Incoming from Rob Fisher: link to a piece in the Independent, about machine learning applied to old telescope data is finding new planets.
A computer was trained to look through the data from the Kepler space telescope, and look for signals that might belong to planets. And it found new planets within existing systems, by spotting signals that seemed to indicate something of interest but were too weak to have been spotted by humans.
That suggests that there might be whole worlds and solar systems hiding within the data we’ve already collected, but which we had not noticed because there are simply so many signals to pick through. Kepler has collected four-years of data from looking at the sky and 150,000 stars – far more than humans could ever look through.
So, exactly what were these weak signals?
The new planets – just like all of the thousands found by Kepler – were spotted by watching the sky for light coming from the stars. When planets pass in front of their stars, scientists can register the dimming as they go, and use the speed and characteristics of that dimming to work out what the solar system might actually look like.
Much of that work relies on pattern recognition, which until now has been done by scientists looking through the data. But the new findings are the result of work between Nasa and Google, which trained machine learning algorithms to learn to spot those patterns itself and so pick through the data much more quickly.
This is good. Keep Skynet busy with harmless hobbies.
Maybe not. Getting Skynet to compile a huge and exhaustive list of all the places in the universe where biology-based life might be, after biology-based life on this planet has been taken care of.
This is maybe how the robot holocaust will happen. We will have been telling them to “take care of” us and our fellow creatures. But they’ll have been watching too many gangster movies, and …