I'm the Computer Now

Wave physics as an analog recurrent neural network
Jan 2020, phys.org

This story is the craziest thing I've ever heard -- mind-numbing omnipotent future-speak. Computing with physical wave media in virtual dimensions? I'm in.

The electronic engineering department at Stanford "trained" physical wave systems to "learn" temporal data features.

If you see the word "train" these days you can bet they're talking about deep learning / neural networks. Usually, the way you train a network is to let it watch thousands and thousands of hours of youtube videos. Or any other data set you have available. (The Alcohol Language Corpus of drunk speech, perhaps?)

The network itself learns certain things about all the data you give it, and it assembles itself, arranges itself, to be able to detect those features in other images, let's say. Or you could have it learn what a "Vietnamese sandwich" is, and then synthesize Vietnamese sandwich versions of other sandwiches. Or you could have it show you the eignevector, the essence of a Vietnamese sandwich. (Spoiler, the essence of youtube is the face of a cat. That's what the network learned most by watching a several lifetime's worth of internet-video).

How is eigencat not a thing? Try that search for yourself, "youtube eigencat"

But these guys are talking about teaching not a network, but a wave of air. Not only is it not alive, it's not even a computer! Then again, the waves are the computer:

As proof of principle, they demonstrated an inverse-designed, inhomogeneous medium to perform English vowel classification based on raw audio signals as their waveforms scattered and propagated through it.
-source

You read that correctly. There are no circuits, no electricity. The physical system itself performs a recurrent processing dynamic, based on the way the waves move through space.

It's the new thing, analog machine learning.

Check out this old news on computers made of other stuff that's not computers, like the Magic Dust Supercomputer, Slime Mold Computer, Crystal Calculators, and of course, the DNA computer.

Growing crystals to generate random numbers
Feb 2020, phys.org

Generating random numbers has always been a tricky problem for computer engineers because computers were designed to be as predictable as possible. 

One of the more pressing applications of random number generation is data encryption—most existing schemes rely on the constant generation of random numbers. 

The process of crystallization is random due to many factors that come into play as chemicals in a liquid solution evolve from a disordered state to one that is very organized. 

A camera took a picture of each of the cups as crystal formation began. Each of the pictures was converted to a zero or a one based on nothing but the geography of the crystal. The zeros and ones were then strung together to form a random number.

New study allows brain and artificial neurons to link up over the web
Mar 2020, phys.org

Novel Nanoelectronics has enabled brain neurons and artificial neurons to communicate with each other using brain-computer interfaces, artificial neural networks and advanced memory technologies (also known as memristors).

Cultivated lab rat neurons from Padova were distributed into memristive synapses in Southampton, and then connected to artificial silicon neurons in Zurich. 
-Alexantrou Serb et al. Memristive synapses connect brain and silicon spiking neurons, Scientific Reports (2020). DOI: 10.1038/s41598-020-58831-9

Advanced 'super-planckian' material exhibits LED-like light when heated
Mar 2020, phys.org

"We believe the light is coming from within the crystal, but there are so many planes within the structure, so many surfaces acting as oscillators, so much excitation, that it behaves almost like an artificial laser material," Lin said. "It's just not a conventional surface."

The robot that grips without touching
Jan 2020, phys.org

Acoustic waves -- method that makes it possible to lift and manipulate small objects entirely without touching them.

Chilling out - Physicists create exotic “fifth form of matter” on board the ISS
June 2020, Ars Technica

Algorithm quickly simulates a roll of loaded dice
May 2020, phys.org

Random number generation:

The algorithm, called the Fast Loaded Dice Roller (FLDR), was created by MIT graduate students and research scientists. 

FLDR can use up to 10,000 times less memory storage space than the Knuth-Yao approach, while taking no more than 1.5 times longer per operation.

Growing crystals to generate random numbers
Feb 2020, phys.org