Your Brain is a Prediction Machine

AKA Your Brain on Music

First, some sound tech:

Surround sound from lightweight roll-to-roll printed loudspeaker paper

Jan 2021, phys.org

https://phys.org/news/2021-01-lightweight-roll-to-roll-loudspeaker-paper.html

Sonorous paper loudspeakers: "Ordinary paper or foils are printed with two layers of a conductive organic polymer as electrodes. A piezoelectric layer is sandwiched between them as the active element, which causes the paper or film to vibrate. Loud and clear sound is produced by air displacement.

via Chemnitz University of Technology: Georg C. Schmidt et al. Paper‐Embedded Roll‐to‐Roll Mass Printed Piezoelectric Transducers, Advanced Materials (2021). DOI: 10.1002/adma.202006437

Image credit: Center for High Entropy Alloy Catalysis - Danish National Research Foundation

Next, some music science:

Hit songs rely on increasing “harmonic surprise” to hook listeners, study finds

Aug 2021, Ars Technica

https://arstechnica.com/gaming/2021/08/hit-songs-rely-on-increasing-harmonic-surprise-to-hook-listeners-study-finds/

The end result is what the authors call "Inflationary-Surprise Hypothesis," which makes music (and all art, and all culture we would assume) to have more surprise over time. And so if you're kind of old, and you think the music of today sucks, and it's annoying, stupid, and makes no sense, then you're probably right. 

In other words, when a song "constantly defies the listener's expectations throughout", that's what you want to hear. Or, from the paper itself, you could say "human perception of tonality is influenced by exposure."

We heard this a long time ago when Leonard Meyer said it --

"A culture, like a musical style, is a learned probability system.” (p17, footnote 21)

-Music, the Arts and Ideas. Leonard B. Meyer, U. of Chicago, 1967

Or take a look at this infographic, titled The Anatomy of a Joke.

What to Expect When the Unexpected Becomes Expected: Harmonic Surprise and Preference Over Time in Popular Music. Front. Hum. Neurosci., 30 April 2021. https://doi.org/10.3389/fnhum.2021.578644

And more of the same:

The brain's 'prediction machine' anticipates the future when listening to music

Aug 2021, phys.org

https://medicalxpress.com/news/2021-08-brain-machine-future-music.html

When a musical phrase has an unresolved or uncertain quality about it our brains automatically predict how the melody will end. ... Like a sentence, a musical phrase is a coherent and complete part of a larger whole, but it may end with some uncertainty about what comes next in the melody. The new research shows that listeners use these moments of uncertainty, or high entropy, to determine where one phrase ends and another begins. ... The participants judged melodies that ended on high-entropy tones to be more complete—and lingered on them longer.

"This study shows that humans harness the statistical properties of the world around them not only to predict what is likely to happen next, but also to parse streams of complex, continuous input into smaller, more manageable segments of information," said Hansen.

via Association for Psychological Science: Niels Chr. Hansen et al, Predictive Uncertainty Underlies Auditory Boundary Perception, Psychological Science (2021). DOI: 10.1177/0956797621997349

For your listening pleasure:

Music Circles: An interactive data visualization tool that helps users discover new music

Mar 2021, phys.org

https://techxplore.com/news/2021-03-music-circles-interactive-visualization-tool.html

It's an advanced music recommendation system. Try it out, called Music Circles:

https://musiccircles.netlify.app/

via Seoul National University: Music-Circles: can music be represented with numbers? arXiv: 2102.13350 [cs.HC]. arxiv.org/abs/2102.13350

Post Script:

Taylor Swift releases a 'perfect replica' of Fearless

Apr 2021, BBC News

https://www.bbc.com/news/entertainment-arts-56687699

Up to now, I thought Taylor Swift was a good musician but more importantly a good young female role model because she handles her celebrity status so well. Now she's in the pantheon of powerful and middle-finger-wielding artists because she straight re-mastered her master copies to fuck her record owners and take back control of her own work.  

If you're not sure why this is such a big deal, see David Byrne's How Music Works (2012) for a good intro to the world of musicians, their music and their rights. 

When you make a song, and then someone uses that song in the opening scene of their superfamous blockbuster movie, somebody gets paid, but it might not be you. It depends who owns the "master copy" of that song. Even if you play your own song live, and record that, and use that in a commercial or movie, it's still the "original" master copy that gets the credit, and the money. Somebody recorded that song in a music studio and made a physical copy of the recording. The music is ephemeral and nobody can own that. The idea for the song even moreso. But that recording, on that day, in that studio, is a physical thing, and it represents all the ephemera that come after, so whoever owns the master, owns it all. 

Swift's re-mastering makes the old master copies worthless, and puts her in full control of her own work, something that rarely happens to an artist as big as her. 

