On artificial intelligence and solitary confinement.

On artificial intelligence and solitary confinement.

512px-Ludwig_WittgensteinIn Philosophical Investigations (translated by G. E. M. Anscombe), Ludwig Wittgenstein argues that something strange occurs when we learn a language.  As an example, he cites the problems that could arise when you point at something and describe what you see:

The definition of the number two, “That is called ‘two’ “ – pointing to two nuts – is perfectly exact.  But how can two be defined like that?  The person one gives the definition to doesn’t know what one wants to call “two”; he will suppose that “two” is the name given to this group of nuts!

I laughed aloud when I read this statement.  I borrowed Philosophical Investigations a few months after the birth of our second child, and I had spent most of his first day pointing at various objects in the hospital maternity ward and saying to him, “This is red.”  “This is red.”

“This is red.”

Of course, the little guy didn’t understand language yet, so he probably just thought, the warm carry-me object is babbling again.

IMG_5919
Red, you say?

Over time, though, this is how humans learn.  Wittgenstein’s mistake here is to compress the experience of learning a language into a single interaction (philosophers have a bad habit of forgetting about the passage of time – a similar fallacy explains Zeno’s paradox).  Instead of pointing only at two nuts, a parent will point to two blocks – “This is two!” and two pillows – “See the pillows?  There are two!” – and so on.

As a child begins to speak, it becomes even easier to learn – the kid can ask “Is this two?”, which is an incredibly powerful tool for people sufficiently comfortable making mistakes that they can dodge confirmation bias.

y648(When we read the children’s story “In a Dark Dark Room,” I tried to add levity to the ending by making a silly blulululu sound to accompany the ghost, shown to the left of the door on this cover. Then our youngest began pointing to other ghost-like things and asking, “blulululu?”  Is that skeleton a ghost?  What about this possum?)

When people first programmed computers, they provided definitions for everything.  A ghost is an object with a rounded head that has a face and looks very pale.  This was a very arduous process – my definition of a ghost, for instance, is leaving out a lot of important features.  A rigorous definition might require pages of text. 

Now, programmers are letting computers learn the same way we do.  To teach a computer about ghosts, we provide it with many pictures and say, “Each of these pictures has a ghost.”  Just like a child, the computer decides for itself what features qualify something for ghost-hood.

In the beginning, this process was inscrutable.  A trained algorithm could say “This is a ghost!”, but it couldn’t explain why it thought so.

From Philosophical Investigations: 

Screen Shot 2018-03-22 at 8.40.41 AMAnd what does ‘pointing to the shape’, ‘pointing to the color’ consist in?  Point to a piece of paper.  – And now point to its shape – now to its color – now to its number (that sounds queer). – How did you do it?  – You will say that you ‘meant’ a different thing each time you pointed.  And if I ask how that is done, you will say you concentrated your attention on the color, the shape, etc.  But I ask again: how is that done?

After this passage, Wittgenstein speculates on what might be going through a person’s head when pointing at different features of an object.  A team at Google working on automated image analysis asked the same question of their algorithm, and made an output for the algorithm to show what it did when it “concentrated its attention.” 

Here’s a beautiful image from a recent New York Times article about the project, “Google Researchers Are Learning How Machines Learn.”  When the algorithm is specifically instructed to “point to its shape,” it generates a bizarre image of an upward-facing fish flanked by human eyes (shown bottom center, just below the purple rectangle).  That is what the algorithm is thinking of when it “concentrates its attention” on the vase’s shape.

new york times image.jpg

At this point, we humans could quibble.  We might disagree that the fish face really represents the platonic ideal of a vase.  But at least we know what the algorithm is basing its decision on.

Usually, that’s not the case.  After all, it took a lot of work for Google’s team to make their algorithm spit out images showing what it was thinking about.  With most self-trained neural networks, we know only its success rate – even the designers will have no idea why or how it works.

Which can lead to some stunningly bizarre failures.

artificial-intelligence-2228610_1280It’s possible to create images that most humans recognize as one thing, and that an image-analysis algorithm recognizes as something else.  This is a rather scary opportunity for terrorism in a world of self-driving cars; street signs could be defaced in such a way that most human onlookers would find the graffiti unremarkable, but an autonomous car would interpret in a totally new way.

In the world of criminal justice, inscrutable algorithms are already used to determine where police officers should patrol.  The initial hope was that this system would be less biased – except that the algorithm was trained on data that came from years of racially-motivated enforcement.  Minorities are still more likely to be apprehended for equivalent infractions.

