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. 

People, Particles and Social Science

aka Sociothermodynamics

Let's not forget that we do live in the era of Big Data, and it's only getting Bigger.

Big data requires new methods of analysis. As we get more and more info about the world around us, we need more basic, underlying frameworks to organize and interpret that data. And that is where physics comes to the rescue. Physicists are used to looking at complex systems with millions, billions and trillions of interactions, and being able to make predictions about their behaviors.

You would think it would show up in the news way more often, because it sure sounds like magic to me. But alas, it's not an everyday headline, so I thought I would spit a few terms up here, just to help stay familiar. This is in relation to urban design and economics.

Inness - the tendency for people to gravitate to the socioeconomic center of a city; this can be correlated to socioeconomic factors, infrastructure factors, and even mortality rates, and an example would be how well-developed cities with multiple socioeconomic centers would have a low inness value.

Betweenness Centrality - a measure of how many things you are in the between of; an example would be how some locations are at the intersection of more than two streets.

These examples are pretty straightforward because they're based on physical objects or locations. Things get a bit more abstract when we start to talk about the spread of disease through a population, or better yet, memetic propagation, which is the spread of ideas.

The practical applications of using network science to predict complex systems are much needed -- the way people move throughout a city is increasingly important when more than half the world's population is now living in cities. But I'd rather hear about how we can predict your chances of adopting a new slang term based on the gesture recognition of your 5 best friends, for example. 

Post Script:
Urban Planning and Big Data
Study finds online restaurant information can closely predict key neighborhood indicators
July 2019, phys.org

What can Wikipedia tell us about human interaction?
June 2019, phys.org

"...if we look at Wikipedia pages about the 2015 Paris terrorist attacks, we can see that the page about the attack is directly connected to the page about Charlie Hebdo magazine, and also to a cluster of pages representing terrorist organizations," Miz explains.

Benzi and Miz call this kind of information-seeking "collective memory," as it can reveal how current events trigger memories of the past.

Hyphens in paper titles harm citation counts and journal impact factors
June 2019, phys.org

Notes:
Can the laws of physics untangle traffic jams, stock markets, and other complex systems?
Mar 2019, phys.org

Here's some other topics at the intersection of science and society:
The hipster effect - Why anti-conformists always end up looking the same
Mar 2019, phys.org

Social media can predict what youíll say, even if you donít participate
Jan 2019, Ars Technica

The 2008 recession associated with greater decline in mortality in Europe
Feb 2019, phys.org

"Periods of macroeconomic recession are associated with lower levels of pollution and fewer accidents in the workplace and on the roads. These are the factors most likely to have the greatest influence on accelerating the decline in mortality. Alcohol and tobacco consumption also fall during periods of greater austerity, as do the prevalences of sedentary lifestyles and obesity. While the underlying mechanisms are still not well established, the findings of some studies also point to the influence of other factors, such as work stress and the fact that healthy habits demand time, something less available to a person working in a full-time job."

And finally, totally unrelated wordporn:
No longer qubits but qutrits!
Complex quantum teleportation achieved for the first time - qutrits

Alpha Control

If this paragraph doesn't make the hair on the back of back of your neck stand at attention, then your ass needs to download an imagination:

DeepMind created five versions of their AI, AlphaStar, and trained them. Their training included footage of human games. A deep neural network trained directly from raw game data by supervised learning and reinforcement learning, said the DeepMind bloggers. AlphaStar, added Coldewey, learned from watching humans play at first, "but soon honed its skills by playing against facets of itself."
-source

Facets of itself.

***
Next up, another item related to botnets and brain-nets, like who wants to be a millionaire but for everything; I'll call it "crowding", but I'm waiting for a better word, which will come in 5 or 6 years, because this will be happening everywhere. It's the wisdom of the crowd, plus artificial intelligence.

Instantaneous crowdsourcing is the way the article refers to it. Real-time remote backup (made of real people) for an otherwise automated system. In this case, that system is air traffic control, and it is built on top of what we've learned so far with automotive traffic control, via the ridesharing explosion happening right now.

The problem is obvious. A car can now drive on its own pretty darn well. But there will always be an extenuating circumstance, an exception to the algorithm, the trolleyology of a real world, high risk situation. And that's where humans have to come in.

Software in the vehicle would analyze real-time vehicle data and electronically guesses 10-30 seconds into the future to estimate the likelihood of a "disengagement"—a situation where the car's automated systems could need human help.

If the likelihood exceeds a pre-set threshold, the system contacts a remotely located control center and sends data from the car.