Artificial Intelligence is Human After All

Medical AI models rely on 'shortcuts' that could lead to misdiagnosis of COVID-19

Jun 2021, phys.org

https://techxplore.com/news/2021-05-medical-ai-shortcuts-misdiagnosis-covid-.html

Very interesting lesson for us humans, in terms of profiling and stereotypes:

The team found that, rather than learning genuine medical pathology, these models rely instead on shortcut learning to draw spurious associations between medically irrelevant factors and disease status. Here, the models ignored clinically significant indicators and relied instead on characteristics such as text markers or patient positioning that were specific to each dataset to predict whether someone had COVID-19.

via  University of Washington: AI for radiographic COVID-19 detection selects shortcuts over signal, Nature Machine Intelligence (2021). DOI: 10.1038/s42256-021-00338-7 

Image credit: Sabine62 via Fractal Forums - Weaving - 2018

Computer scientist researches interpretable machine learning, develops AI to explain its discoveries

Nov 2020, phys.org

https://techxplore.com/news/2020-11-scientist-machine-ai-discoveries.html

Finally a deep learning machine that can explain how it got its results (something previously not available, hence the term black box AI). Or is this just mansplaining? Nipsplaining.

This Looks Like That: Deep Learning for Interpretable Image Recognition, Chaofan Chen et al, 33rd Conference on Neural Information Processing Systems (NeurIPS 2019), Vancouver, Canada.

https://papers.nips.cc/paper/2019/file/adf7ee2dcf142b0e11888e72b43fcb75-Paper.pdf

Figure 1: Image of a clay colored sparrow and how parts of it look like some learned prototypical parts of a clay colored sparrow used to classify the bird’s species. -source

DeepMind's AlphaZero breathes new life into the old art of chess

Sep 2020, phys.org

https://techxplore.com/news/2020-09-deepmind-alphazero-life-art-chess.html

Maybe disturbing? We need the robot to help us learn to be human again? To be ... something? Again?

On the topic of playing chess vs practicing chess, and of course, on being human (vs being a computer) -- As chess grandmaster Vladimir Kramnik recently told Wired magazine, "For quite a number of games on the highest level, half of the game—sometimes a full game—is played out of memory. You don't even play your own preparation; you play your computer's preparation."

The solution? Change the game. With the help of AlphaZero, they discovered new variations on the game of chess that would force players to play again for the first time, I guess. Such changes made were to forbid castling, or introduce self-capture, or allow pawns to move two spaces at once. We evolve together. 

Assessing Game Balance with AlphaZero: Exploring Alternative Rule Sets in Chess, arXiv:2009.04374 [cs.AI] 

Post Script:

Research finds some AI advances are over-hyped

June 2020, phys.org

https://techxplore.com/news/2020-06-ai-advances-over-hyped.html

An article in Science magazine assessing the study cites a meta-analysis of information retrieval algorithms used in search engines over a decade though 2019 and found "the high mark was actually set in 2009." Another study of neural network recommendation systems used by streaming services determined that six of the seven procedures used failed to improve upon the simpler algorithms devised years earlier.

But isn't the new wave of neural networks about size? The algorithms themselves are rather simple, but it's the size of the network of gpu's times the size of the dataset that make the system so powerful.

Also, this sure sounds like an argument against the big dogs in tech inhibiting innovation through monopolistic control. 

Alpha Takes All

This isn't a post about the adversarial neural network that was programmed to play a video game against itself until it took the Grandmaster Championship from the world's best (human) players. It's about a person who beat the computers, and took out a huge portion of the stock market in the process.

Come back with me for a moment, to the year 2010: Smart phones are just now in everybody's hand, the word "application" is just about to be replaced by the word "app," and an algorithm is still just an anagram of the word logarithm.

Wall Street, on the other hand, is just about to get rocked by an event called the Flash Crash of 2010. The market lost and then regained trillions of dollars in 30 minutes, due to a bunch of trading algorithms getting stuck in each other's code like a pack of leashed-up dogs trying to sniff each other's butts. It was described as one of the most turbulent periods in the history of financial markets, and for quite a bit after, nobody knew what the heck happened.

It was also one of the first alarms to be rung about the dangers of artificial intelligence.

"No place for a human" is how the stock market was now described after the takeover by high frequency trading algorithms.
-Jerry Adler, WIRED, 2012

That comment means something different all of the sudden, in light of the Boeing 737 Max's MCAS failure that killed almost 350 people in 2019.

But taking it back to the stock market, the story got a nice twist this year when we were reminded about the Hound of Hounslow. He's a human, but he doesn't exactly belong in the stockmarket either. By 2015, this self-taught quant was held on multiple counts of doing bad things to the stock market, and by 2020, he was given his (lenient) sentence.

Why wasn't he put in jail for life for causing a trillion-dollar market maelstrom? For one thing, he's helping authorities catch other HFT-wielding market manipulators.