And a new artificial intelligence algorithm could be used to determine whether a crime was “gang related.”  The consequences of error can be terrible, here: in California, prisoners could be shunted to solitary for decades if they were suspected of gang affiliation.  Ambiguous photographs on somebody’s social media site were enough to subject a person to decades of torture.

Solitary_Confinement_(4692414179)When an algorithm thinks that the shape of a vase is a fish flanked by human eyes, it’s funny.  But it’s a little less comedic when an algorithm’s mistake ruins somebody’s life – if an incident is designated as a “gang-related crime”, prison sentences can be egregiously long, or send someone to solitary for long enough to cause “anxiety, depression, and hallucinations until their personality is completely destroyed.

Here’s a poem I received in the mail recently:

LOCKDOWN

by Pouncho

For 30 days and 30 nights

I stare at four walls with hate written

         over them.

Falling to my knees from the body blows

         of words.

It damages the mind.

I haven’t had no sleep. 

How can you stop mental blows, torture,

         and names –

         They spread.

I just wanted to scream:

         Why?

For 30 days and 30 nights

My mind was in isolation.

On theft and bullying.

On theft and bullying.

We aren’t born equal.

We never have been, not humans or any other animals.  Among species that birth in litters, the baby that leaves the womb largest has a lifelong advantage.  In the modern world, where a baby happens to be born dictates its future prospects to a huge degree.  Maybe it’ll be born in the United States, instantly reaping the benefits of citizenship.  Or maybe it’ll be born into a war-torn nation, amidst strife caused by climate change… which was itself caused by the creation of wealth and prosperity for all those U.S. babies.

And we use our initial advantages to further tilt the scales.  The largest mammal in a litter pushes the others aside and takes the most milk.  The rich get richer.

The same principal holds true in astrophysics.  The more dense a black hole, the better it will be at grabbing additional matter.  Densely-arrayed galaxies will keep their neighbors longest: because empty space expands, the rate at which stars drift away from each other depends upon their initial separation.  The farther away you are, the faster you will recede.  The lonely become lonelier.

But we are blessed.  Through the vagaries of evolution, humans stumbled into complex language.  We can sit and contemplate the world; we can consciously strive to be better than nature.

We can read Calvin and Hobbes and think, “Hey, that’s not fair!”

calvin 9 - 11.png

It is perfectly natural for Moe to get the truck.  He is bigger, after all.  He is stronger.  In the world’s initial inequalities of distribution, physical prowess determined who would reap plenty and who would starve.  After all, not all territories are equivalently bountiful for hunters or gatherers.  Now we have sheets of paper that ostensibly carve up the world amongst us, but in past eras raw violence would’ve staked claims.  Human mythology brims with accounts of battle to gain access to the best resources… and we humans still slaughter each other whenever insufficient strength seems to back the legitimacy of those papers.

When the threat of contract-enforcing state violence in Syria waned, local murder began.  And we lack an international state threatening violence against individual nations – inspired partly by the desire for resources, George W. Bush initiated a campaign of murder against Iraq.

Except when we’ve banded together to suppress individual violence (the state as Voltron), the strong take from the weak.

At least we know it’s wrong.

We’ve allowed other forms of bullying and theft to slip by.  After all, differing physical prowess is only one of the many ways in which we are born unequal.  If it is wrong for the strongest individual to steal from others, is it also wrong for the most clever to do the same?

weaponsofmathdestructionFrom Cathy O’Neil’s Weapons of Math Destruction:

But the real problem came from a nasty feeling I started to have in my stomach.  I had grown accustomed to playing in these oceans of currency, bonds, and equities, the trillions of dollars flowing through international markets.  But unlike the numbers in my academic models, the figures in my models at the hedge fund stood for something.  They were people’s retirement funds and mortgages.  In retrospect, this seems blindingly obvious.  And of course, I knew it all along, but I hadn’t truly appreciated the nature of the nickels, dimes, and quarters that we pried loose with our mathematical tools.  It wasn’t found money, like nuggets from a mine or coins from a sunken Spanish galleon.  This wealth was coming out of people’s pockets.  For hedge funds, the smuggest of the players on Wall Street, this was “dumb money.”

the math was directed against the consumer as a smoke screen.  Its purpose was only to optimize short-term profits for the sellers.  And those sellers trusted that they’d manage to unload the securities before they exploded.  Smart people would win.  And dumber people, the providers of dumb money, would wind up holding billions (or trillions) of unpayable IOUs. … Very few people had the expertise and the information required to know what was actually going on statistically, and most of the people who did lacked the integrity to speak up.

