Artificial intelligence beats humans at crossword solving and assesses the health of ‘singing’ coral reefs

27 May 2022 Michael Banks

Sounds good: a healthy coral reef in Sulawesi, Indonesia. (Courtesy: Tim Lamont)

Many people around the world enjoy doing crosswords, taking a few moments out of their day to test their knowledge and wordplay by answering clues to find words that complete the crossword grid. Crosswords are a puzzle that humans particularly excel at, so much so that we can outperform machines — or so we thought.

One of the most prestigious crossword competitions is the annual American Crossword Puzzle Tournament and the best that a machine could do in 2017 was 11th place thanks to a programme called Dr Fill. But machines are now making a comeback and in 2021 the Berkeley Crossword Solver – built by a team at the University of California, Berkeley – managed, for the first time, to outscore every human player.

The programme, which is outlined in a paper on arXiv, is based on “neural question answering, structured decoding, and local search,” which involves finding solutions to questions and then using databases and natural language artificial intelligence (AI) to refine the answers. And despite such progress, there is still room for improvement. The Berkeley Crossword Solver has an “exact puzzle accuracy” of 82% for crossword puzzles set by The New York Times – compared to 57% in previous incarnations. Crossworders beware.

Singing coral reefs

Another novel application for AI that we came across this week is using the technology to assess the health of coral reefs. An international team led by researchers at the UK’s University of Exeter and Lancaster University have used AI to analyse the sounds emitted by reefs. These “songs” are made by the rich variety of fish and other creatures that live on the reef – and the presence or absence of these organisms is related to the health of the ecosystem.

The team trained their AI system using audio recordings made at the Mars Coral Reef Restoration Project in Indonesia. The recordings were from both healthy and damaged coral reefs and once trained, the system was able to identify the health of a reef 92% of the time.

“Coral reefs are facing multiple threats including climate change, so monitoring their health and the success of conservation projects is vital,” says Exeter’s Ben Williams. He adds that assessing the health of a reef using conventional visual and audio techniques can be labour intensive.

“Our findings show that a computer can pick up patterns that are undetectable to the human ear. It can tell us faster, and more accurately, how the reef is doing,” adds Williams.

Lancaster’s Tim Lamont adds, “In many cases it’s easier and cheaper to deploy an underwater hydrophone on a reef and leave it there than to have expert divers visiting the reef repeatedly to survey it – especially in remote locations”.

The research is described in Ecological Indicators.

Michael Banks is news editor of Physics World magazine

from physicsworld.com    29/5/2022