The science fiction (SF) novels of Iain M Banks are set against the backdrop of an anarchic interstellar civilisation that calls itself The Culture. Insofar as it's governed, The Culture is run by artificial intelligences, which are vastly more intelligent than humans.
The Culture doesn't have a currency. It doesn't need one. Nobody has to work. The Culture is resource-rich and intelligent or semi-intelligent machines do any necessary tasks. Food, clothing, technology, transport, etc., are available for the asking.
The concept of machines that are self-aware, auto-didactic and capable of outthinking human beings have been standard elements of SF for a long while. The examples range from the ultra-hostile Skynet, to the madly malign HAL9000 to the benign Multivac.
Moral ambiguities aside, silicon intelligences currently outperform their makers in many ways. Programs are excellent financial traders. They are better drivers and pilots. They are better at calculating how to pack cargo into containers. They are better at weather forecasting. They are also better clerks than any Macaulay-putra could ever be. Programs also play much better chess and draughts.
Another bastion of human superiority crumbled when the AlphaGo program beat Lee Seedol, one of the world's best players of Go (the game is capitalised here to distinguish it from a command at dog obedience school). AlphaGo recently won a five-game match by 4-1 with Seedol winning the fourth game.
Go is way more complex than chess. It is played on a 19x19 board (chess is 8x8). There are 361 ways to start a game and roughly 2x10ˆ170 games are possible (2 followed by 170 zeros). The average game lasts 200-odd moves (by each player). Counting may be required to assign a clear result to each game. Unlike chess which has a maximum number of 32 pieces, a theoretical maximum of 361 stones (181 black, 180 white) might be in play.
In contrast, chess has "only" 20 opening moves; an average chess game lasts 40 moves and the total number of possible chess games is about 10ˆ47 (1 followed by 47 zeros). The number of stars in the observable universe are estimated at 3x10ˆ23. So we are talking very (repeated umpteen times) large numbers.
There are other problems with programming a machine to play Go. One is a lack of "hierarchies of value". In chess, the value of the kings are infinite and the relative strength of different pieces is easily distinguished. This makes it relative easy to program good chess moves. In Go, every stone has the same value. That makes the evaluation of Go positions much more difficult.
Humans learn to play Go by pattern recognition. They watch and note patterns that work better. There are prodigies as in music, chess and mathematics. The world's best Go player is an 18-year-old, Ke Jie (AlphaGo would be #2 if this victory had been in an official match).
AlphaGo was developed by DeepMind Technologies, an UK-based Google subsidiary founded by Demis Hassabis, a former chess prodigy who is also a strong player of Shogi (Japanese chess) and poker. The program uses "deep learning". The method is suspiciously close to being human in approach. The program is shown example games and learns to play without being "taught" good or bad moves. It is a learning approach that can be applied to many tasks, quite apart from other games.
Software programs written with a more conventional approach are not too strong at several other games. Bridge for example, requires an ability to communicate with a wide variety of people and the best humans are a little better than the best programs. Shogi has the concept of captured pieces changing sides and "turn-coating" is also a hard concept to program.
But if deep learning can handle Go with this level of efficiency, it could probably learn bridge, or Shogi with high competence. it could prepare a tax return, or a legal brief. It may even be a good diagnostician if it eavesdrops on enough medical consultations. Could this be the beginning of the end for a vast range of professions?
Twitter: @devangshudatta
Disclaimer: These are personal views of the writer. They do not necessarily reflect the opinion of www.business-standard.com or the Business Standard newspaper