O’Neil was right to feel queasy – after all, she had become Moe.  All the high-frequency traders – who are lauded as brilliant despite often doing no more than intercepting others’ orders, buying desired products a millisecond before anyone else can, and re-selling them at a profit – are simply thieves.  Sometimes they are stealing because they are more clever.  Other times, they are stealing because their pre-existing wealth allows them to buy access to lower-latency computer servers than anyone else.

In any case, Calvin would disapprove.

calvin 9 - 20.png

Thankfully, O’Neil quit stealing (although she doesn’t mention returning her prior spoils).  After all, that is part of our blessing – we cannot change the past, but…

… and here it’s worth mentioning that Ludwig Wittgenstein was clearly incorrect when he wrote that “One can mistrust one’s own senses, but not one’s own belief.  If there were a verb meaning ‘to believe falsely’, it would not have any significant first person present indicative.”  Most physicists believe in free will and the mutability of the future, despite also knowing that, according to the laws of physics, their beliefs should be false…

… we can always fix the future.

(As a special treat – here is one of the most beautiful comic strips about opening your eyes to change.)

021217-600x202

On perception and learning.

On perception and learning.

Cuddly.

Fearful.

Monstrous.

Peering with the unwavering focus of a watchful overlord.

A cat could seem to be many different things, and Brendan Wenzel’s recent picture book They All Saw a Cat conveys these vagrancies of perception beautifully. Though we share the world, we all see and hear and taste it differently. Each creature’s mind filters a torrential influx of information into manageable experience; we all filter the world differently.

They All Saw a Cat ends with a composite image. We see the various components that were focused on by each of the other animals, amalgamated into something approaching “cat-ness.” A human child noticed the cat’s soft fur, a mouse noticed its sharp claws, a fox noticed its swift speed, a bird noticed that it can’t fly.

All these properties are essential descriptors, but so much is blurred away by our minds. When I look at a domesticated cat, I tend to forget about the sharp claws and teeth. I certainly don’t remark on its lack of flight – being landbound myself, this seems perfectly ordinary to me. To be ensnared by gravity only seems strange from the perspective of a bird.

theyallsawThere is another way of developing the concept of “cat-ness,” though. Instead of compiling many creatures’ perceptions of a single cat, we could consider a single perceptive entity’s response to many specimens. How, for instance, do our brains learn to recognize cats?

When a friend (who teaches upper-level philosophy) and I were talking about Ludwig Wittgenstein’s Philosophical Investigations, I mentioned that I felt many of the aims of that book could be accomplished with a description of principal component analysis paired with Gideon Lewis-Kraus’s lovely New York Times Magazine article on Google Translate.

My friend looked at me with a mix of puzzlement and pity and said, “No.” Then added, as regards Philosophical Investigations, “You read it too fast.”

wittgensteinOne of Wittgenstein’s aims is to show how humans can learn to use language… which is complicated by the fact that, in my friend’s words, “Any group of objects will share more than one commonality.” He posits that no matter how many red objects you point to, they’ll always share properties other than red-ness in common.

Or cats… when you’re teaching a child how to speak and point out many cats, will they have properties other than cat-ness in common?

In some ways, I agree. After all, I think the boundaries between species are porous. I don’t think there is a set of rules that could be used to determine whether a creature qualifies for personhood, so it’d be a bit silly if I also claimed that cat-ness could be clearly defined.

But when I point and say “That’s a cat!”, chances are that you’ll think so too. Even if no one had ever taught us what cats are, most people in the United States have seen enough of them to think “All those furry, four-legged, swivel-tailed, pointy-eared, pouncing things were probably the same type of creature!”

Even a computer can pick out these commonalities. When we learn about the world, we have a huge quantity of sensory data to draw upon – cats make those noises, they look like that when they find a sunny patch of grass to lie in, they look like that when they don’t want me to pet them – but a computer can learn to identify cat-ness using nothing more than grainy stills from Youtube.

Quoc Le et al. fed a few million images from Youtube videos to a computer algorithm that was searching for commonalities between the pictures. Even though the algorithm was given no hints as to the nature of the videos, it learned that many shared an emphasis on oblong shapes with triangles on top… cat faces. Indeed, when Le et al. made a visualization of the patterns that were causing their algorithm to cluster these particular videos together, we can recognize a cat in that blur of pixels.

