The long and winding Path of Go

The Path of Go is a nice puzzle-based game that has just appeared in the Xbox LIVE Arcade.

If you download it, it will cost you just 400 Microsoft Points. You could therefore be fooled into thinking that it is just a simple little title, knocked up in a couple of months by a games development team.

But the story of the Path of Go is a long, arduous and intriguing one - and one that could have much wider implications in the world of Artificial Intelligence.

For the Path of Go is merely a by-product of years of AI research by the Microsoft boffins - and Pocket-lint was invited behind the scenes so they could tell us all about it.

The game's main developer Markus Jost won a game developing competition set up by Microsoft and, as a result, was offered an internship within the Microsoft Research team - where he began work on The Path of Go.

However, the Artificial Intelligence required to power the computer players in Go is so intricate that the title soon took on significant importance within the department. Microsoft Research isn't really in the business for bashing out Xbox LIVE games, but the Path of Go represented a challenge in terms of the advancement of AI.

"It provided an abstract test bed that was perfect for a number of key AI principles", said Joaquin Quiñonero Candela, the project lead and senior applied researcher at Microsoft Research Cambridge.

So why exactly is Go so complex? After all, developers seemed to have cracked the AI aspect to Chess years ago.

"We still compare it in terms of complexity to the game of chess in the western world", said Robert Mateescu, Microsoft post doc researcher and semi-professional Go player. "But it's a mind game of high complexity and it might take your entire life to master the game".

Comparing it to chess, where there are an average of 20-30 move options per turn (with 20 starting moves), just from the starting position in Go there are 361 different combos on a 19 x 19 board, with a total number of Go scenarios being 10 to the power 700. So, the AI needed was far more advanced than that of Chess games in the past.

What the Microsoft Research team came up with was an amalgamation of a number of advance technologies including TrueSkill (as used across many Xbox LIVE games), a machine-learning system, F# - a functional programming language used to develop the AI, and XNA which is a set of tools with a managed runtime environment.

The AI on board the game is very similar to that which can be found in search engine algorithms. Thore Graepel, Artificial Intelligence and TrueSkill senior researcher explains:

"A user going to a page is presented with a number of ads, and he decides to click one ad, and thereby tells us that for him, that ad was better than all of the other ads.

"On the board, the professional player has a choice of all the legal moves he could play and they pick one particular move and we know then from the game records that one particular move is the best move, similar to the user who clicked the ad. We model these methods on the same process".

The result, for now at least (this tech is bound to form the basis of future Microsoft Research) is a nice strategy-based game that will take even advanced Go players five or so hours to complete - with the time for novices being more like 15-20 hours (all on the easier 9 x 9 board).

There are tutorials to teach you the techniques required, and a nice adventure-story based narrative that throws in full games of Go against various opponents, with mini-puzzles to solve along the way.

The graphics are nice (the lake with the Koi carp is very impressive) and there are lovely touches when it comes to the materials of Go boards and the actions that your avatar can do.

So will the game be a success? According to Candela, that is the one thing that isn't too clear.

"Is Go going to work on the Xbox, for people who are used to blowing other people up?" he said. "The answer is, we'll see".

You can see for yourself - The Path of Go is out now on Xbox LIVE Arcade, for 400 Microsoft Points.

 



>