AI Rediscovers Einsteins Time Dilation, Solar Flares in the Lab, Nanosurgery & More

This Week's Science News

[This is a transcript of the video with references for those who’d rather read.]

Welcome everyone to this week’s science news. Today we’ll talk about artificial intelligence that rediscovered Kepler’s laws, solar flares in the laboratory, nano-surgery with tiny magnets, a candidate for a strange star, what the new JUICE spacecraft will look for, how much air pollution is avoided by nuclear power, a software that creates 3d models from 2d drawings, an estimate for how much rare earth metals the energy transition will need, and of course the telephone will ring.

AI Rediscovers Laws of Nature

A team of machine learning scientists from the U.S. and the UK has developed an artificial intelligence that rediscovered Kepler’s 3rd Law of planetary motion using equations.

The idea to use artificial intelligence to discover natural laws from data is not new and it’s been done before. Using software to extract patterns from data and then extrapolating them is a straightforward application of machine learning. But we’d like to have a neat set of equations that describes the data, and not some big machine learning code that we don’t really know what it does.

In the new work they have now made a big step towards that. They’ve developed an algorithm that works like you expect a theoretical physicist to work. I mean, someone’s gotta do it, right?  This algorithm analyses the data, develops hypotheses, and then tries to find a compact formula to express the hypothesis so that it supports the data.

They fed their program with three real-world data sets: NASA’s fact sheet on planets from our own solar system, NASA’s data from the solar-system TRAPPIST-1, and records of the orbits of five binary stars.  

The program then tried different mathematical relations between the input parameters – addition, subtraction, multiplication, square roots and so on – and checked how well any of those explain the data. It correctly hit on Kepler’s third Law, that’s a relation between the distance between two bodies – the d in the formula here – their mass, that’s m, and their orbital periods, that’s p. The error values were impressively small. 

They called their method AI-Descartes, after the French mathematician René Descartes who championed logical deduction as a scientific method. The system also correctly reproduced Langmuir’s Nobel-Prize-winning work on gas molecules, and Einstein’s law of relativistic time-dilation. Albert is shocked.

I think that’s all fine until the AI starts making science news about their research.

Solar Flares in the Laboratory