Just presently, during the health hiatus, I am playing perhaps six games of Scrabble a week, probably as many in total as I had clocked up in the previous seventy years.
So prompted by reading once again of computers which play checkers and of computers which play chess, what follows is some speculative thinking about what a robot or a computer might do in the way of playing this quite different game, with the general idea being given in the snap above. Between two and four players each play with a bank of seven tiles, drawn at random from the stock. All played words, other than the first word, have to be connected to one or more existing words and have to be a recognised word, usually in some dictionary based, dictionary defined sense.
A previous venture into the field was noticed at reference 1. For Scrabble more generally, see references 2 and 3.
The first thought is that the computer would win hands down. A modern computer, even a hand-held one, would have a perfect vocabulary and be able to do a perfect search of the move space in game reasonable time – winning which might be regulated down by means of some kind of handicap system so that the computer did not lose interest. A handicap system which might work by introducing a degree of randomisation into the computer's plays, by it not always making the best play available. Such randomisation would have the side effect of confusing the human players’ responses; they would have even less knowledge about what the computer was going to do next.
But there are plenty of other possible complications.
First, what could constitute a perfect vocabulary? With most people not liking the simple rule of bold head words in a standard English dictionary, mainly because of the likely exclusion of all kinds of grammatical derivatives – for example biter, bitten and bites from bite, hospitals from hospital and sheaves from sheaf. House rules and a sense of fair play can deal with most of this – but would a computer understand either?
One way to handle this would be for the computer to hold a curated list of words which were allowed and which could be inspected and updated; a list which would gradually get better, gradually align with the house rules and the sense of fair play.
Second, do we allow the game theory people in? So does the computer just put down the highest scoring word available at the time, or does it have regard to what that might be giving its opponent or opponents?
Does it have regard to its own next play? That is to say, does the computer take account of what it might be able to do when its next turn came around? Does it assess the risk of its opponent making use of the opportunity that it was creating for its own use?
Does the computer even know about its opponents and their moves – or does it just know what is on the board when its turns comes around? Is the computer allowed to allow for the fact that this particular opponent is likely to make this particular sort of mistake or to overlook this particular kind of opportunity?
It is also true that the humans could do this sort of thing too. They might, for example, detect and exploit weaknesses in the computer’s play.
Lastly, it would be straightforward for it to know about the tiles that have been played and the tiles in its own hand. The starting set of tiles is known. And from all of this it knows something about the tiles left in the bag and the tiles in the hands of the opposition. All information which might be used to its advantage; information which the average human player might find difficult to make use of. Not enough mid term memory.
Information which might also be useful in the case that the computer was thinking of exchanging some dud tiles with the bag, rather than putting down a word.
Third, there would need to be some rules around the time taken for a move. This would probably include taking so long that opponents got bored and took their minds off the game. But would this also include forbidding the other wheeze of slapping down one’s move immediately after that of an opponent, thus giving the opponent no time to rest? Thus putting the opponent under some psychological pressure?
Fourth, the rule for challenges would probably need to be modified. In which connection one should note that most players do not allow dictionaries to be browsed during play; they may only be used to resolved challenges. Rules perhaps modified to say that the computer had the last word on what words were allowed. That is not to say that the computer would always use a legitimate word when it was in play mode: it might play with its opponents by daring them to challenge plausible but illegitimate words. Or implausible but legitimate words. And it might chose not to challenge an illegitimate word which gave it a good opening. But it would have to be sporting and tell the truth when there was a challenge.
What we have suggested so far could be handled by a personal computer being looked after by a human, with a human feeding in information about tiles and human moves and reading out the computer’s moves. So, fifth, do we want a full on robot, able to see the board for himself, to pick his own tiles from the bag and to place his own tiles on the board. Various wheezes that such capability make possible come to mind.
A robot that could read his opponents, probably on the basis of visual cues, and know when one of them had a coup in the offing which needed to be blocked?
A robot that could deduce information about the tiles an opponent was holding by the way he sorted them in their rack, rather in the way that some card players can tell something about the cards an opponent is holding?
A robot that could deduce information about an opponent’s next move by the way he looked at the board? Perhaps a robot that could see through attempts to put him off the scent by conspicuously looked at the wrong part of the board.
And less sporting, a robot with very sensitive fingers able to discretely feel up the letters in the bag, so improving his selection?
A robot that could decipher stray reflections containing images of opposition tiles?
All of which might be useful when one has trouble getting to sleep. The thinking person’s alternative to counting sheep or going through the alphabet with things like the highest mountains in the world or the coastal towns of England.
References