The computer learns in a way vaguely analogous to the formation of social cliques in a middle school cafeteria. Each kid is a beautiful and unique snowflake, sure, but there are certain properties that cause them to cluster together: the sporty ones, the bookish ones, the D&D kids. For a neural network, each individual is only distinguished by voting “yes” or “no,” but you can cluster the individuals who tend to vote “yes” at the same time. For a small grid of black and white pixels, some individuals will be assigned to the pixels and vote “yes” only when their pixels are white… but others will watch the votes of those first responders and vote “yes” if they see a long line of “yes” votes in the top quadrants, perhaps… and others could watch those votes, allowing for layers upon layers of complexity in analysis.

three-body-problem-by-cixin-liu-616x975And I should mention that I feel indebted to Liu Cixin’s sci-fi novel The Three-Body Problem for thinking to humanize a computer algorithm this way. Liu includes a lovely description of a human motherboard, with triads of trained soldiers hoisting red or green flags forming each logic gate.

In the end, the algorithm developed by Le et al. clustered only 75% of the frames from Youtube cat videos together – it could recognize many of these as being somehow similar, but it was worse at identifying cat-ness than the average human child. But it’s pretty easy to realize why: after all, Le et al. titled their paper “Building high-level features using large scale unsupervised learning.”

Proceedings of the International Conference on Machine Learning 2010
You might have to squint, but there’s a cat here. Or so says their algorithm.

When Wittgenstein writes about someone watching builders – one person calls out “Slab!”, the other brings a large flat rock – he is also considering unsupervised learning. And so it is easy for Wittgenstein to imagine that the watcher, even after exclaiming “Now I’ve got it!”, could be stymied by a situation that went beyond the training.

Many human cultures have utilized unsupervised learning as a major component of childrearing – kids are expected to watch their elders and puzzle out on their own how to do everything in life – but this potential inflexibility that Wittgenstein alludes to underlies David Lancy’s advice in The Anthropology of Childhood that children will fair best in our modern world when they have someone guiding their education and development.

Unsupervised learning may be sufficient to prepare children for life in an agrarian village. Unsupervised learning is sufficient for chimpanzees learning how to crack nuts. And unsupervised learning is sufficient to for a computer to develop an idea about what cats are.

But the best human learning employs the scientific method – purposefully seeking out “no.”

I assume most children reflexively follow the scientific method – my daughter started shortly after her first birthday. I was teaching her about animals, and we started with dogs. At first, she pointed primarily to creatures that looked like her Uncle Max. Big, brown, four-legged, slobbery.

IMG_5319.JPG
Good dog.

Eventually she started pointing to creatures that looked slightly different: white dogs, black dogs, small dogs, quiet dogs. And then the scientific method kicked in.

She’d point to a non-dog, emphatically claiming it to be a dog as well. And then I’d explain why her choice wasn’t a dog. What features cause an object to be excluded from the set of correct answers?

Eventually she caught on.

Many adults, sadly, are worse at this style of thinking than children. As we grow, it becomes more pressing to seem competent. We adults want our guesses to be right – we want to hear yes all the time – which makes it harder to learn.

The New York Times recently presented a clever demonstration of this. They showed a series of numbers that follow a rule, let readers type in new numbers to see if their guesses also followed the rule, and asked for readers to describe what the rule was.

A scientist would approach this type of puzzle by guessing a rule and then plugging in numbers that don’t follow it – nothing is ever really proven in science, but we validate theories by designing experiments that should tell us “no” if our theory is wrong. Only theories that all “falsifiable” fall under the purvey of science. And the best fields of science devote considerable resources to seeking out opportunities to prove ourselves wrong.

But many adults, wanting to seem smart all the time, fear mistakes. When that New York Times puzzle was made public, 80% of readers proposed a rule without ever hearing that a set of numbers didn’t follow it.

Wittgenstein’s watcher can’t really learn what “Slab!” means until perversely hauling over some other type of rock and being told, “no.”

We adults can’t fix the world until we learn from children that it’s okay to look ignorant sometimes. It’s okay to be wrong – just say “sorry” and “I’ll try to do better next time.”

Otherwise we’re stuck digging in our heels and arguing for things we should know to be ridiculous.

It doesn’t hurt so bad. Watch: nope, that one’s not a cat.

16785014164_0b8a71b191_z
Photo by John Mason on Flickr.