The leniency also comes from the fact that he's running an Asperger cortex, and he saw the stock market as a video game, and like AlphaStar, he learned how to play that game really well, and how to "beat" his opponents.  He noticed that the other trading algorithms were doing strange things, and getting entrained into each other's code. He thought he could tap into that groupthink and orchestrate the whole lot, and he did, triggering a cascade of buy/sells that crashed the market.

Notes
Hounslow trader avoids jail in 'flash crash' case
Jan 2020, BBC News

Quantitative Financial Analyst

High Frequency Trading

Post Script
Circa 2010's commentary on high frequency trading:
Raging Bulls - How Wall Street Got Addicted to Light-Speed Trading
Aug 2012, WIRED

Keeping track of algorithms in the news:
Algos
Network Address, 2012

Better Recognize

The majority of the news in deep learning neural nets comes from face recognition applications, which makes sense, because face recognition is a thing that we do. Jennifer Aniston neuron anyone?

DeepGestalt is a face-recognition software that identifies rare genetic disorders evidenced as imperceptible facial alterations (think Down Syndrome but imperceptible). What's cool about this is that it uses your phenome, not your genome.

This is cool because it's totally non-invasive; you don't need genetic material like blood. It's scary because you don't need genetic material to get genetic data about the people you're looking at; I'm thinking surveillance here.

The software was developed by a company called FDNA, and trained on their own database of 500,000 faces (from 10,000 people). Facebook has the biggest face database there is, but this comes in at #2.

And if you're a monkey and feeling left out, don't fret, we're coming for you too. In a shift of species, pictures of cute chimps are being used to train a system that crawls social media posts looking for matches of missing monkeys. (People who buy trafficked monkeys do publicize it, because why else would you want a monkey if not to show your Friends.) It's called ChimpFace, and it's definitely not the only animal-face-rec software out there; elephants, lemurs, lions and pets in general.

So it seems like deep learning and face-recognition are a dynamic duo. But face-rec can be anything-rec. It's pattern recognition. Orgasm recognition? The sound of a fake orgasm? Anything.


Notes:
AI can diagnose some genetic disorders using photos of faces
Jan 2019, Ars Technica

Facial recognition tool tackles illegal chimp trade
Jan 2019, BBC News

Art and AI

Frederic Bazille’s Studio 9 Rue de la Condamine (1870) and Norman Rockwell’s Shuffleton’s Barber Shop (1950)

When A Machine Learning Algorithm Studied Fine Art Paintings, It Saw Things Art Historians Had Never Noticed
The Physics arXiv Blog via Medium, Aug 2014
Source document: Toward Automated Discovery of Artistic Influence

There's some stirring in the dusty world of art history, with the rise of encultured robots threatening human livelihoods. A promising young algorithm is set upon the world, fed with centuries of art imagery, design principles, and historical documentation. Our little algorithm then grows up and learns how to identify patterns in the art world better than its teacher.

In the two compared images above, this little art-historian algo recognized similar compositional patterns that had never been seen before - a hidden Norman Rockwell, see above.

First of all, as an art history major in college, I look at all the compared/related images discovered by the AI, and I am not so impressed. Maybe the general concept is what fails to impress me. When you follow the art world long enough you get to know something about how influence works, and about the power that one thing can have on an artist's work. And I say that there is no such thing as one thing.

The nature of the artist is to take the world at large, a fuck-tonnery of pre-filtered miasma, and to make sense, or at least to fight with it in a way that leaves a record of the battle, and for the benefit of humankind. To say that one painting influenced another because they have similar stylistic elements or design principles is kind of silly. I do understand that subconscious influence has its way with the creative process. But that refers to life as well as art. The new style checker cab, or Triangle shirtwaists, or bubble tea or middle-hipster Americana folk music or The Beatles or African masks or even syphilis could influence an artists' work.

Charge of the Lancers - Umberto Boccioni - 1915

Take ^Futurism, for example. It is inspired by, among other things, the fragmentation of society, be it from national upheaval circa the World Wars, or from the way the landscape looks while riding a speeding train which propelled people faster past the countryside than they had ever moved before. How does an algorithm find that?

I heard Picasso's mistress Françoise Gilot, in her bio of Pable Picasso, say some of his lobster paintings were a response to her hard-shelled personality which came to a head prior to their separation. Algorithms can see that? Nah man.

I know someone can come on here and argue with me, successfully, that artists do influence each other in simple visual ways, and at times, the visual connections can supplement a lack of historical data surrounding their work. But still, there is a need for socio-biographical data in all this, and I wonder if our little algo could be even better trained.

Now, all this having been said, I just finished watching this: Davos talk about the future of artificial intelligence, with IBM CEO Ginni Rommetti. She says that the goal of IBM's artificial intelligence (Watson, by the way, in case you forgot) is to extend human faculties, not replace them. According to her premonitions, the art historian is not doomed, rather it will be enriched and extended by our algorithmic overlords.