This is my second submission attempting (and failing) to become a regular contributor at the fine web-aggregator known as 3QD — you can read my first here. On rereading this more recent attempt, I am not sure I would have selected it myself had I been one of the judges — it perhaps expects too mych “British” knowledge of its readers, who are as likely to come from Hoboken or Pakistan as from Oxford (hat-tip there to my two 3QD friends, Bill Benzon & Omar Ali). For those of you, therefore, who may not know what conkers is, here’s a brief video introduction, with the actual game demo starting around the 1’35” mark:

I’ll add a fantastic passage from Seamus Heaney at the end of my post, to give it the final mahogany polish a conkers post deserves.

____________________________________________________________________________________________

Three Quarks Daily submission:

Conkers is for kids. So what does it have to do with computers, AI, “deep learning” or robots?

It’s a very British game, conkers, played with the seeds of horse chestnut trees, of which Britain counts almost half a million, pierced and strung on string. I suppose you could think of it as a primitive form of rosary bashing, with each player’s rosary having only one bead, but that might give others the wrong impression since conkers (the game) is sacred the way taste is sacred before anyone has told you, “this is strawberry”, not the way sanctuaries are sacred after someone has put an altar rail round them, or screened them off with a rood screen.

Even giving the game the name Conkers with a capital C puts it on a pedestal, when all it’s about is gleefully finding a suitable conker, maybe loose on the ground or maybe encased in its prickly green shell, drilling it through with some sort of skewer or Swiss Army device, threading it (stringing it) on string, finding a gleeful or shy playmate, and whazzam! conking their conker with your own so their falls apart and yours remains triumphant. Battered perhaps, but victorious.

An artificial general intelligence, left to its own devices and skilled in “deep learning”, will surely figure out that play plays a significant role in learning, by reading Johan Huizinga perhaps, or figuring out “the play’s the thing” – or noting that play is how learning develops in mammalian infancy and expresses itself in human mastery. And since learning is what artificial general intelligence is good for and would like to be even better at, you can bet your best conker that artificial general intelligence will want to learn to play.

Okay, computers have played, and beaten humanity’s best, at such games as Pong, Space Invaders, Draughts, Backgammon, Chess, Jeopardy, Rock-Paper-Scissors, and Go –

–

— and apparently one set of AI researchers is considering Texas Hold’em as a plausible next challenge. I want to see them try their hands (hands?) and minds at conkers, though.

Page 1 of 4 | Next page