Unnatural Algorithms?

There is growing societal concern over how companies are making use of our private data. A couple of decades ago, most such data was held on paper, and it wasn’t obvious what companies might do with the data even it was digitized. Today, vast amounts of our personal data are held in sophisticated relational databases which can be mined by sophisticated machine learning algorithms in order to … well, nobody is quite sure, and that is where much of the concern lies. What are these algorithms, and toward what ends may they be employed? With access to so much of our private data, are these algorithms going to control more and more aspects of our daily lives, until we realize, too late, that the machines are in charge? Such concerns might be allayed, or at least complicated, if we take the time to think about what an algorithm actually is, and about the many algorithms that already control our lives – indeed, the lives of all living things.

Contrary to popular conception, an algorithm is not something that runs in a computer, but rather, a recipe for doing something. Your morning routine is an algorithm: get out of bed, make coffee, and so on. That routine is made up of smaller algorithms. Get out of bed involves: throw back sheets, sit up, yawn. Throw back sheets involves: move right arm, grab sheets, move right arm again. However, in general algorithms are more sophisticated than a simple recipe, involving incoming information, and actions conditional on that information. Sit up; if light, move to door, otherwise turn on lamp then move to door; once at door, if door open, walk down stairs, otherwise open door then walk down stairs; if sleepy, make coffee. Computers are particularly efficient at carrying out some forms of algorithms. Notably, those involving only data, mathematics and information, such as machine learning algorithms. But computers were pre-dated by a much more complex, much more general, and much more efficient algorithmic machine called life.

To a large extent, the evolution of life on Earth can actually be defined as the creation and refinement of algorithms, dictating everything from how to grow a single cell; to how to grow a whole animal or plant; how to change your pupil size in response to light; how to react to predators; how to find and attract mates; how to work in a group to hunt down a gazelle. Every such process, at every scale of life from the microscopic to the enormous, is dictated by a recipe, which can legitimately be called an algorithm. Each such algorithm involves the integration of various forms of incoming information, which is fed into computational processes that make a decision based on that information. Once again, the algorithms tend to be made up of smaller algorithms, just as computer programs are made of smaller sub-routines. The algorithms might be executed by neurons in a brain; by genes within a genetic network; by a population of cells exchanging hormones; or by a population of animals exchanging communication signals. But they are algorithms all the same.

Overarching the whole of life is a very special kind of algorithm called evolution by natural selection. The role of this most uber of all uber-algorithms has been to refine the lower-level algorithms of life, weeding out those that lead to reproductive failure, and bulking up (and mutating) those that lead to reproductive success. Natural selection has been unwavering and utterly heartless in its pursuit of this simple goal. Yes, it has led to cooperation, mutualistic behaviour, and parental care of offspring, but only because this increases – you guessed it – reproductive success. In the pursuit of that same goal, it has also created an arsenal of aggressive weaponry and tactics that would rival anything in the history of human conflict. Just read about parasitoids. Actually, don’t.

The algorithms that control our bodies and brains have undergone selection in a very wide variety of environments, beginning right back with the very first appearance of life on Earth. This wide variety of environments has, in turn, selected for algorithms that are robust and adaptable, which explains why the algorithms have been able to cope with the unprecedented challenges thrown up by our modern environment. Who could have predicted that our navigation algorithms could cope with driving a car, or that our motor control algorithms could cope with typing on a mobile phone? Having said this, some of our algorithms are no longer fit for purpose. We now want to live longer than we need to just to reproduce, so we don’t like the fact that our cellular algorithms were not optimized to prevent cancer or dementia in old age. We want to live healthy, happy, productive, lives, so we are annoyed at the algorithms that make us eat too much food when it is on offer, or make us get depressed when we are overstimulated. We have developed concepts of human rights and justice, so we don’t like the fact that some of our mental algorithms make us unconsciously biased by gender, race and a host of other factors. We want to see an end to war, but deep-seated social algorithms, which lead to an unwarranted sense of us vs them, keep starting it up again.

All of which would be fine if only we could reprogram life’s algorithms. But we can’t, because we don’t understand the computational machinery of life sufficiently to do so. The miracle of modern medicine – and miracle it is – has not, in general, come about through a deep understanding of our internal systems, but through educated guesswork, trial and error.

This is about to change. We are on the brink of a completely new level of understanding of natural computational systems. We will one day understand the algorithms that enable trillions of genes and cells to coordinate themselves in order to build a perfect human baby, or trillions of neurons to work together to outthink the world’s data centers using less energy than a 40W bulb. Using this new understanding, there will be, for the first time in the history of all life on Earth, a species that is not at the mercy of the algorithms handed to it by natural selection. We will, if we only choose to, reprogram ourselves to cure cancer, dementia, mental illness, sexism, racism and war.

Ironically, this new understanding of life’s algorithms will come about through the application of machine learning and AI algorithms, running in silico, and utilizing various forms of biological data taken from the likes of you and me. Should we be worried about the dangers of these algorithms, and the other ones working away at our financial data, movement data or browsing history? Can we understand and control them sufficiently to head off those dangers? We should be concerned. Indeed, we are, which is why the world’s nations, companies are beginning to put in place all kinds of safeguards around the use of such data. But, as I hope this essay has made clear, we humans are, and have always been, at the mercy of algorithms. What is different about the new algorithms is not that they can’t be controlled – but that, in principle, they can.

This essay was inspired by the panel session ‘Gods In Boxes’ at Techonomy 2015.

© 2017 Drew Purves All Rights Reserved