Spinning Data into Thought

Spinning Data into ThoughtHow Computers Think: An IntroductionSimon CarryerBlockedUnblockFollowFollowingJan 6Anyone who has seen a ball of wool, a length of flax fibre, or a bundle of raw cotton, could, in theory, imagine how to twist these fibres into yarn, how to knit a woollen jumper, or how to weave a cotton blanket or a canvas sail.

The art and science of textiles does not exploit any hidden property of thread; no obscure chemistry or quirk of physics need be discovered to understand how a fibre can be twisted and knotted to produce a flexible, strong fabric.

It is just the geometry of the object which allows this, plain for anyone to see.

That’s in theory.

But theory, as always, exceeds practice.

Because no one person in history did just look at a length of fibre and figure all that out.

There is no single person who could figure all that out, just by looking, or if there was, they had something else on their mind.

Textile technology has developed incredibly slowly across history, the product of hundreds of thousands of individual human beings’ ingenuity, hard work, and moments of brilliance.

Archeological records tell us the first fabrics are nearly thirty thousand years old, known only by the impressions they left on tiny pieces of clay.

Since that time, techniques have been invented, forgotten and rediscovered all over the world, and the technology continues to develop as further convolutions of the geometric properties of a simple thread are unravelled.

A complex pattern emerges from the repetition of simple stepsThe most sophisticated knitting patterns now resemble lines of computer code, and in fact, the resemblance is more than superficial.

Knitting patterns are instructions for a series of simple steps which, repeated, produce an extraordinarily complex result.

Both knitting patterns and computer programs encode logical operations which are interpreted (by a human or a computer) into a process which transforms a simple input into a vastly more useful output.

This is the definition of the word “algorithm”.

The word conjures images of greek letters scrawled on a chalkboard by some bearded, tweed-wearing professor, or the inscrutable inner workings of an ineffable machine.

But the reality is much more accessible.

“Algorithm” simply means “a series of repeatable, logical steps”.

Textiles are, in a very literal sense, products of mathematical operations, and advances in humans’ ability to perform those operations at scale has had a profound effect upon our lives.

The Industrial RevolutionA spinning wheel is a surprisingly simple device.

It allows a worker to feed fibres onto a revolving wheel, which cunningly twists the fibres into thread, and collects this thread on a spindle.

In various forms, the spinning wheel has been known for at least the last thousand years.

In 1764, according to a legend which is almost certainly untrue, James Hargreaves, a weaver and carpenter from Lancashire, England, saw an overturned spinning wheel, and noticed that this orientation would allow multiple spindles to be placed in a row, allowing a single wheel to spin fibres onto each of them.

His subsequent invention, the “spinning jenny”, allowed one spinner to do the work of eight.

Around the time of Hargreaves’ invention, it is estimated there were fewer than 50,000 spindles producing thread in England.

By the 1820s, there were over seven million.

This small shift in perspective, this application of simple geometry at a vast scale, was one of the first steps in what became Britain’s Industrial Revolution.

The Spinning JennyThe effects of the Industrial Revolution on our society are almost too vast to understand.

What began as a quantitative change — one worker doing the job of eight, then sixteen, then hundreds — became a qualitative shift in the social fabric.

Entire new industries were born out of the ashes of old ones.

Products that were once the preserve of the wealthy became available to the masses.

As transportation networks improved, the world grew effectively smaller — markets on the far side of the continent from each other were suddenly only a few days’ travel away.

Power fundamentally shifted, from land-owning nobility, to a factory-owning middle-class.

Not all of these changes were positive, and not all people benefited from the radical social changes sparked by the rapid pace of technological development.

While the whole country was wild for the new cheap fabrics, and the possibilities they enabled, there was at least one group amongst whom these innovations were substantially less popular, and they started a social movement whose name still has currency today, though not quite in the way they hoped.

They called themselves “Luddites”.

When a new process means that one person can do the work of eight, the result is not eight people doing an eighth as much work, but rather, one person does the same amount of work, and the other seven go hungry.

Stockingers, the skilled artisans who produced finely-knitted fabrics for stockings and hosiery, were not thrilled by this maths.

The stocking frame, invented in 1589, but improved and widely adopted late in the eighteenth century, threatened their way of life.

A stocking frame allowed the fine work of stocking-making to be performed faster and more reliably, and by a relatively unskilled operator.

The stockingers, protecting their livelihoods and their way of life, revolted.

Taking inspiration from a (probably fictional) story of a weaver, Ned Ludd, who smashed his knitting frame, the stockingers’ protests often took the form of destroying the machinery that was destroying their livelihood.

The Luddites were born.

From 1812, “Machine Breaking” was punishable by deportation or deathIn popular conception the Luddites have become something of a metaphor for fear of technology, their name synonymous with a stubborn resistance to progress.

But their actual position was far closer to that of a traditional labour movement.

Their concern was not the rise of new technology per se, but rather the effect that these new machines had on their livelihoods, the conditions in which they were expected to work, and the pay they would receive for that work.

The tide of history was not kind to the Luddites.

Stocking frames were manufactured faster than the protestors could smash them, and the craft the stockingers had learned became obsolete, replaced by unskilled labourers sweating in hot, dangerous factories.

Perhaps the stockingers were able to find new work, in one of the many new industries that emerged in the wake of the industrial revolution — as stokers, lamplighters, bloomers and so on — but for many of them this must have been a life-altering, traumatic transition.

Artificial IntelligenceGiven the context of this essay, you might be able to guess the analogy I’m drawing.

We are, in the early decades of the twenty-first century, in the midst of a similar transitional phase.

Automation of a huge range of tasks, replacement of skilled workers with artificial intelligences, new processes that replace difficult tasks with the press of a button, all echo the changes of the Industrial Revolution.

Whether the social changes that accompany these technological developments will be as profound as those of the Industrial Revolution remains to be seen, but I think the comparison is instructive.

Like the techniques of knitting and weaving had been known many years before they were industrialised, the mathematics that underpin artificial intelligence are not, for the most part, particularly new.

Also like textile crafts, they are simply manipulations of the geometric properties of a raw material.

For textiles, this is thread.

For artificial intelligence, this is data.

The crucial shift that has taken place is not in the development of new techniques (though many new techniques have been developed).

Rather, it is the speed with which these operations can be performed, and the amount of data available to them, which has qualitatively changed what they are able to achieve.

By “artificial intelligence” I don’t mean the pop-culture image of all-knowing computer overlords, or sinister robot assistants.

I mean the broad range of techniques that allow us to transform digital information into more useful products.

These techniques are already widely adopted in all classes of consumer product.

We can turn information about people’s shopping habits into predictions about next year’s fashion trends, or into recommendations about what someone might want to purchase next.

From accident and injury data, we can build models that classify people as high or low insurance risks.

We can turn digitised documents into search algorithms that identify relevant legislation or research.

We can even build programs which mimic and respond to human language.

Many of these tools do not fit neatly into the idea of “intelligence” as we apply it to human thinking.

But in aggregate, they produce the kinds of results that have previously only been achievable with human minds.

Essentially, the changes we are experiencing are quantitative changes.

We have always been able to predict fashion trends, search through documents, to speak and listen.

What is changing is the speed and scale of our ability to perform these operations.

From being limited by the speed and attention span of a human operator, we’ve moved to the (in some ways) less limited power of computers.

As with the Industrial Revolution, technological developments have increased the productivity of the individual worker, and just like the Industrial Revolution, this is having increasingly radical effects on which skills are in demand, and which are becoming obsolete.

We share our world with computer intelligences.

These intelligences think very differently to how you or I do, if they can be said to “think” at all.

To describe how they work as intelligence, to relate to them as thinking, perceiving beings, is at best a strained metaphor, a tool for helping us understand something new by relating it to something familiar.

But it is a metaphor that is becoming increasingly apt.

Like a human intelligence, artificial intelligences are the product of their history.

They are their parents’ children, and they reflect the biases, foibles, and also the strengths and values of their creators.

They are neither more nor less innocent than the people who build them.

What is perhaps the greatest difference between human intelligence and artificial intelligence, the most important hurdle to us perceiving them as thinking “like us”, is that their processes can, for the most part, be laid completely bare.

The inner workings of a human mind may be forever mysterious to us, probed only by the inexact sciences of psychology and anthropology.

The soft meat of a human brain tells us little about the inner life it contains.

But a machine intelligence, complex though it may be, is always reducible to its constituent parts.

When the artificial intelligence is a “black box”, which conceals from us how it arrives at its pronouncements, we can imagine it is the work of some abstract personality.

But exposing the cogs in the machine, unravelling the knots in the tapestry, we can see the simple rules, the ones and zeroes, the knits and purls, that have built this complex fabric.

Like glimpsing behind the Wizard’s curtain, seeing how the trick is performed dispels the magic.

But it is a magic that should be dispelled.

To understand these new, artificial intelligences, to relate to them on something closer to their own terms, we need to form better expectations of what they can and cannot achieve, we need to speak a little of their language.

Like the stockingers, we face a choice: We can emulate Ludd, and attempt to smash or hobble these new machines and preserve the way of life we currently enjoy.

Or perhaps we can try to negotiate a new relationship with them, to find a home for them in our lives.

In the following essays, I will explain in simple terms the processes that artificial intelligence algorithms use to spin data into a mimicry of thought.

I’ll start with the simplest of these algorithms, and introduce some of the key concepts in machine learning.

I’ll also explore other kinds of artificial intelligence, such as that used to drive the actions of computer-controlled characters in video games.

I’ll also venture into the world of “deep learning”, exploring neural networks which attempt to mimic human brain structures.

Along the way, we’ll learn about some fun topics related to the data I’m using.

We’ll find answers to a few questions which are increasingly relevant in everyone’s lives: What is artificial intelligence?.How does it work?.What can it do well, and what does it do badly?.What even is intelligence?.And finally, I hope, we’ll learn a little more empathy, a little more understanding for these new computer intelligences with which we are sharing our lives.

Go to Part One: Decision Trees and Dinosaurs.. More details

Leave a Reply