The control center's system analyzes the car's data, generates several possible scenarios and shows them to several human supervisors, who are situated in driving simulators.

The humans respond to the simulations and their responses are sent back to the vehicle.

The vehicle now has a library of human-generated responses that it can choose from instantaneously, based on information from on-board sensors.
-source

***
This stuff freaks me the out, because there is only one thing on this Earth more powerful than the human brain, and that's more than one human brain, connected together, and operating in real time. That's scarier than AI; it's the combination of humans and artificial intelligence. Semibots is a decent word for it, because it's a little bit of both.

And what's more, we lack the framework for predicting what happens when these hybrid neural networks mature. It's called the science of sync, and it should describe and predict the nature of emergent, synchronized complex networks, except that it doesn't, either because we don't understand it enough, or because such things are impossible to predict.

***
I'm not sure how this will pan out in the near term, but keep in mind there are aspirations for an internet in outerspace; the One-Web satellite constellation was just deployed.

Combine that with the emergent synchronization of the human-robot union described above, and we're getting just beyond the point in our future where we can still grasp what the heck is going on.


Notes:
AlphaStar hungry for world domination in StarCraft II fights
Jan 2019, phys.org

Air traffic control for driverless cars could speed up deployment
Feb 2019, phys.org

OneWeb satellite internet mega-constellation set to fly
Feb 2019, BBC

Physicists discover surprisingly complex states emerging out of simple synchronized networks
Mar 2019, phys.org

Views of the Future

The expanding brain meme is in order.

Look, I'm not in love with Elon Musk or anything, but I got pretty pissed when I read a Futurism article about his interview with Joe Rogan.

I can obviously care less about whether or not somebody "inhaled," I mean, I was around during the Bill Clinton administration, after all. Not to mention, what to hell do people think is going to happen on the Joe Rogan show.

The part I'm upset about is where people seem to think Musk is being a loopy headfreak when he's asked about future-trending topics like AI sentience.

And that's where I'll cut - he's asked when he thinks AI will become sentient. And he responds, very slowly like every other answer he gives during this kind of painful interview - he says something that sounds to many like, "We're all trees, man."

You're thinking, "he didn't answer the question at all." But I think you're wrong. His actual answer went like this:

Joe: How far are humans from creating sentient AI?

Musk: You could argue that any group of people — like a company — is essentially a cybernetic collective of human people and machines. That’s what a company is. And then there are different levels of complexity in the way these companies are formed and then there is a collective AI in Google search, where we are also plugged in like nodes in a network, like leaves in a tree. 

We’re all feeding this network without questions and answers. We’re all collectively programming the AI and Google. […] It feels like we are the biological boot-loader for AI effectively. We are building progressively greater intelligence. And the percentage that is not human is increasing, and eventually we will represent a very small percentage of intelligence.

This is coming from a guy who is making, besides reusable rocket ships and statewide underground transport tunnels and consumer blowtorches, a neural interface system. And in case folks forget, because this was a while ago, he made PayPal. So he knows how things work. He may not be Jaron Lanier but he understands how these things work, and he's telling us that in regards to AI and sentience, we're asking the wrong question.

I'm going to elucidate what I think he was trying to say when he answered this, or at least what maybe he should have said considering his audience in this particular venue.

First, there is a fine line between us and AI. When you consider that we as users are the training program for many of these tools, such as a massive search engine, it should make you wonder which part of it is us and which part the machine.

Second, when we realize that we have essentially been programming these things since the dawn of what I will call for lack of a more neutral word, the surveillance state, we must admit the sentience is already here.

So, even more briefly, what he could have said to the question - when will AI become sentient - was that AI is us, and it's already happened.

All this being said, I spent a good three hours watching a debate on the floor of the Parliament about the real threats of AI, not to mention having watched a good thousand hours of other scientists, researchers, engineers, and other professionals talk about their work in the field. With that, I concluded to myself that a General Artificial Intelligence will not be here for another couple generations, and that the threat will be from ourselves, not a sentient system.

Musk seemed to be in concert with the latter part of this, although he's still pretty spooked about what could happen.

This brings me to my next point, which is that Musk is a guy for whom the world moves way faster than most. We have to remember, if not for him, a working reusable rocket would have taken another 50 years at least. For example, if NASA of some other government agency were in charge of the project. It doesn't mean that because of him the world moves faster, but that when he talks about things, he sees them from a very different perspective, one where amazing things you can barely imagine can happen tomorrow. And he probably assumes that you are in his head with him, which you are not, and Joe Rogan is not, and apparently this writer for Futurism is not.

But fellow speculative fiction enthusiasts are, and hopefully they would agree that it's quite entertaining to listen to this guy talk about things. (Again, if you can get around the 10-second pause before answering every question.)

Finally, full disclosure, I along with Musk and Neil deGrasse Tyson am a no-case guy (no cellphone case that is).

Post Script
The comment about how we're probably living in a simulation - I'm not a proponent of this, as there are more convincing arguments against it than for it coming from the physics world. However, his comment on how in the future we will all be living in a game, and it will be indistinguishable from reality, well, I'll ask folks to read Charles Stross' Halting State, or Vernor Vinge's Rainbow's End or ... many scifi novels written in the past thirty years? (And also a Network Address post from wayback.)

The Beginnings of the Mass Transference Device

AKA Metabots, Mindborgs, and Organic Computing

source

This now makes three-people babies the second scariest thing I've ever heard.

Wiring Monkey Brains Together Has a Point, Say Scientists
WIRED, July 2015

"Today, researchers at Duke University announced they have done nearly that, wiring animal brains together so they could collaborate on simple tasks. Network monkeys displayed motor skills, and networked rats performed computations.

"That’s right. They made a botnet out of brains."

"To build the monkey network, Nicolelis’ team first implanted electrodes in rhesus macaque brains, positioned to pick up signals from a few hundred neurons. Then they connected two or three of the macaques to a computer with a display showing a CG monkey arm. The monkeys were supposed to control the arm, directing it toward a target like a boat crew rows forward. When the monkeys got the arm to hit the target, the researchers rewarded them with juice. (“Each monkey had different juice preference,” says Nicolelis. “We had to do a preference test beforehand.”) To be clear, the monkeys don’t think “move my arm” and the arm moves—they learn what kind of thinking makes the arm move and keep doing that—because monkeys love juice.

"The rat study was even weirder. For this one, the neuroscientists directly wired four rats’ brains together—using the implants to both collect and transmit information about neural activity—so one rat that responded to touch, for example, could pass on their knowledge of that stimulus to another rat. Then the researchers set the rats to a bunch of different abstract tasks—guessing whether it might rain from temperature and air pressure data, for example, or telling the difference between different kinds of touch-stimuli. The brain collectives always did at least as well on those tests as an individual rat would have, and sometimes even better. And in a successful effort to squidge people out, the researchers called these rat-borg collectives “organic computers” or, even worse, “brainets.”"

Animal brains connected up to make mind-melded computer
New Scientist, July 2015

Building an organic computing device with multiple interconnected brains
Nature

Computing Arm Movements with a Monkey Brainet
Nature

Animals Are People Too

Everytime I see articles like this, I think to myself, not that maybe animals are just like people, but instead that the things we consider so special, and the things that 'make us human', are not so.

Social insights from whale chatter
BBC News, Sep 2015

[modified from article]
These great beasts live in very tight social networks, chatting amongst each other using a system of clicks like Morse code - patterns of three to 12 or 15 clicks that vary in rhythm and tempo

The researchers have shown how separate clans - perhaps numbering thousands of animals - will use their own particular subtle sequence of noises.

These vocalisations are not innate - they must be learned, and may indicate the whales are behaving in ways that at some level mirror the operation of human cultures.

"In one clan we call the 'regular clan', we heard regularly spaced clicks, but in another vocal clan that we call the 'plus-ones', the coda types they make have an extended pause at the end before the last click."
-PhD student Maurício Cantor

Crowdsourcing

Meet the Genius Behind the Trillion-Dollar Coin and the Plot to Breach the Debt Ceiling

RYAN TATE 01.10.13

modified:

“It was really a pointless conversation” 

...underlines how ad-hoc online communities, like an anonymous international band of commenters, are increasingly able to move their ideas from the fringes into the middle of political debate.

Goliard Figures, in Umberto Eco, The Infinity of Lists, 2008. p. 263.

in Crowdsourcing Reform, 2010

It may have started as a game, but Beowulf and his pals are poised to inject an important new tactic into oversight of the government’s monetary institutions.

The coin hack even surprised and impressed former U.S. Mint director Philip Diehl, who co-authored the law that enabled the platinum loophole in the first place.

“When I first heard about the idea to mint a trillion-dollar coin, I was very surprised,” says Diehl. “But because I know that law backwards and forwards, I knew immediately that the guy who came up with the idea was right."

http://www.wired.com/business/2013/01/trillion-dollar-coin-inventor

But it wasn't 'a guy' who 'came up with' the idea, it was many.