Humans are dumber than computers—that’s the consensus after that Watson thing last year, right? Not if they can’t beat Will Shortz, says Matt Ginsberg, creator of a computer program that’s designed to best even the most cunning human cruciverbalists. To prove it, he’s entering his so-called Dr. Fill in next month’s American Crossword Puzzle tournament.
What makes Dr. Fill so smart, so worthy of competition with those speedy word nerds? Memory. He’s basically memorised the entire cannon of crossword puzzles to plug-and-play as needed. Unlike Watson, Dr. Fill wasn’t programmed to understand the language of the clues, but rather to play the game, explains The Economist:
His software contains all clues and answers for crossword puzzles created since 1990, and can break down unknown clues into long lists of potential answers that fit the required word or phrase length. Dr Fill fills in squares with its most confident pick, and then works through potential crosses. The trick in making this large problem smaller is the crossed words, he says, as it reduces the universe of possibilities in comparison to open-ended questions.
But even if this programming method works, is doing crossword puzzles as human a task as playing chess or answering questions on Jeopardy? Will we have lost even more ground in the robot-human brain race? Probably not; tapping a database is a far cry from logical reasoning. That, and I suspect even Dr. Fill will have a hard time parsing NY Times crossword guru Will Shortz’s stabs at hip-hop culture clues. [Economist]