WHAT is an algorithm? Ask a coder or a mathematician, and they will tell you it is basically a recipe 鈥� a step-by-step list of instructions. Innocent enough. But to some, the term has begun to sound quite sinister.
Algorithms are everywhere. They decide what results you see in an internet search, and what adverts appear next to them. They choose which friends you hear from on social networks. They fix prices for air tickets and home loans. They may decide if you鈥檙e a valid target for the intelligence services. They may even decide if you have the right to vote.
Much of this goes unremarked by those it affects. But when people become aware of it, the reaction is apt to be hostile 鈥� as it was last year when news broke that Facebook had experimented with manipulating its users鈥� emotions through minor changes to their newsfeeds.
Advertisement
This mistrust isn鈥檛 helped by the baffling, often absurd results when algorithms don鈥檛 work as anticipated. Minor goofs, from odd translations to eccentric suggestions, are popular when shared on social media. But bigger glitches can have serious consequences, from 鈥渇lash crashes鈥� on stock markets to fire sales on shopping sites.
And as algorithmic systems grow more tangible, concern is mounting that they are getting too powerful. An offbeat web ad is one thing; a driverless car that ploughs through a playground is another. These systems don鈥檛 even have to malfunction to provoke unease: their decisions can run counter to our ideas of fairness. Algorithms don鈥檛 see humans the same way other humans do.
Put it all together, and the image that comes to mind is of HAL 9000, the inscrutably murderous supercomputer in 2001: A Space Odyssey, and its impassive, unblinking red eye. (The AL in HAL is a contraction of algorithmic, after all.)
No wonder that some people are now trying to check whether algorithms are working as their creators intended. Few of these self-appointed auditors have inside access: rather, they鈥檙e judging algorithms by the results they produce (see 鈥�No one in control: The algorithms that run our lives鈥�).
Some argue that we need something akin to the regulators that oversee financial services and utilities. Perhaps. But what would such watchdogs do? Vetting algorithms in advance isn鈥檛 practical: many are too complex for their outputs to be predicted.
We could start by recalling that algorithms don鈥檛 do anything themselves: they are just recipes. The doing is by systems built to act on their suggestions, often without human intervention.
So a watchdog could specify the kinds of actions that require humans in the loop. It could also act on the 鈥渃lickwrap鈥� agreements that covertly give the creators of such systems licence to make free with our data 鈥� and crack down on government and public services that assume our consent. This is not dissimilar to the intent of the information commissioners some countries have 鈥� although the practice would have to be quite different if it is not to hamstring algorithmic systems鈥� ability to deliver real and sizeable benefits.
Ultimately, those benefits may prove inseparable from their unfamiliar side effects. But then, much the same could have been said of the machines introduced during the Industrial Revolution. Most of us don鈥檛 understand how descendants of those machines bring us modern life, and we may never really grasp the alienness of algorithms. But that doesn鈥檛 mean we can鈥檛 learn to live with them.
聯The benefits of algorithms may be inseparable from their side effects 鈥� as with the Industrial Revolution聰
This article appeared in print under the headline 鈥淟oving the algorithms鈥�
